55 Commits

Author	SHA1	Message	Date
Kgothatso Ngako	1bd5ad960f	docs: record what phase 2 built, and why the audit changed shape ... The plan's phase 2 was written before the migration ran and assumed the work was mostly a sweep for off-grid numbers. It was not: 10dp and 20dp dominate the tree and both are already on the M3 scale, so only 89 of 527 were off-grid at all. The section now says what was actually built -- the scale, the two migrations, and the reason the audit's value-based classification had to become a shape-based one -- along with the 0.03% pixel diff that shows the sweep moved nothing. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-08 00:37:30 +02:00
Kgothatso Ngako	8cd0f3f44d	refactor: take the last 353 spacing literals onto the scale, and reach zero ... Phase 2, final step, of docs/material-design-conformance.md. Every `.dp` literal in a spacing position in the UI tree is now a token. 431 reads of `MaterialTheme.spacing.*`, one reasoned exemption, and `m3-spacing-positions.py` exits 0. **Shape decides the token, not just the value.** The migration script grew a per-shape mapping because the same number means different things in different positions: 8dp of padding is `compactPadding`, 8dp of gap is `itemGap`, and 8dp under a `Spacer` is neither of those and stays `space100`. Where the pair determines the meaning the semantic name is used, and nowhere else: padding + 8dp -> compactPadding 10 sites padding + 16dp -> containerPadding 10 gap + 4dp -> relatedGap 8 gap + 8dp -> itemGap 10 That is 38 of 353. The rest take the raw stop, and deliberately: assigning a semantic name needs somebody to have read what the container *is*, and a name that asserts a meaning the code does not have is worse than a stop that asserts none. `screenMargin` in particular is unassignable mechanically -- it is 16dp of padding, exactly like `containerPadding` -- so it has no call sites yet and gets them when someone reads the screens. **Two spacers were standing in for zero.** `WriteNewNoteScreen` renders `Spacer(Modifier.height(1.dp))` twice, in the `LazyColumn` item that shows a reply preview when there is one. There is nothing to show and the item still has to render something; 1dp was the placeholder. Now `space0`, with a comment, because a 1dp gap that nobody intended is the kind of thing that gets copied. **One value is exempt, and says so at the site.** `SovereignWalletStartupScreen`'s `Spacer(Modifier.height(128.dp))` is room to scroll the last wallet clear of the bottom of the window -- reserved space, not a step in the rhythm. The scale tops out at `space900` (72dp) and rounding to it would put the row back under the edge. Rather than exempt it in the script by value, the classifier now honours an inline `// m3-spacing-exempt: <reason>` comment on the lines directly above. Exemptions belong at the call site: the reason travels with the code, a reviewer sees it in the diff that adds it, and the tool stops accumulating a list of numbers that mean nothing on their own -- the mistake the first version of this audit made with `DIMENSION_EXEMPT`. **Where the tokens landed.** `space125` (10dp) 128 times and `space250` (20dp) 107 -- the two values that already dominated the tree, now named. `space600` (48dp) 52 times, which is the empty-state spacer from the previous commit. The long tail is 2, 4, 6, 12, 14, 16, 24, 32, 40 and 64dp. **Verified that nothing moved.** The landing screen was captured on emulator-5554 before and after and compared pixel by pixel on a 4px grid: **47 differing samples out of 162,000, 0.03%**, and they are the status bar clock. The sweep is a rename. **Budget ratcheted 353 -> 0**, dated in the file. Phase 8 wires `--check` into CI, at which point a new `.dp` in a `padding()` fails the build. **Tests.** 942 pass, 594 jvm over 72 classes and 348 android over 44, unchanged -- `SpacingScaleTest` already asserts the scale, and there is nothing to assert about a call site having been renamed that the compiler does not. `:composeApp:compileDebugKotlinAndroid` builds, the debug apk installs and runs, `m3-audit.sh --check` exits 0. 75 files, 432 insertions, 348 deletions. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-08 00:36:59 +02:00
Kgothatso Ngako	f7a732d68a	refactor: move the 89 off-grid spacing values onto the M3 scale ... Phase 2, second step, of docs/material-design-conformance.md. 77 of the 89 literals that were off M3's spacing scale sat in spacing positions and now read `MaterialTheme.spacing.spaceNNN`; the remaining 12 are dimensions and are out of scope. One drifted corner moved onto the shape scale. **The mapping, and why each is the nearest stop rather than the nicest number.** 5.dp x10 -> space50 (4dp) padding and gaps in dense rows 15.dp x14 -> space200 (16dp) card and dialog padding, two gaps 30.dp x1 -> space400 (32dp) the spacer under LoadingDataIndicator's spinner 50.dp x52 -> space600 (48dp) the spacer above an empty or error message Nearest-stop throughout, so the largest move is 2dp and most are 1. `5.dp` is equidistant between `space50` and `space75`; it goes to 4dp because `spacedBy(4.dp)` is already the idiom elsewhere in the tree and a scale with two answers for the same input is not one. The 52 at 48dp are the same three lines copied into 16 files -- a `Spacer` pushing "Something went wrong" down the screen. Phase 5 retires them into a shared empty-state composable; migrating them first means that composable inherits a token rather than another literal. **One shape, and it is the argument for having a scale at all.** `RoundedCornerShape(30.dp)` in `TextNoteEventDetail` was the only hand-written corner off the M3 scale, at 30dp against `extraLarge`'s 28. Two units: invisible beside any single other card, and exactly the drift that happens when the value is a literal. It is now `MaterialTheme.shapes.extraLarge`, the first call site for the scale `Shape.kt` documented. **Rewritten by a script that reads call shapes, not values, and it is checked in.** `docs/scripts/m3-migrate-spacing.py` brace-matches three call shapes -- `padding(...)`/ `PaddingValues(...)`, `Arrangement.spacedBy(...)`, and a `.height()`/`.width()` whose enclosing call is `Spacer(` -- and rewrites only literals that fall inside one. A `.size(18.dp)` icon, a non-Spacer `.height()`, a `RoundedCornerShape` or a `BorderStroke` can never be caught, which a regex over `\\d+\\.dp` would have done to all of them. It inserts the two imports where they are missing and skips comment lines. Dry run by default. **The audit was measuring the wrong thing, and this is where that showed.** It split literals by value against a hardcoded `DIMENSION_EXEMPT` list -- and the split is not a property of the value. `16.dp` is a spacing stop *and* a plausible icon size. `50.dp` was a `Spacer` height in 52 places and a divider width in one, and no list of numbers separates those. `docs/scripts/m3-spacing-positions.py` replaces it with the same brace-matching parse the migration uses, so the audit and the migration agree by construction; the audit now reports **353 spacing literals** left and 76 dimensions out of scope, and the exemption table is gone. That reframes phase 2's acceptance criterion into something checkable: spacing positions to zero, dimensions untouched. The script exits 1 while any spacing literal remains. **What is left off-scale, and why none of it is a defect.** Twelve dimensions: avatar sizes at 35, 55, 70 and 75dp, icon sizes at 18 and 22dp, and a 50dp divider width. Avatar and icon sizing is a component-spec question rather than a spacing one -- M3 gives icons 18/20/ 24/40/48 and says nothing about avatars -- and the plan puts per-component specs after the adaptive phase. They are reported rather than exempted so the number stays visible. **Tests.** 942 pass, 594 jvm over 72 classes and 348 android over 44, unchanged -- this commit adds no assertions, and the ones it could add (`SpacingScaleTest`) landed with the scale. `:composeApp:compileDebugKotlinAndroid` builds, `m3-audit.sh --check` exits 0. Pixels move by at most 2dp, in 30 files. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-08 00:32:27 +02:00
Kgothatso Ngako	47bdb6f976	fix: promote the two pill colours to extended roles, fixing both contrast failures ... Phase 1, step 5 of docs/material-design-conformance.md. `BluePill` and `RedPill` were raw `Color` values in `Color.kt`, paired at the call site with `Color.White` and `Color.DarkGray` by eye. Both pairings were below the 4.5:1 floor, and one of them was not the colour it looked like. **This step could not leave the pixels alone, and it is the only one so far that changes them.** `Color.DarkGray` on `BluePill` measures **2.90:1**. `RedPill` was `Color(230, 32, 32, 191)` -- the four-Int constructor, whose last argument is alpha, so it is `#E62020` at 0.749. Opaque, white on it is 4.57:1 and passes; composited over the surface as it actually renders, it is **3.50:1** and does not. Any correct version of these two buttons is a visible change, so "adds, does not restyle" does not apply here and the plan already said the call sites would move in this step. **What M3 asks for here is an extended colour**, not a literal: a brand colour promoted to a full role family -- `color` / `onColor` / `colorContainer` / `onColorContainer` -- so that contrast is a property of the family rather than a decision repeated at each use. `ColorFamily` was already declared in `Theme.kt`, unused, alongside an `unspecified_scheme`; Material Theme Builder emits both, and this is what they are for. **Derived by the same rule as the gold palette**, which the fixed-roles commit established and verified: maximum in-gamut chroma at the source colour's Lab hue, sampled at M3's role tones. BluePill's hue is 277.0 and RedPill's is 36.3. role light dark blue light red light color tone 40 tone 80 #0060AB #C00012 onColor tone 100 tone 20 #FFFFFF #FFFFFF colorContainer tone 90 tone 30 #D7E2FF #FFDAD3 onColorContainer tone 10 tone 90 #001C39 #390C00 The buttons take `color`/`onColor`: 6.46:1 for the red pill and 6.44:1 for the blue, from 2.90 and 3.50. **A side effect worth having.** At tone 40 the two pills are the same lightness, so they now read as a matched pair. Before, `#E62020` sat beside `#5D8DD6` -- a saturated red next to a soft periwinkle -- and the blue looked like the lesser option. On a screen whose whole content is "commit, or wipe and leave", weighting one choice by accident is a defect of its own. **They travel on a composition local, not on `isSystemInDarkTheme()`.** `ColorScheme` has no slot for extended colours, so `LocalExtendedColors` is provided by `TorchTheme` from the same `darkTheme` it chooses the scheme with. Reading `isSystemInDarkTheme()` at the call site would have been one line shorter and subtly wrong: it ignores a caller who passed `darkTheme` explicitly, so a preview forcing dark would show light pills. The local defaults to the light families rather than to `unspecified_scheme` -- nothing composes outside `TorchTheme` today, and an invisible button is a worse way to discover that than a light-themed one. **No medium- or high-contrast variants, deliberately.** The entire surface is two buttons on one screen, and the light family's weakest pair is 6.44:1 -- clear of the floor by more than the contrast schemes would add. 32 more values for that would be out of proportion, and the comment in `Color.kt` says so rather than leaving the omission to be read as an oversight. **`QRCodeView` lost its constructor default.** `QRCodeBackgroundPainter` defaulted `backgroundColor` to `BluePill` -- a colour picked outside the theme for a surface that is almost never seen, since at the default `padding = 0.dp` the logo painter covers the rect it fills. The default is gone and the one call site passes it, so the choice is visible rather than buried. **Two new assertions, one of which is about the constructor.** `ColorSchemeContrastTest` grows to 9. The first checks both pairs of every extended family at 4.5:1. The second checks that every extended role is **opaque**, because `RedPill`'s alpha is what made the first assertion insufficient: a translucent container has no ratio of its own -- it has one only once composited -- so a contrast test would have measured a colour the user never sees. That is the bug this commit fixes, and it would have passed a naive contrast test. **The audit stopped counting its own commentary.** Fixing these call sites left a comment *explaining* what `Color.White`/`Color.DarkGray` had been, and `m3-audit.sh` counted it as a hardcoded colour -- so the file stayed in the report after being fixed. The script now drops comment lines before counting. Budget ratcheted 11 -> 9: the two real sites, plus the false positive the filter removes. **Tests.** 930 pass, 588 jvm over 71 classes and 342 android over 43, up from 926/586/340. `:composeApp:compileDebugKotlinAndroid` and `:composeApp:compileKotlinJvm` build, `m3-audit.sh --check` exits 0. The nine remaining hardcoded colours are phase 3's, and are listed by the audit. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-08 00:23:31 +02:00
Kgothatso Ngako	52b57769e1	feat: adopt MaterialExpressiveTheme, and give shape, type and motion a named home ... Phase 1, steps 3, 4 and 6 of docs/material-design-conformance.md. `TorchTheme` passed `MaterialTheme` a colour scheme and a typography and nothing else, so shape and motion were whatever the library defaulted to and there was nowhere to write down what any of it was for. **Expressive, by decision rather than by drift.** The plan deliberately left `MaterialExpressiveTheme` vs `MaterialTheme` open, because it changes component defaults app-wide and is a product call. Put to the product owner on 2026-09-08 and answered expressive. The pinned material3 1.10.0-alpha05 ships the whole set -- `ButtonGroupKt`, `SplitButtonKt`, `FloatingToolbarKt`, `LoadingIndicatorKt`, `ShortNavigationBarKt`, `WideNavigationRail`, `MaterialShapesKt` -- and the tree already opts into `ExperimentalMaterial3ExpressiveApi` in 66 places, so this makes explicit what the imports had already assumed. **All four slots are passed explicitly, and that is the point.** `MaterialExpressiveTheme` defaults its colour scheme to `expressiveLightColorScheme()` and its shapes and typography likewise -- Material's values, not this app's. Leaving any slot to that default is the same class of accident as the twelve unassigned fixed roles fixed two commits ago: it compiles, it renders, and it renders somebody else's design. **No visual change on the screens checked, and that is worth stating rather than assuming.** Measured on the API 36 emulator: the "Invite a Friend" button is byte-identical before and after -- same fill `#4E5E8B`, same 357px box at the same y -- because the expressive default for a `Button` at default size matches the baseline in this version. What expressive actually buys is elsewhere: `LocalUsingExpressiveTheme` gating component behaviour, the three increased shape steps, the fifteen `...Emphasized` type roles, and the components phases 5 to 7 are built on. **`MantraShapes` is baseline `Shapes()`, on evidence.** The corners hand-written across the tree already land on the M3 scale -- RoundedCornerShape(4.dp) x3 = extraSmall RoundedCornerShape(12.dp) x11 = medium RoundedCornerShape(16.dp) x2 = large RoundedCornerShape(30.dp) x1 ~ extraLarge (28dp) -- so overriding the scale would restyle the app for no reason. What is wrong is that they are literals, which is how the last one drifted two units off the scale and why none of them can move per breakpoint later. `Shape.kt` documents the eight steps and what each is for; migrating those seventeen call sites is a later phase, and this is what they migrate onto. Declaring it explicitly rather than relying on the default gives the note somewhere to live. **`MotionScheme.expressive()` is wired and unused.** Nothing in the app animates today -- one `animateScrollToPage`, no `AnimatedVisibility`, no navigation transitions -- so this buys nothing yet. It is here so that when the motion phase starts, every spec comes from the scheme rather than from a literal `tween`, and the app's feel is one decision instead of forty. **`Type.kt` left `com.example.ui.theme`.** It has been declaring that package while living under `press/mantra/compose/ui/theme/`, one of three namespaces holding live UI code in this tree. The move is mechanical; the doc comment on it is not. It records what each type family is *for* -- `display*` for a screen's identity, `headline*` for section tops, `title*` for headers and list headlines, `body*` for anything read as a sentence, `label*` for **component text only** -- because the audit's finding is not that the scale is wrong but that 92 of 240 reads are `label*` while `display*` and `headline*` carry 9 between them across 43 screens. A UI at one pitch. The file stays baseline; the rule now has a home for the sweep that fixes the call sites. **The desktop unlock screen renders in the app's theme for the first time.** `PassphraseGate` sat in the `else` branch beside `MantraApp`, which applies `TorchTheme` itself -- so the gate composed under the default `MaterialTheme` and its `colorScheme.error` and `typography.headlineSmall` were baseline M3. It is the first screen a desktop user sees. `TorchTheme` now wraps both branches. That wraps the unlocked branch twice, deliberately. `MantraApp` keeps its own `TorchTheme` because android and ios enter through it and would lose the theme entirely if it moved out; a second application of identical values costs one `CompositionLocalProvider` composition. The comment says so, since the redundancy looks like an oversight. **Dynamic colour stays on, by decision.** Also put to the product owner: on Android 12+ `dynamicColor = true` wins unconditionally, so the six schemes are used only below Android 12, on ios and on desktop, and a modern phone paints the wallpaper palette. Answered keep as-is. A comment on the selection in `TorchTheme` now says this outright, because otherwise the next person to change `Color.kt` and see nothing happen on their phone will assume the change did not work. **Tests.** 926 pass, 586 jvm over 71 classes and 340 android over 43, unchanged -- this commit adds no assertions, because what it changes is either a library default (nothing to assert that the compiler does not) or a doc comment. `:composeApp:compileDebugKotlinAndroid` and `:composeApp:compileKotlinJvm` build, the debug apk installs and runs on emulator-5554 under the expressive theme, `m3-audit.sh --check` exits 0. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-08 00:17:41 +02:00
Kgothatso Ngako	21d57eba54	feat: honour the platform's contrast setting, reaching four schemes that were dead code ... Phase 1, step 2 of docs/material-design-conformance.md. `Color.kt` has carried medium- and high-contrast variants of both themes since it was generated -- 156 colour values, wired into `lightColorScheme`/`darkColorScheme` in `Theme.kt`, and never selected. `TorchTheme` chose between `darkScheme` and `lightScheme` and nothing else, so a user who turned contrast up in Accessibility settings got no change at all. M3's accessibility foundation leads with *honour individuals*: "supporting varying preferences and choices that allow individuals to address how their changing conditions, individual knowledge, and varying needs are met." The work to do that was already done and disconnected. **The expect/actual boundary moved, because it was in the wrong place.** `themeColorScheme` took four arguments and did two unrelated jobs -- decide the contrast-free light/dark scheme, and decide whether to prefer a wallpaper palette. Adding contrast to it would have meant passing six schemes across the boundary and repeating the selection table in three actuals. It splits instead into `platformThemeContrast()` and `dynamicColorScheme()`, each answering one narrow platform question, with the six-way table as a plain function `appColorScheme(darkTheme, contrast)` in common code. `dynamicColorScheme` returns null rather than falling back internally so the fallback stays in one place. **Android reads the setting and listens for changes.** `UiModeManager.getContrast()` is API 34; the app's minSdk is 26, so below that the answer is Standard. The float is snapped to the nearest of the platform's three documented positions rather than matched exactly, so a future finer-grained slider degrades to the closest scheme this app has instead of falling back to Standard. The `ContrastChangeListener` is the part that is easy to leave out and matters most. A contrast change does not restart the activity and does not arrive as a `Configuration` update, so without it the new setting would take effect on the next cold start -- which is precisely the case the setting exists for. `context.mainExecutor` rather than `ContextCompat.getMainExecutor`: it needs API 28, this branch is already gated on 34, and composeApp does not declare androidx.core -- it only arrives transitively through activity-compose, which is not a dependency to lean on. **iOS observes the notification for the same reason** -- `UIAccessibilityDarkerSystemColorsEnabled` plus `UIAccessibilityDarkerSystemColorsStatusDidChangeNotification`. It is a boolean, not a slider, so iOS reports High or Standard and never Medium. **Desktop is honest rather than complete.** Windows publishes high contrast as the `win.highContrast.on` AWT desktop property and fires a property change when it is toggled, so that path is real and live. macos "Increase contrast" and the linux desktop equivalents do not reach AWT, and reading them means a native call per platform, so on those two the answer is Standard and the file says so. This is the right place for a user-overridable preference later; a desktop app cannot always see what the desktop was told. **Verified on an emulator, at the pixel.** API 36, dynamic colour temporarily switched off (see below for why that is necessary), sampling the `onPrimaryContainer` pixel of the "Skip for now" label as `settings put secure contrast_level` moved: standard (0.0) #848484 onPrimaryContainerLight medium (0.5) #A7A7A7 onPrimaryContainerLightMediumContrast high (1.0) #D0D0D0 onPrimaryContainerLightHighContrast The three declared values exactly, and **the app was not restarted between them** -- only the setting changed, four seconds apart. That is the listener working end to end. The probe that switched dynamic colour off is reverted in this commit; the emulator's contrast_level is back at 0.0. **A finding that came out of the verification, and is not fixed here.** `TorchTheme` defaults `dynamicColor = true`, and on Android 12+ dynamic colour wins unconditionally -- so on essentially every current Android device **none of the six schemes is used at all** and the app renders in whatever the user's wallpaper produced. The first screenshot of this session shows the onboarding screen in Material lavender; switching dynamic colour off reveals the black-and-gold brand for the first time. Nobody on a modern Android has been seeing this app's palette. That is a product decision, not a conformance one, so it is recorded in the plan's "What this plan does not cover" rather than changed. It does bound what this commit buys: on Android 14+ with dynamic colour on, contrast is honoured by the platform anyway (the `system_*` resources shift with it, confirmed on the same emulator -- buttons went slate-blue to near-black navy). What this commit reaches is Android below 12, Android 12-13, iOS, and desktop. **Three new assertions.** `AppColorSchemeSelectionTest` covers the table itself, because its failure mode is silent and specific: a scheme wired to the wrong cell still renders a complete, plausible UI, and somebody who turns contrast up and gets the medium scheme back cannot tell it apart from a high-contrast scheme that is not very high. It asserts each of the six cells by identity, that all six are distinct objects (a copy-paste leaving two cells on the same scheme would pass the first test only if it also mislabelled one), and that `onSurface` on `surface` never *falls* as contrast rises -- the one direction that must hold, and deliberately not the full monotonicity assertion that ColorSchemeContrastTest explains is false. **Not compiled: the iOS actual.** The ios targets are declared only on macos (see docs/jvm-target.md), so `Theme.ios.kt` is written against the UIKit and Foundation bindings rather than checked by a compiler. Its file comment says so. The android and jvm actuals of the same two functions are compiled, and the android one is verified on a device. **Tests.** 926 pass, 586 jvm over 71 classes and 340 android over 43, up from 920/583/337. `:composeApp:compileDebugKotlinAndroid` and `:composeApp:compileKotlinJvm` build, `m3-audit.sh --check` exits 0. The 54 existing `TorchTheme { }` call sites are untouched -- the new parameter is defaulted. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-08 00:10:06 +02:00
Kgothatso Ngako	86c9628eee	fix: assign every ColorScheme role, so no component can fall back to Material lavender ... Phase 1, step 1 of docs/material-design-conformance.md. `Theme.kt` assigned 36 of the 49 roles `androidx.compose.material3.ColorScheme` declares. The other thirteen took `lightColorScheme()`/`darkColorScheme()` defaults, and for twelve of them that default is the Material baseline palette: `primaryFixed` -> `ColorLightTokens.PrimaryFixed` -> `PaletteTokens.Primary90` -> **#EADDFF**. Lavender, in an app whose primary is `#000000`, in both themes, in all six schemes. Nothing in the tree reads a fixed role today, which is why nobody has seen it. That also means it could not have been found by looking at the app -- it springs the first time an expressive component reaches for one, and it will look like a rendering bug rather than a missing assignment. **The tones were computed, not chosen.** M3 defines the family by tone: `xFixed` = tone 90, `xFixedDim` = 80, `onXFixed` = 10, `onXFixedVariant` = 30, and ColorLightTokens and ColorDarkTokens carry identical values for all twelve -- theme-independence is what "fixed" means. Tone is CIE L*, so for a chroma-0 palette a tone is exactly the sRGB grey at that L*, and inverting L* -> Y -> sRGB reproduces this palette's own greys **to the byte**: tone 0 #000000 primaryLight tone 10 #1B1B1B primaryContainerLight, onSurfaceLight tone 20 #303030 onPrimaryDark, inverseSurfaceLight tone 40 #5E5E5E inversePrimaryDark tone 80 #C6C6C6 primaryDark, inversePrimaryLight tone 90 #E2E2E2 onSurfaceDark, surfaceContainerHighestLight tone 95 #F1F1F1 inverseOnSurfaceLight tone 100 #FFFFFF onPrimaryLight Eight independent hits. The primary and tertiary palettes are the standard M3 neutral tonal palette at chroma 0, so their fixed families are derived rather than invented. **The secondary palette is gold at Lab hue 87.5 degrees, and its dark half is maximum in-gamut chroma at that hue.** Generating tones off that ramp regenerates `onSecondaryDark` (#3D2F00, tone 20) and `secondaryLight` (#745B00, tone 40) byte for byte, which is what licenses using it for tones 10 (#241A00) and 30 (#584400). Its tones 90 and 80 are **reused rather than regenerated**. The palette already ships #FFDE82 at tone 90 (as `secondaryDark`) and the brand gold #EFBF04 at tone 80 (as `secondaryContainer`, identical in light and dark -- someone hand-set it, no generator emits that). Regenerating would have produced #FFDF99 and #F1C100: a second gold two units from the one already on screen, indistinguishable in isolation and wrong beside it. A near-duplicate brand colour is worse than none. **Sanity check on the whole derivation.** The four ratios these families produce land within 0.1 of M3's own baseline fixed family -- onFixed on Fixed 13.30 (baseline 13.32) onFixedVariant on Fixed 7.17 (baseline 7.23) onFixed on FixedDim 10.08 (baseline 10.08) onFixedVariant on Dim 5.44 (baseline 5.47) -- because tone, not hue, sets the ratio. Two palettes with nothing in common landing on the same four numbers is the check that the tone mapping is right. **Containers hold across the contrast setting; content darkens.** That is the move `Color.kt` already makes everywhere else -- `onSurfaceLight` goes #1B1B1B -> #111111 -> #000000 while `surfaceLight` stays #F9F9F9 through all three -- so the fixed family follows it: content tones 10/30, then 5/20, then 0/10. The weakest pair ladders 5.44 -> 7.73 -> 10.08. Shifting the containers instead would have moved the brand-visible half for a setting that is about legibility. **`surfaceTint` is the thirteenth, and it was never a defect.** Its default is `primary`, which is correct: `surfaceColorAtElevation` composites it over `surface` at 2-8% alpha, so an elevated light surface darkens toward primary and an elevated dark one lightens -- M3's own behaviour, and this app sets no elevations anywhere, so nothing reads it. It is assigned explicitly anyway, with that reasoning in a comment, so that "every role is assigned" is a property a reader can check by looking rather than by knowing which omissions were deliberate. m3-audit.sh reports the two kinds apart for the same reason. **Three new assertions, and the two that matter cannot be satisfied by accident.** `ColorSchemeContrastTest` grows from 4 to 7: - both content roles on both fixed containers at 4.5:1, across all six schemes; - the fixed roles are the same colour in light and dark, which is the definition and would otherwise only fail on a screen that puts one beside a themed surface; - no role is left at the Material baseline palette -- the twelve baseline hex values read out of `PaletteTokens.kt` and asserted absent. Verified by deleting `primaryFixed = primaryFixed,` from `lightScheme` alone: two tests fail, naming the role and printing back `Color(0.917, 0.866, 1.0)`. Reverted. **Audit budget ratcheted 12 -> 0**, dated in the file. Per the header's contract that is the only direction a budget moves, and the commit that lowers it is the one that earns it. **Tests.** 920 pass, 583 jvm over 70 classes and 337 android over 42, up from 914/580/337 -- three new assertions counted once per target. `:composeApp:compileDebugKotlinAndroid` builds, `m3-audit.sh --check` exits 0. No visual change: every role that had a value keeps it, and the thirteen that gain one were rendering baseline defaults nothing reads yet. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-07 23:59:34 +02:00
Kgothatso Ngako	2b0ce8d73b	test: measure M3 conformance instead of asserting it, with a budgeted audit and a contrast test ... Phase 0 of docs/material-design-conformance.md. Every count in that document was produced by hand, which makes the eight phases after it opinions rather than work with acceptance criteria. This is the harness that turns them back into numbers. **`docs/scripts/m3-audit.sh` regenerates the whole audit, and can fail a build.** Plain invocation reports; `--check` exits 1 when a budget at the top of the file is exceeded. The budgets are the tree as it stands -- 11 hardcoded colours, 33 bare `.clickable`, 18 null content descriptions, 12 unassigned colour roles -- and the contract written into the header is that they ratchet **down**, in the same commit that earns the reduction, and are never raised. Counts a phase has not reached yet are `-1`, which reports but never fails. Phase 8 wires `--check` into CI, at which point a raised budget is the diff a reviewer is looking for. Verified both directions: `--check` exits 0 on the clean tree, and appending a single `Color(0xFF00FF00)` to LoadingScreen.kt makes it exit 1 naming the budget. **Two counts are reported apart from each other on purpose.** Thirteen ColorScheme roles are never assigned in Theme.kt, and reporting that as one number would overstate it. Twelve are the `*Fixed*` family, which default to `ColorLightTokens.PrimaryFixed` -> `PaletteTokens.Primary90` -> `#EADDFF`, so a monochrome app renders Material baseline lavender the moment anything reads one. The thirteenth is `surfaceTint`, whose default is `primary` -- correct, and not a defect. The script labels the first group "lavender" and the second "not a defect". The `.dp` histogram splits three ways for the same reason. 527 literals: 419 on the M3 spacing scale, 19 dimensions rather than spacing (a 1dp hairline, an avatar, an image height), and 89 genuinely off-scale. The naive split reported 101 off-scale by counting 1dp borders as bad spacing, which would have sent phase 2 chasing hairlines. `DIMENSION_EXEMPT` is deliberately short and the header asks for a justification in the commit that lengthens it. **`ColorSchemeContrastTest` walks the real schemes, which cost a visibility keyword.** Four assertions over all six declared schemes: every content role on its container at 4.5:1, `onSurface` on each of the seven tonal surfaces at 4.5:1, `outline` against every surface it is drawn on at 3:1, and `primary`/`error` against `surface` at 3:1. WCAG relative luminance from first principles -- the 0.03928 knee and the 2.4 exponent, not a gamma-2.2 approximation, because the approximation moves borderline pairs by enough to change a verdict and the tightest pair in this tree is 4.56:1. `Theme.kt`'s six schemes went from `private val` to `internal val` so the test can see them. The alternative -- rebuilding the schemes inside the test from `Color.kt`'s public values -- keeps production visibility untouched and was rejected: it would assert the palette and miss the wiring, and the wiring is the half that fails silently. `surfaceContainerHigh = surfaceContainerHighestLight` is a one-character slip, compiles, and reads fine in review. A comment above the first scheme says this, so the keyword is not quietly widened back. **Verified that it bites.** Nudging `onSurfaceVariantLight` from `#4C4546` to `#9C9496` -- a plausible "soften the secondary text" edit that nothing else in the build would object to -- fails with `light: onSurfaceVariant on surfaceVariant is 2.29:1`, naming scheme, pair and ratio. Reverted; the committed value is unchanged. **Monotonicity across the contrast ladder is deliberately not asserted.** The obvious invariant -- high-contrast beats medium beats default for every pair -- looks right and is false. Ten pairs move the other way, and correctly: in the light high-contrast scheme `surfaceContainerHighest` goes darker to separate it from `surface`, which drops its ratio against `onSurface` from 13.30 to 12.29 while raising the separation that the change exists for. `onErrorContainer on errorContainer` drops 7.24 -> 5.19 from default to medium for the same kind of reason. Asserting the ladder would have meant either a red test or nine exemptions; the floor is the real invariant and every one of those values is comfortably above it. The test's doc comment records this so the next reader does not add the assertion. **Also not asserted: `outlineVariant`, and the call sites.** `outlineVariant` reads 1.61:1 against surface, which looks alarming and is not a defect -- M3's own baseline sits in the same range and the role is a decorative divider, so `outline` is what gets the 3:1 assertion. The seven call-site pairings that are genuinely below threshold, including the 1.00:1 one in ProposalListScreen, belong to phase 3; adding them now would mean checking in a red test. **Doc reconciled to the script rather than the other way round.** Three hand counts were wrong and are corrected in docs/material-design-conformance.md: 520 `.dp` literals -> 527 (the earlier figure omitted the exempt dimensions), 90 `label*` typography uses -> 92 (it missed `labelSmallEmphasized` and `labelLargeEmphasized`, which are label roles too), and 101 off-scale -> 89. The phase 0 section is rewritten from a plan into what was built, including what was decided against. **Tests.** 914 pass, 580 jvm over 70 classes and 334 android over 42 classes, up from 906/576/69 and 330/41 -- the four new assertions, in one new class, counted once per target because commonTest flows into both. `:composeApp:compileDebugKotlinAndroid` builds. No app behaviour changes: the only production edit in this commit is `private` -> `internal` on six vals. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-07 23:51:34 +02:00
Kgothatso Ngako	1c02b25a07	docs: measure the UI against the M3 foundations, and phase the work that follows ... A plan, not a change: what m3.material.io/foundations asks for as of its May 2026 revision, what these 43 screens actually do, and eight phases ordered so that each one makes the next mechanical rather than judgemental. **The spec was read, not remembered.** m3.material.io is a client-rendered SPA -- WebFetch returns an empty `<main>` and the tab URLs 404 on direct navigation -- so the numbers here came out of a real browser session clicking through the tab controls. That mattered: the May 2026 revision renamed window size classes to **breakpoints** and there are now five of them rather than three (compact / medium / expanded / large / extra-large, at 600 / 840 / 1200 / 1600dp), renamed responsive design to adaptive design, and published the spacing system as tokens on an 8dp scale where `space100 = 8dp`. Writing this from memory of older M3 would have produced a plan against a vocabulary the current spec no longer uses. **The palette is fine; the call sites are not.** Every `onX`-on-`X` pair in all six declared schemes clears 4.5:1, the tightest being `onPrimaryContainer` on `primaryContainer` at 4.61:1 light and 4.56:1 dark. So the generated scheme is not the problem and this plan does not propose a repalette. What fails is colour decided locally, seven pairings of it, and the worst is not visible to a reviewer: Card(colors = CardDefaults.cardColors(containerColor = primaryContainer)) { ListItem(colors = ListItemDefaults.colors(containerColor = Color.Transparent), `cardColors(containerColor = ...)` does derive `contentColor = contentColorFor(...)`, so `LocalContentColor` inside the card is correct. But `ListItem` does not read `LocalContentColor` -- its headline comes from `ListTokens.ItemLabelTextColor`, which is `onSurface` -- and the call site overrides only `containerColor`. In the light scheme `onSurface` and `primaryContainer` are both `#1B1B1B`. That is **1.00:1**, and it is applied exactly to `proposal.awaitsYou`, so the proposals waiting on your signature are the ones rendered invisible. `HomeScreen`'s `titleContentColor = primary` on `containerColor = primaryContainer` is the same mistake at 1.22:1. Ratios were computed rather than eyeballed; the script is in the Phase 0 deliverable. **Twelve colour roles fall through to Material baseline lavender.** `Color.kt` never assigns `primaryFixed`, `primaryFixedDim`, `onPrimaryFixed`, `onPrimaryFixedVariant` or the secondary/tertiary equivalents, so `lightColorScheme()` defaults them to `ColorLightTokens.PrimaryFixed` -> `PaletteTokens.Primary90` -> `#EADDFF`. Nothing reads them today, which is why it has never been noticed; the trap springs the first time an expressive component does. Read out of the pinned `material3-desktop-1.10.0-alpha05-sources.jar` rather than assumed. **Four of the six declared schemes are unreachable.** The medium- and high-contrast variants are written out in full in `Color.kt` -- 78 colour values -- wired into `lightColorScheme`/`darkColorScheme` in `Theme.kt`, and then never selected: `TorchTheme` chooses between `darkScheme` and `lightScheme` only. The work to honour a platform contrast setting is already done and disconnected. **10dp and 20dp are not the problem they look like.** They are the two dominant spacing values (132 and 115 uses) and both are *on* the M3 scale, as `space125` and `space250`. The plan says so rather than proposing a sweep that would change nothing. What is wrong is that none of the 520 `.dp` literals records whether it is padding, a gap or a margin -- the three categories the spec gives different rules to -- so nothing can be adapted per breakpoint later. About 101 are off-scale (50dp x 53, 15dp x 14, 5dp x 10 and so on), and `Modifier.height(50.dp)` appears 49 times as the same copied spacer above the same copied error message. **Findings that were measured and then dropped.** `outlineVariant` reads 1.61:1 against surface and `secondaryContainer` 1.65:1, both of which look alarming and neither of which is a defect: M3's own baseline sits in the same range, and the 3:1 rule the spec gives is for clustered interactive containers, not dividers or tonal surfaces. `onSurface.copy(alpha = 0.38f)` is the specified disabled opacity and the spec exempts disabled states from contrast entirely. Reporting these would have padded the count and cost the reader trust in the rest. **The rest of the audit, in counts.** 334 string literals in composables against 2 `stringResource` calls, with title case throughout ("Edit Profile", "New Chat") where the style guide asks for sentence case. Zero `Snackbar` across 26 `Scaffold`s. 16 copies of `Text("Something went wrong")`, none of which offers a retry. 90 of 240 typography reads on `label*` roles, which are for component text, while `display*` and `headline*` carry 9 uses between them across 43 screens. 33 bare `Modifier.clickable` with no minimum target, two of them text-height. Two `BoxWithConstraints` and no window-size handling at all, on a project with a desktop target whose own entry point already says so in a comment. **Eight phases, ordered by what each unblocks.** 0 baseline harness, 1 theme, 2 spacing tokens, 3 accessibility floor, 4 content, 5 states and feedback, 6 adaptive layout, 7 motion, 8 guard rails. Tokens come before the call sites that consume them; the accessibility floor comes before the adaptive work that would otherwise double the surface to fix; guard rails come last so they lock in real state rather than aspiration. Phase 6 is the only one that cannot be done mechanically and the only one marked not reversible alone. **What it deliberately does not decide.** Whether the target is `MaterialExpressiveTheme` or `MaterialTheme` -- the pinned material3 ships the full expressive set and the code already opts into `ExperimentalMaterial3ExpressiveApi` in 66 places, but it changes default component shapes and sizes app-wide, so it is a product call and Phase 1 raises it rather than answering it. Also out of scope: whether the monochrome palette is right, the per-component specs, iOS (which only builds on a mac, and whose HIG asks 44dp where M3 asks 48dp), and the three package namespaces the UI currently lives across. No code changes. `docs/README.md` gains the row and the closing paragraph's note on how this one relates to the others. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-07 23:44:32 +02:00
Kgothatso Ngako	98f766fcd3	fix: put the invite in the room, so a stuck one can be seen ... Inviting a member to a group that already had members put nothing whatsoever in the transcript. Not "put it in late" -- nothing, and nothing ever if the invite did not complete. So the one failure the user is best placed to notice, an invite that never reached the person it was made for, was the one the app kept to itself. The line existed. It was written by `MarmotOutboundDao.deliveryWelcome`, which is the wrong place for it, and the reason is the two paths through `inviteMember` that docs/marmot-membership.md already describes. A group that is still only its creator has nobody to inform, so its Welcome goes out immediately and `deliveryWelcome` runs inside the invite. A group that has members must broadcast a commit first, and its Welcome waits for a relay to acknowledge it -- `DatabaseNostrRepository.broadcastProcessed` picks the stored `MarmotCommitResult` back up and delivers then. Every invite after a group's first therefore wrote its transcript line a relay round trip away from the invite, if at all. **Four separate silences, not one.** Worth listing because only the first is about the deferral, and fixing that alone would have left the other three: 1. The deferred path wrote nothing until the ack, and nothing ever without one. 2. The write hung off `getMarmotKeyPackageById(...)?.let { getProfileByPublicKey(...)?.let { ... } }`. Those two lookups were there to *name* the invitee, and a miss on either cost the whole line rather than just the name. 3. `deliveryWelcome` wraps its body in `catch (e: Throwable) { logger.e(...) }` and returned Unit, so a Welcome that could not be built reached the log and no further. 4. `inviteMemberToChatRoom` is `@Transaction`. An invite that threw -- no MLS state for the room, a credential identity that does not match the peer -- rolled its line back with everything else, which is right, and left no account of the refusal anywhere durable. And the line it did write was `messageType = "message"`, `isUserMessage = true`, so it rendered as a chat bubble: "Invited Bob to chat", attributed to the inviter as something they said. **Three membership types, and the line moves to invite time.** `ChatMessage.MEMBERSHIP_TYPES` -- `memberInvited`, `memberInviteSent`, `memberInviteFailed` -- rendered by the transcript as system notices through `RitualNotice`, the way the ceremony, signing and chronicle lines already are. `memberInvited` is written by `inviteMember` and by `addMembersToChatRoom`'s batch path, *when the invite is made*, and deliberately **inside** the caller's transaction. Both halves of that matter and they pull opposite ways: written any later and an invite waiting on an ack that never comes shows nothing, which is the bug; written outside the transaction and an invite that does not survive `addMember` leaves the room claiming one was made. `memberInviteSent` is written by `DatabaseNostrRepository` alone. It is not written on the immediate path, and that is not an oversight: there the Welcome goes out in the same breath as the invite, so one line is the whole truth. It would also be a line the transcript could not order -- `MantraConverters` stores `Instant` as `epochSeconds`, the room query is `ORDER BY createdAt DESC`, and two rows written in the same second tie. Only the deferred path separates the two events in time, so only it owes a second line. `memberInviteFailed` carries the reason, because it is the only copy the user gets. `deliveryWelcome` now returns `Boolean` and files this line from its own catch before returning false -- its callers had no other way to see a failure it had already swallowed, and on the deferred path there is no invite screen left to fail back to. `addMembersToChatRoom` reads that answer instead of a `runCatching` that could never catch anything. **The refusal is written from outside the transaction that rolled it back.** `DatabaseChatRepository.inviteMember` catches, calls `announceInviteFailed`, and rethrows. The throw is what puts a message on the invite screen now; the line is what is still there tomorrow. Swallowing it instead would have popped the user back to the chat as though the invite had gone out, which is the bug the existing `runCatching` in `AddMemberToChatRoomConfirmationViewModel` was added to stop. **No schema change.** `messageType` is a free-form string column with a default, so new values need no migration -- unlike the chronicle rename, which had to rewrite the ones already stored. Nothing reindexes these either: they carry no `marmotGroupEventId`, so `getResolvedMarmotGroupEventIds` cannot see them and `UNRESOLVED_MARMOT_TYPES` does not name them. Six new tests. Four on the DAO: the immediate path leaves a line naming the invitee where the old code left none, the deferred path leaves one *and* claims no Welcome sent before any ack, everything an invite writes is a membership type rather than something the transcript would render as a bubble, and a refused invite leaves no claim that one was made. Two new ones on `DatabaseChatRepository`, which had no test file: a refused invite is written into the room, and the caller still gets the throw. Still open, and now said plainly in the doc rather than implied: a Participant row carries no state saying where its invite got to. The transcript narrates it; the `TODO: Update status of participant Invitation.PENDING -> Invitation.SENT` is untouched. Nor does an invitee with no published key package reach the room at all -- that fails in the view model, before there is an invite to write a line about. 806 tests pass -- 509 jvm, 297 android. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 18:16:26 +02:00
Kgothatso Ngako	19d57ef3a5	Merge branch 'mantra' into claude/happy-gauss-dbe258 ... Brings in the Chronicle rename and the deprecation of the row rebuild, and carries the supersession fix across into the new vocabulary. Git followed every rename on its own -- `ArchiveManager` -> `ChronicleManager`, the tests, the docs -- and auto-merged all three files my fix had touched. What it could not do is rename identifiers inside the hunks it merged, so the fix arrived speaking the old language: `ChronicleAssemblyJvmTest` still called `ArchiveManager.assemble` and `ArchiveEvent.decodePage`, which does not compile, and six doc comments in `ChronicleManager` and `GroupSignedEvent` still said "archive" -- the exact ambiguity with archiving a chat that the rename exists to remove. One real conflict, in the design note, and it is the same sentence twice: my correction of "a retranslated passage archives once" against the rename of the uncorrected claim. Resolved to the correction, in the new vocabulary -- the property still holds, it just stopped being free the moment the chronicle was read from `GroupSignedEvent` rather than rebuilt from rows, and `ChronicleManager.currentTranslationsOnly` is what holds it up. `compileKotlinJvm` passes over a test file that does not compile, so it was no evidence here; `compileTestKotlinJvm` is. And the filter was re-checked the way it was written: removing it fails the same three tests, so the merge did not quietly neuter them. 503 jvm tests and 297 android unit tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 16:37:12 +02:00
Kgothatso Ngako	4890906b24	Merge branch 'mantra' into claude/rename-archive-chronicle-a4a8e0 ... The rebuild deprecation landed on mantra while the rename was in flight, and it touched the same files by their old names. Git matched the renames itself, so the only conflict was `ChronicleRoundTripTest`'s header, where both sides had rewritten the same paragraph: mantra's says this file is now the gate on a deprecated fallback rather than on the only path, which is the newer and truer claim, so it wins and the rename is applied on top of it. Everything the merge brought in went through the same substitution as the rest: the nine `@Deprecated` messages and the "Retiring the rebuild" checklist all name `ChronicleManager`, `ChronicleRoundTripTest` and docs/member-chronicle.md, which are the files that now exist. 797 tests pass -- 500 jvm, 297 android. The five new ones are the migration's. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 16:29:23 +02:00
Kgothatso Ngako	ea11e8b233	refactor: call it a chronicle, and keep "archive" for what a user does to a chat ... Archiving a chat is an ordinary thing a user will want to do to a conversation, and it is not this. This is the group's signed record, handed to a member who joined after the work was done so their room stops being empty. Two unrelated meanings of one word in one app is a bug waiting to be written, and `ChatRoom.archiveRequestedAt` is exactly where they would have met: a column on the chat row, named for the thing that is not the chat. So the whole feature is Chronicle now -- `press.mantra.compose.nostr.chronicle`, `ChronicleEvent` (30327), `ChronicleRequestEvent` (30328), the three tags, `ChronicleManager`, `docs/member-chronicle.md`. The kind numbers do not move; only the words do. **The wire tags move too**, `archiveId` -> `chronicleId` and `archivePage` -> `chroniclePage`, which is free exactly once. Both kinds are new and there is no old build to stay compatible with -- the design note says so in as many words -- so the alternative was carrying the old spelling on the wire forever to save a rename that costs nothing today. The recipient tag stays `p`; it was never ours. **Schema v14, because two things had the old word written into stored data.** `ChatRoom.archiveRequestedAt` becomes `chronicleRequestedAt`, renamed rather than dropped and re-added: while it is set it is the only record that a device with an empty room has already asked the group for its history, and a device that lost it mid-flight would ask again on its next launch, and the one after that. The three `ChatMessage.messageType` strings become their `chronicle*` spellings, rewritten rather than left to a legacy constant the way `dkgApprovalNeeded` was. These lines cannot be regenerated -- a chronicle is announced once, when it is requested, sent and applied -- and an unrecognised type is not skipped by the transcript. It renders as an ordinary chat bubble, so "Caught up on 12 items" would come back attributed to a member as something they said. `MIGRATION_13_14` does both, because Room can rename a column and cannot rewrite rows in the same breath. `ALTER TABLE ... RENAME COLUMN` needs SQLite 3.25, which `getRoomDatabase` guarantees by pinning `BundledSQLiteDriver`, and the column is in no index, no foreign key, and there is not a view or trigger in the database -- so nothing has to move with it. Five tests hold the two halves apart: the value survives, the column keeps its position, a room that never asked still reads as never having asked, the three types are rewritten, and every other type is left alone. **`isArchivable` is `isChroniclable`**, on the "recyclable" pattern, and it keeps its job unchanged: the allowlist that stands between a replayed `GroupKeyStateEvent` and the apply path. No behaviour change beyond the rename. 797 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 16:27:36 +02:00
Kgothatso Ngako	87ef129de5	fix: archive the translation that stands, not every draft of it ... Follow-up to reading the archive out of `GroupSignedEvent` rather than rebuilding it from rows. The two sources do not hold the same thing, and one place where they differ reaches the archive. `ChatMessage.applyInnerEvent` supersedes a translation chunk: retranslating a passage changes the text and so the event id, so the arm drops the row it replaces -- newest by the timestamp the group signed at, id breaking a tie. The record does not, and should not: a signature is the group's statement and discarding one is not that table's business. So a passage translated three times leaves one row and three events. While the archive was rebuilt from rows that difference was invisible, because a sender simply had nothing but the group's current answer to each passage. Read from the record it is not: measured on a seeded room, one retranslation leaves one row and puts **two** payloads in the archive, and it compounds -- every draft a group ever signed would travel in every archive it ever sends, for as long as the room exists. **The rule is the applying arm's, restated rather than approximated.** An archive that shipped one translation as current while the recipient settled on another would have both validly signed and nothing downstream to notice they disagree, so `currentTranslationsOnly` groups by `(translationChapterId, chunkId)` and keeps the maximum by `(createdAt, id)` -- the same comparison, spelled the same way. Dropping the drafts is safe precisely *because* the recipient applies that rule too. This is not what keeps them correct; it is what stops them being sent work they would discard on arrival. Grouped per passage rather than per chapter, or retranslating one passage would take every other passage's translation with it. A translation naming no passage is left alone rather than lumped in with the rest: it is unappliable either way, and letting one stand in for a whole passage would let a malformed event suppress a good one. Three tests, and all three fail if the filter is removed: a retranslated passage leaves one row, two recorded events and one payload; three translations signed in the same second settle on the same id the row keeps, which is what pins the tiebreak to the applying arm's; and two passages each keep their own, which is what a group-by-chapter mistake would fail. Two documentation corrections alongside it. docs/member-archive.md said "a retranslated passage archives once" as a property of the rows, which stopped being true the moment the record became the source -- it now says what makes it true again. And `GroupSignedEvent.verifies()` said a false means the row's columns have drifted from the event they came from. That is the reading worth acting on and it is not the only one: a room never derived from its group's key signs as the bare threshold key rather than as its own id, so a perfectly good event there fails and cannot be made to pass, the key it would need being absent from the row and unreachable from one. 498 jvm tests and 297 android unit tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 16:19:12 +02:00
Kgothatso Ngako	c66f085681	refactor: deprecate the row rebuild, and write down what goes with it ... `assemble` reads `GroupSignedEvent` now and rebuilds from `Mantra*` rows only what that table does not hold, which is work signed before it existed. The rebuild is therefore on its way out rather than merely second in line, and this says so where a reader will actually meet it -- at the call site, from the compiler -- instead of only in a paragraph they have to find first. **Nine `@Deprecated` markers, and they are load-bearing as documentation.** The eight `toXEvent()` methods and `ArchiveManager.rebuiltEventsOf`, each carrying the same sentence: this is the fallback for pre-v13 work, read the event off the table instead, and it goes when the last such install does. That raises nine warnings in `commonMain` today, all of them inside the walk itself, so the deprecation is visible in every build without anything failing over it. The level is `WARNING` deliberately -- the code is still called, still correct, and still the only thing standing between an older room and an empty archive. **The checklist is a new section in docs/member-archive.md**, because the interesting part of this removal is not the eight methods, it is everything around them that is easy to take out by association or leave behind by accident. *What goes*: the walk and the version-label recovery inside it, the union in `signedEventsOf`, the eight rebuilds, and `ArchiveRoundTripTest` entire -- all ten cases, which exist to hold the rebuild up and cover nothing else. Its own header still opened with "signed events are not stored as events", which stopped being true two commits ago, so it now says what it is: the gate on a deprecated fallback, deleted with what it guards. *Two already-dead cousins to sweep at the same time*, named because they will look like part of the rebuild to whoever does the removal and are not: `MantraTranslation.toTranslationEvent`, which nothing has ever called, and `MantraTranslationChunkProposal.toTranslationChunkEvent`, on a model that is not even a `@Database` entity. *The tests that seed without recording*: in `ArchiveAssemblyJvmTest` the `apply`-only seeding **is** the rebuild path, and two of its cases are about the union specifically and mean nothing without it. `ArchiveApplyJvmTest` seeds its sender the same way but is testing delivery rather than assembly, so it needs the recording call *added* -- otherwise it quietly starts asserting against an empty archive, which is the same silent-success failure this whole feature is about. **What only looks like it goes, which is the half worth writing down.** The `isArchivable` filter in `signedEventsOf` is not part of the rebuild and becomes the only thing standing. It is there *because* of the record: the walk could only ever produce document kinds, so nothing needed filtering while it was the source, and the table holds every kind the group has signed -- starting with the `GroupKeyStateEvent` every room signs as its first act. Dropping it with the walk turns every room's archive into an `IllegalArgumentException` from `ArchiveEvent.build`. Two cases fail with exactly that if it goes, which is the guard against removing it by association rather than by decision. The verify filter in `assemble` stays too. With the rebuild gone it checks events that were verified before they were recorded, so it cannot fail in practice -- which is the argument for keeping it, not against. "Cannot happen" is the state it exists to preserve. `Mantra*.signature` and `Mantra*.publicKey` are explicitly *not* on the list. They were what made a row rebuildable, and since v13 `groupSignedEventId` says whether the group signed a row and points at the proof -- so they are arguably redundant. But four test files assert on them and `MantraTranslationContributor` builds a contributor list out of one, and it is a twelve-table migration with its own tests to rewrite. It should be decided on its own merits, not ride along. **The precondition cannot be checked, and the section says so plainly.** No query answers "does any install still hold pre-v13 work" -- a device that upgraded is indistinguishable from one that never had any, and the rows that need rebuilding are on other people's devices. What is observable is the `signedEventsOf` log line, which fires only when the rebuild actually contributed something; fleet-wide silence is evidence and not proof. The cost of getting it wrong is named as well, because it is not loud: the member keeps their own rows and reads the room normally, and only loses the ability to *answer* a request with the older half of the group's work -- so a newer member asks, is answered, and receives an archive that is quietly short. No behaviour change. 495 jvm tests and 297 android unit tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 16:06:17 +02:00
Kgothatso Ngako	a69d80d38f	feat: archive the events the group signed, not rebuilds of its rows ... `assemble` read the archive out of `Mantra*` rows, rebuilding each payload with `toXEvent()` and standing or falling on that rebuild being byte-identical to what was signed. It had to: nothing kept the events. `GroupSignedEvent` keeps them now, so `signedEventsOf` reads the record first and rebuilds only what the record does not hold. **The rebuild stays, as the fallback, keyed by id.** A room whose work predates v13 has no events on file, and dropping the walk would silently empty its archive -- the failure mode being that a member asks for the history, a member answers, and nobody notices the answer was blank. So both sources are read and unioned by event id, which is also what a half-upgraded room needs: older work only the rows remember, newer work on file, and neither half complete on its own. The fallback can go once no install still carries pre-v13 work, and `ArchiveRoundTripTest` is what holds it up until then. **The allowlist does real work on the way out now, and this is the part that would have bitten.** The rebuild could only ever produce document kinds, because those are the only rows it walks. The record holds every kind the group has ever signed -- and every room signs a `GroupKeyStateEvent` as its first act, so one is on file in every room that has signed anything at all. `ArchiveEvent.build` refuses a non-archivable kind with `require`, so an unfiltered read does not quietly ship a key state: it throws, and the room's entire archive fails on the one event every room has. `signedEventsOf` therefore filters on `isArchivable` before anything else, which is the same rule `applyPage` applies on the way in. Removing that one line fails two tests with exactly that exception, which is how I know they are load-bearing rather than passing for the reason I expected. **An artifact whose initial version row is missing now archives.** The rebuild has to recover the version label from that row -- `fromArtifactEvent` drops it, so it is not on the artifact -- and logs and gives up without it, which is a hole in the archive for any device that applied half a batch. Read from the record there is nothing to recover: the label never left the event. That is the case that makes the record the better source rather than merely the faster one, and it has a test of its own. **One verify filter over both sources**, because the rule is per event and not per source: nothing leaves that the recipient could not check for themselves. A drop still means different things on each side -- a member's own rumor sitting in the same table as the group's work, versus a row that has drifted from the event it recorded -- and the comment now says so, since the log line cannot. **Ordering is unchanged where it matters and looser where it does not.** `inApplyOrder` is a stable sort by dependency rank, so the union only affects order *within* a rank: a room holding some work both ways can order two chapters differently from a member holding one way only. Pages are idempotent and applied payload by payload, and two members already differed by the order their rows were written in, so this costs nothing. `rebuiltEventsOf` still runs on every archive even where it contributes nothing, because there is no way to tell a complete record from a partial one without doing the walk, and it is a handful of indexed queries against a room's own rows. 495 jvm tests and 297 android unit tests pass. Five new cases in `ArchiveAssemblyJvmTest`, which seeds through the real inbound path and now records the same batch the way `FrostSigningManager.complete` does: payloads compared byte-for-byte against what was signed, work held both ways travelling exactly once, a genuinely room-signed key state left behind, a signed kind the archive has no arm for left behind, and the artifact the rebuild has to leave out archiving from the record. The existing assembly and end-to-end tests seed without recording, so they go on covering the rebuild fallback unchanged -- which is why they all still pass, and why that is evidence rather than luck. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 15:51:16 +02:00
Kgothatso Ngako	8f9e4de82e	feat: keep what the group signed, and the path it signed as ... A quorum signing something is the most expensive thing this app does and, until now, the least recorded. `FrostSigningManager.complete` verified the signature, handed the event to `ChatMessage.applyInnerEvent`, and let it go. What survived was whatever the row it became happened to keep -- an artifact keeps its `signature` and `publicKey`, a translation contributor list keeps nothing at all because that arm is still a TODO, and a kind this build has no arm for keeps nothing anywhere. The signature is the group's statement; the rows are one reading of it. `GroupSignedEvent` is where the statement itself now lives, at schema v13 behind an `AutoMigration(12, 13)`. **The columns are `NostrEvent`'s, not a summary of one.** `id`, `publicKey`, `kind`, `tags`, `content`, `signature` and the event's own `created_at` as `createdAt`, so what is stored is an event rather than a description of one. That is what makes `verifies()` answerable from the row alone: it delegates to `GroupKeyStateEvent.isSignedByRoom`, which asks whether the author is the room, whether the id is the hash of the fields sitting next to it, and whether the signature checks out. No ceremony, no key state and no path have to be on hand first -- which is exactly the position a member added after the ceremony is in. **The derivation path is the point of the exercise.** `publicKey` is the group's threshold key walked to `derivationPath`, and for a room that walk is also the room id -- see docs/shared-key-derivation.md, where those are one value. Without the path there is no way back from a signature to the ceremony behind it: a threshold key alone does not say which of a group's rooms signed, and a room id alone cannot be walked backwards. `GroupKeyState` records the path for the room; this records it for the event, so an event stays checkable after the room's state is gone or was never known. Null means the untweaked threshold key, the same meaning it carries on `FrostSigningSession.derivationPath`, which is where the signing path is copied from -- resolved from the room by `signingPath`, never from a proposer. **Two writers, and both file only what they have already checked.** `FrostSigningManager.recordSignedEvents` files a whole batch in one write, after every item's signature has verified and before any of them is applied -- a session's events are one decision by one quorum, so half a batch on file is a state no reader should have to reason about. `ArchiveManager.applyPage` files each payload it accepts, after the allowlist and `GroupKeyStateEvent.isSignedByRoom`, reading the room's path once per page from `GroupKeyState` rather than once per payload. Neither failure is the caller's: recording throws are logged and swallowed, because a ceremony that succeeded must not be reported as failed over a row this device could not write down. **The archive half is what makes a recipient more than a dead end.** A member handed their history used to end up holding the rows and none of the events -- able to read the group's work, unable to prove any of it, and unable to build a page for the next member to arrive. Now the events land too. **`record` merges rather than overwrites, and that direction is deliberate.** The same event reaches a device twice by design: once when the session that made it completes, once from any archive page carrying it. The second arrival is the poorer one -- an archive knows no session, and on a member who joined after the ceremony no derivation path either -- so the incoming row fills gaps and never empties them. The event's own fields are not merged because they cannot disagree: the id is the hash of them, so two rows under one id either hold the same event or one of them is not the event it claims to be. **Every `Mantra*` row points back at it.** `groupSignedEventId` on all twelve entities that carry `marmotGroupEventId`, stamped by `ChatMessage.applyInnerEvent` through a new defaulted parameter. On a group-signed row it is the only provenance there is: both Marmot ids are null, because there is no group event and no inner event behind one -- a signed event authored by the threshold key cannot travel as an inner event at all, since the outbound pipeline re-authors rumors as their sender and would strip the signature off. The column is only set when the record actually landed, so a row never points at an event that is not there. **`ArchiveManager`'s own doc said something that is no longer true.** It opened with "signed events are not stored as events", stated as present-tense fact and load-bearing for the paragraph under it. Corrected there and noted at the head of the same section in docs/member-archive.md, which is a phase history and so gets a note rather than a rewrite. Assembly still rebuilds payloads from rows via `toXEvent()` and the round-trip gate still holds it up: a room whose work predates v13 has no events on file, and rebuilding is the only way to reach it. Reading assembled events from the table is worth doing once that fallback can be dropped. **Two things this deliberately does not touch.** `ChatMessage` gets no such column -- it is not a `Mantra*` row and already carries `frostSigningSessionId` for the lines that need to name a session. `MantraTranslationChunkProposal` has a `marmotGroupEventId` but is not a `@Database` entity and nothing in `composeApp/src` references it, so it was left as the dead code it is rather than grown a column. Rows are not backfilled by the migration. The events they came from are gone, and minting an id for one would point a row at a signature nobody can produce; null reads as "this device does not hold the event behind this row", which is true of every row written before today. 490 jvm tests and 297 android unit tests pass. `GroupSignedEventDaoJvmTest` is eight cases against a real 2-of-3 quorum rather than a stub signature, because a fake one would satisfy every column assertion and prove nothing -- it covers the round trip, the path walking back to the row's own author, the merge in both directions, batch ordering, and a row edited after the fact no longer verifying. `SignedGroupKeyStateTest` adds the end-to-end claim over two devices: a batch of three signed in one session lands as three events on both, each at `m/9420/0/0` that neither device was told and both derived from the room they stand in. `ArchiveApplyJvmTest` asserts the receiver ends up holding the events and not only the rows, and that the four forgeries in its adversarial page become no signed-event rows either -- a forgery filed there is one the recipient goes on to hand to everybody else. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 15:43:18 +02:00
Kgothatso Ngako	47aa79ebc7	feat: archive the translated text too, now that the group signs it ... The merge brought in two commits that close the gap this feature was written around, so the allowlist grows from six kinds to eight. `feat: sign an artifact's first version with it, not derive it after` makes the version the second item of the artifact's own signing batch. `feat: ask the group to sign a chunk's translation, not just save it` puts a quorum behind the prose. Both were done for their own reasons and neither was about the archive, but they are exactly what the archive was missing: an archive can only carry what its recipient can check, so a derived version and a member-authored translation could not travel. A new member got the whole structure and none of the words. **30301 and 30309 do not go on the end of the list.** The order is the foreign keys: a version sits between its artifact and the chapters hanging off it, and a translated chunk hangs off both a source chunk and a translation chapter, so that one really is last. **`toArtifactVersionEvent` had the bug this predicted it would.** It emitted [artifactId, alt] where `build` emits [alt, artifactId], so the id did not round-trip -- the same fault fixed on `MantraArtifact.toArtifactEvent` in Phase 3, in the second of the three unused rebuilds, and for the same reason: nothing had ever called it, so the "tag order matches build" claim in its comment was never checked. `toTranslationChunkEvent` was already correct. Both now have a round-trip case, which is what makes the difference between a rebuild that is right and one that has not been contradicted yet. **`signedEventsOf` walks two steps further**, emitting each version and the translation chunks under each translation chapter. A retranslated passage archives once: the arm that applies a translation chunk drops the one it supersedes -- newest by the timestamp the group signed at, id breaking a tie -- so what a sender holds, and therefore what travels, is the group's current answer to each passage rather than its drafts. **The seeds had to change with it.** Both database tests derived the artifact's first version by applying the artifact, which is exactly what stopped happening; they now sign it through `ArtifactVersionEvent.initialVersionOf`, the way the batch does. That also removes the one exception in the end-to-end assertion: every archived row is now authored by the room and carries a signature, where the artifact version used to have to be excused for having neither. 480 tests pass. The plan's Phase 3 table, its built-vs-plan table and its "what this does not do" section are updated -- what an archive cannot do is down from two things to one, and the remaining one is that it still cannot make its recipient able to sign. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 15:13:43 +02:00
Kgothatso Ngako	f3984e838c	test: prove the catch-up row by row, and say what an old build makes of a page ... Phases 8 and 9 of docs/member-archive.md. The tests ran in the phases where the code they cover first existed -- the way the batch-signing note's did -- so this is what was missing from them, plus the rollout note, plus the plan marked built. **Compared row by row, not by count.** The end-to-end test asserted the two databases held the same *number* of artifacts, chapters and chunks. That is not the claim: two databases can hold the same counts and disagree about every row, and a rebuild that lost the group's signature -- or re-authored a row as whoever sent it -- would pass a count and fail the only thing an archive is for. It now compares `(id, author, signature)` per row across every archived kind, and then asserts each one is authored by the room and carries a signature. The artifact version is the one exception, and it has to be: nobody signs it, it is derived from the signed artifact on arrival. Which is exactly why it is not archived, and why a chapter's foreign key survives without it. **An old build does not ignore an archive page, it renders it.** Phase 9's first draft said an old build "files it as unsupported, exactly as it does today for anything it does not know" -- true, and it reads better than it lives. An unsupported row's content is `event.toJson()` and it renders as an ordinary chat bubble, so every member on an old build sees each archive page as a raw-JSON bubble of up to `MAX_PAGE_BYTES`, once per page. Nothing breaks and nothing is lost, but a group mid-upgrade gets a genuinely unpleasant transcript, and that is worth knowing before the first archive goes out. So the rollout rule is stated rather than implied: the receiving half ships safely on its own -- phases 1-4 send nothing -- and no member starts sending until every member understands kind 30327. The mitigation if that ever proves unacceptable is the one the appendix rejects for other reasons, and it is named there so the trade can be weighed rather than rediscovered. **The plan is marked built**, with a table of the five places the implementation chose differently from the plan and why: nine archivable kinds became six, a count cap that could never fire, queueing moved a phase later, a re-read that was never needed, and the rollout note above. Phase 8 also records the three tests that were not in the first draft, each written because something passed for the wrong reason -- a cap that could not fire, an out-of-order test on an archive that was never out of order, and a sweep whose "still missing" count included failures a later pass had already fixed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 14:37:23 +02:00
Kgothatso Ngako	a315b86918	feat: say in the transcript that a member is being caught up ... Phase 7 of docs/member-archive.md, in part. Three chat types -- `TYPE_ARCHIVE_REQUESTED`, `TYPE_ARCHIVE_SENT`, `TYPE_ARCHIVE_RECEIVED` -- so a room that fills itself in explains itself once. Without this the archive is entirely silent by design: it files no line per applied payload, because `ChatMessage` has an `autoGenerate` primary key and every payload would mint a fresh row on every pass of the sweep. The result was a member joining a working group and watching a room populate with no account of where any of it came from, which is worse than the noise it avoided. **One line per archive, not per page.** The received line is written when the request stamp is cleared, which is as close as this can get: an archive's pages are not distinguishable from each other at apply time, and clearing the stamp is exactly the moment a catch-up stops being pending. There is a test that delivers a payload per page, backwards, so the sweep runs repeatedly over many pages, and asserts the transcript holds two lines. **A push behind a Welcome writes nothing**, because the room was never asked. It lands before the member has opened the room, and "caught up on work you have not seen yet" is a line about nothing. Also tested. **The received line names no sender.** An archive can be assembled from pages sent by more than one member, so attributing the catch-up to one would be a guess dressed as a fact. The sent line does name its recipient, written into the content the way the invite line writes one -- which does not follow a rename, and is the accepted cost for a line about something that happened once. **Content is whole sentences**, so these stay out of the AUTHORED sets and nothing prefixes a name to them. And they are added to `ARCHIVE_TYPES` with a matching arm in the transcript, because the failure mode for a missed set is silent: the line renders as a chat bubble, looking exactly like a member having said "Caught up on 12 items". Icons per type rather than the `PanTool` fallback. **Two items from this phase are deliberately not done**, rather than written without the app in front of me: the banner saying a room is catching up, and a "Send history" action on the member row. The first is UI state plumbed through a view model into a layout and the transcript line covers the same ground; the second is a convenience, since both real paths are already automatic. Both are written up in the plan as outstanding, along with the thing this phase was also meant to say and does not: that an archive does not make its recipient able to sign, and does not carry the translated text. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 14:34:19 +02:00
Kgothatso Ngako	6d8866c2b0	feat: offer a new member the group's work alongside their welcome ... Phase 6 of docs/member-archive.md, and deliberately the phase after the one that makes it unnecessary. `deliveryWelcome` now queues an archive for the member being invited, so in the ordinary case they have the group's signed work before they think to ask for it. **This is a latency optimisation, not the mechanism.** A page queued behind the Welcome is not delivered after it: they are different transports -- a relay-borne gift wrap and a kind:445 -- with no ordering between them, and a page that overtakes the Welcome is from an epoch ahead of the invitee's, so `MarmotInboundManager` drops it outright rather than deferring it. Nothing retries and the inviter sees a success. That is the failure in docs/marmot-membership.md wearing new clothes, and the only thing that closes it is the invitee asking once they are demonstrably in the group, which Phase 5 already does on their first open of the room. So nothing here reports failure to the inviter. A push that does not land is the ordinary case the pull exists for, and it sits inside `deliveryWelcome`'s own catch alongside the Welcome it rides behind. A room with nothing signed queues nothing and still invites. **One call, two occasions.** `ArchiveManager.answer` becomes `sendTo`: answering a request and pushing behind a Welcome are the same operation and differ only in who decided, so it is named for what it does rather than for either occasion. Also corrects the plan. Phase 6 claimed the room had to be re-read between the invite and the assembly, for the same reason sequential invites re-read it. It does not -- that rule is about the MLS snapshot a commit is built on, and this runs downstream of the commit over `Mantra*` rows, which no commit touches. Two tests against a real `deliveryWelcome`: inviting into a room with signed work queues exactly one archive page addressed to the invitee, and inviting into a room with none queues no page while still writing the Welcome's gift wrap. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 14:29:11 +02:00
Kgothatso Ngako	ed4410b972	feat: apply an archive a member is sent, and sweep what arrived too early ... Phase 4 of docs/member-archive.md, and the half where the security lives. A member who holds no share, took part in no signing session and cannot decrypt a word of the room's history now ends up with the same rows as everybody else -- and gets there without trusting whoever sent them. **Intercepted in `fromGroupEventResult`, not in `applyInnerEvent`.** An archive is neither a document nor a submission, and deciding whether to act on one needs the active key, which `applyInnerEvent` has no business knowing. That is the same reason the gift wrap above it is handled there, so it sits next to it. **Verification per payload, framing per page.** A forged payload costs itself and nothing else -- the rule `MarmotInboundManager` already uses for a forged direct message, and for the same reason: this runs inside the inbound transaction and one bad event must not take the room down with it. Refusing the whole page would also let a single forgery deny an entire archive. The page's own framing stays all-or-nothing, because a page that will not parse has lost the thing that says what it contains. **The allowlist runs before the signature check, and it is not a formality.** Verification admits an event to the apply path on the strength of the group's signature, which makes every kind the group has ever signed replayable by any member at any time. There is a test that puts a genuine, still-verifying `GroupKeyStateEvent` in a hand-rolled page -- `ArchiveEvent.build` refuses to make one, which is the outbound half of the same rule -- and asserts the receiver's key state does not move. **The chat line is dropped, deliberately.** `ChatMessage` has an `autoGenerate` primary key, so there is no id to dedupe on and every applied payload would mint a new row: a synthetic transcript dated now, and another one on every pass of the sweep. The archive restores the work. The conversation is forward secret and stays gone. **A device that is not the named recipient does nothing.** It can read the page -- it is an ordinary group message, and it is the group's own history -- but it already holds the work, and re-applying would rewrite every one of its rows to point at an archive page rather than at the event that introduced it. That is also what bounds the sweep: only the member being caught up ever builds the list. **The sweep needs no table.** Pages arrive over relays in no order, so page 3 can land before page 2 and its chunks have no chapter to hang off. Those throw a foreign key violation and would be lost -- except the inbound path already stores every inner event it decrypts, so re-reading them is the same shape `FrostSigningManager.replayStoredMessages` has, for the same reason: nothing was lost, it just had nowhere to go at the time. Two things about the loop, the second found by a test: Progress is measured by *failures falling*, not by rows written. "Repeat while a pass applied something" does not terminate, because every write is an upsert and succeeds forever. What strictly decreases is the count that threw. And the result is the last pass rather than the sum of them. Accumulating counts a payload once per pass it survived and reports failures a later pass went on to fix, so `failed > 0` stops meaning "still missing" -- which is the only question a caller asks it. Caught by strengthening the out-of-order test to assert that the page completing an archive leaves nothing behind, rather than only that the rows matched: without that, the test passed while reporting fourteen failures on a fully converged database. Seven tests over two real databases with the pages carried by hand. The one that matters puts four forgeries in a page beside one honest dialect -- the room's id as author with a made-up signature, a real quorum of another group, an event edited after signing, and a member's own rumor, which is what everything on the wire looks like today -- and asserts the receiver ends with exactly the honest one. The rest: a full catch-up matches the sender row for row with the group's signature intact, pages delivered backwards converge and are asserted to have really failed first so the test cannot pass for the wrong reason, an archive files no chat lines, a bystander applies none of it, and applying the same archive twice changes nothing. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 14:19:15 +02:00
Kgothatso Ngako	acff66a22e	feat: rebuild the group's signed record out of the rows it left behind ... Phase 3 of docs/member-archive.md. `ArchiveManager.assemble` walks a room's rows, rebuilds each into the event the group signed, drops anything it cannot prove, and cuts the rest into pages. Nothing sends one yet. **The gate found a real bug, which is why it was the gate.** Signed events are not stored as events -- `FrostSigningManager.complete` applies one and what survives is a `Mantra*` row -- so an archive has to rebuild them with `toXEvent()` and stands or falls on that being byte-identical to what was signed. Every `toXEvent()` in the codebase turned out to be unused in production, written for exactly this and never called, so the "tag order matches build so the event id round-trips" comments on them were claims nothing had ever checked. One was wrong. `MantraArtifact.toArtifactEvent` put the alt tag last where `ArtifactEvent.build` puts it first, and left out the version metadata tag altogether -- because that tag is not on the artifact row at all. `fromArtifactEvent` reads the artifact's own fields and drops the version label, which `applyInnerEvent` has by then turned into the artifact's first `MantraArtifactVersion`. So the label is now a parameter, read off the initial version: the one whose `createdAt` is the artifact's, since `initialVersionOf` derives it from the same event. Neither fault would have surfaced as an error. Both produce a well-formed artifact whose id no longer matches its fields, which every receiver drops as a forgery, silently, one kind at a time. `ArchiveRoundTripTest` now signs each archivable kind with a real quorum, files it as a row, rebuilds it and asserts the signature still covers what comes out -- plus the negative case, that rebuilding with the wrong version label fails as a forgery rather than as a mistake, which is why the assembler reads the label rather than defaulting it. **The allowlist narrows from nine kinds to six, and this is the finding to read.** Only six of the thirteen nip30303 kinds ever reach a signing session; the rest travel as member rumors, vouched for by the MLS frame they arrived in and by nothing that survives leaving it. An artifact version is derived rather than signed -- which is fine, because applying the archived artifact derives it again and the chapters hanging off it keep their foreign key. Nothing builds a `TranslationEvent` at all. The contributor lists have no arm in `applyInnerEvent` that writes a row. And `TranslationChunkEvent` -- **the translated text itself** -- is submitted by `MantraDao.saveTranslation` as its author's rumor, because a translation is one member's work rather than a group decision. So an archive restores everything a translation hangs on and not the translation: a new member gets the dialects, the artifacts, the chapters, the source chunks, which translations exist and their chapter scaffolding, and none of the prose. That is a real limit rather than a detail, so it is written into the allowlist's own doc comment, into the plan's "what this does not do", and into a test named after it -- with the three ways out sketched and none of them taken here, because the cheapest gives up the property the rest of this rests on and the best is a product decision about whether translating is an act of the group or of a member. **Nothing unverifiable leaves.** Every rebuilt event is checked with `isSignedByRoom` against the same room id the recipient will use. Not politeness -- the receiver checks anyway -- but so the page count says what will actually arrive: a row from a member's rumor is dropped here rather than by the recipient. **Walked down the tree, not queried per kind.** Only dialects and artifacts have a by-room query and the rest hang off a parent, and the walk is also what puts an artifact's version label within reach. Order is settled afterwards by `inApplyOrder` rather than by the walk, since the walk groups by artifact and the foreign keys are by kind. **Paging is greedy against both caps**, because they bind different archives: a room of one-line dialects hits the count first and a room of chapters hits the bytes. An event too large for a page of its own is dropped with a log rather than failing the archive -- a chapter nobody can archive is a hole, a member who gets nothing is a bigger one. Assembling only; queueing moved to Phase 5, where the thing that decides when to send lives. That keeps this testable against a real database with no outbound path in the way. Seven tests over a real in-memory database seeded through `applyInnerEvent` itself, so what is archived is what a member's device really holds rather than rows built to suit the test: every payload verifies, all six kinds appear exactly as often as they were signed, the whole archive is in dependency order end to end, a member's unsigned dialect sitting in the same room is left out, an empty room archives nothing without failing, and two archives of identical rows do not share an id -- which is what stops two members answering one request from having their pages counted towards each other's total. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 14:12:29 +02:00
Kgothatso Ngako	6e04f6c7af	feat: give the group's signed record an envelope it can travel in ... Phase 2 of docs/member-archive.md. Two kinds, three tags, a codec and two caps. Nothing sends or applies one yet -- that is phases 3 and 4 -- so this changes no behaviour at all. `ArchiveEvent` (30327) carries a page of the group's signed events, each whole, keeping its own id, author and signature so the receiver checks it rather than believing it. `ArchiveRequestEvent` (30328) is how a device with none of it asks. **Why not one `SubmissionEvent` per event.** The envelope fits and the meaning does not. A submission is an *act* -- this member is putting this event in front of this group -- and an archive asserts nothing; it re-delivers what the group already agreed. On one kind a four-hundred-event backfill is indistinguishable from four hundred new submissions and every device has to guess which it is reading. It would also be one inner event and one kind:445 per payload where a page is one, and the submission arm of `applyInnerEvent` files a chat line per payload, which an archive must not. **Why 3032x and not 30313.** 30313 is free beside the nip30303 document kinds and is not used, on `FrostSigningEvents`' own advice: the DKG's 30310-30316 already overlap that range and are told apart only by living in NIP-17 gift wraps instead, which it calls "an accident of routing rather than a decision, and the next family added should not rely on it." This is that next family. 30327 is also the right neighbourhood on the merits, next to `GroupKeyStateEvent` at 30326 -- an archive is a statement about the record rather than a document kind. **One list is the apply order and the allowlist both**, because a separate allowlist is one more thing that can disagree with the order it is applied in. The order is Room's rather than nostr's: every archivable kind has a foreign key on the one before it, and kind order is not dependency order -- a dialect (30304) has to land before an artifact (30300), and a translation chapter (30308) hangs off a translation artifact version (30306) which hangs off an artifact version (30301). So it is a list, not a `sortedBy { kind }`, and there is a test that fails if anybody makes it one. It is an allowlist first. Verification admits an event to the apply path on the strength of the group's signature, which makes every kind the group has ever signed replayable by any member at any time. A `GroupKeyStateEvent` is group-signed and passes verification perfectly, so an archive carrying an old one is a validly signed statement about what the room signs with, replayed by whoever kept a copy. Nothing but this list stops it. The contributor-list kinds (30305, 30307, 30310) are left out on the same principle from the other side: `applyInnerEvent` has no arm that writes a row for any of them, so archiving them would cost bytes and restore nothing. **All-or-nothing parsing, per-payload verification.** These are not in tension; they answer different questions. A page that will not parse has lost its framing, and one silently shortened by an element would report a complete archive on its page count while holding less than it says. A payload whose signature does not verify is a well-framed page with one bad event in it, and costing its honest neighbours would let a single forgery deny an entire archive. **The count cap was 256 and 256 can never fire.** An event carries 64 characters of id, 64 of pubkey and 128 of signature before it says anything, so the floor is about 370 bytes and a 64 KB page cannot hold much past 170 of them -- the byte cap always binds first and the count cap is a check that never runs. Found by writing the test that a page at exactly the cap still decodes, which failed. Now 128, where both bind something: the count stops a page of many small payloads, the bytes stop a page of few large ones. That test is what fails if somebody later raises one number without the other, and the doc comment says they have to move together. **The `p` tag is a hint, not access control**, and `ArchiveRecipientTag` says so where it is defined. The page is an ordinary group message and every member can read it, which is right, because it is their own history going back to them. What it decides is who *acts*: a device that is not named applies nothing, since it already holds the work and re-applying would rewrite every one of its rows to point at an archive page rather than at the event that introduced it. `ArchivePageTag` refuses an index outside its own count rather than clamping it. The pair is how a receiver decides it has everything, so a repaired one would let a truncated archive read as complete. Twenty-one tests over the codec, both caps, the allowlist, the order and the tags. Also corrects the phase-2 section of the plan, which still said 256. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 14:00:30 +02:00
Kgothatso Ngako	3ee1676a04	docs: plan handing a new member the group's signed history ... A member added after the work was done sees none of it, and nothing in the app will ever show it to them. Two independent reasons, and the second is the one that surprises people. MLS gives no history: a Welcome carries the ratchet tree at the current epoch, not the transcript, and `MarmotInboundManager` drops anything from an epoch it holds no keys for. That is forward secrecy working rather than a gap to close. But group-signed events never travel at all. `FrostSigningManager.complete` says so in as many words -- a signed event authored by the threshold key cannot go out as an inner event, because the outbound pipeline would re-author it as its sender and strip the group's signature off -- so every device *derives* the finished event from its own `FrostSigningItem` rows. A member who was not in the session has no items, and no later message carries the event. So the second problem does not follow from the first and is not fixed by fixing it: even a member who could decrypt the whole back-transcript would still hold nothing an artifact, chapter or chunk could be built from. Which makes an archive not a convenience but the only path, and fixes the line the design has to hold: **it carries what the group signed, never the chat.** Restoring the chat would undo forward secrecy on purpose, and a signed event is the only thing a new member can check for themselves. **The property the whole plan rests on is already true.** A room's id *is* the group's threshold key derived at the room's path -- `GroupKeyState.verifies` and `FrostSigningManager.signingPath` hold that invariant from their own ends -- so `isSignedByGroup`'s three checks collapse to `event.pubKey == chatRoomId`, an id check and a signature verify. No key state row, no threshold key, no path, no lookup. A member who can name the room can verify its signatures, which is exactly the position a new member is in, and it means the sender of an archive does not have to be trusted at all. **Two guards the plan makes non-negotiable.** Nothing on the inbound nip30303 path verifies a signature today, and that is currently correct: rumors carry an empty sig and are authenticated by the MLS frame, so nothing on the wire has ever claimed group authorship. An archive is the first thing that does, so the verify is the feature's entire security rather than hardening on top of it. And verification turns "group-signed" into an admission ticket for the apply path, which is a wider door than it looks: a `GroupKeyStateEvent` is group-signed and would pass perfectly, so an archive could replay a genuine old one and re-point what the room signs with. The archive therefore carries an allowlist of document kinds, checked outbound and independently inbound -- the same shape, and the same reasoning, as the cap on `k` in frost-batch-signing.md. **Push and pull, in that order of appearance and the reverse order of importance.** Pushing an archive after the Welcome is what the question asked for, and on its own it fails the way marmot-membership.md describes: it is an application message in the epoch the add created, so one that beats the Welcome there is dropped rather than deferred, silently, while the inviter sees a success. So the joiner asks instead -- a request is proof it has processed its Welcome, and it covers the reinstall and the second device, which no invite-time push can. The push stays as a latency optimisation, deliberately phased after the thing that makes it safe. Nine phases: the verifier, the events, assembling an archive, applying one and the sweep that lets pages arrive out of order, the request, the push, UI, the cross-device tests, and rollout. The sweep needs no new table -- the inbound path already stores every inner event it decrypts, so it is the shape `FrostSigningManager.replayStoredMessages` already has. Also written down, because it is the first thing this will be reported as a bug for: an archive lets a new member *read* everything and does not let them sign anything. `proposeSigningBatch` wants a secret share and a place in the ceremony, and a group that re-runs its ceremony derives a different room rather than re-keying this one. Closing that needs share resharing, which is a great deal more work than this and is the thing to build after it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 13:52:33 +02:00
Kgothatso Ngako	dbdff55ee0	feat: open the proposal a transcript line is about, not the room's latest ... Tapping a signing line in the transcript opened whichever session the room was running, resolved by `liveSessionForChatRoom` -- the newest one not yet finished, or failing that the newest one at all. That is a guess, and it was a good one exactly as long as a room had one proposal to guess at. With a chapter and its translation open together, half the lines in the transcript led to the other proposal. The line now says which session it belongs to, so there is nothing left to guess: it carries `frostSigningSessionId` through to the route. A line written before that column opens the room's proposal list instead, which is the honest answer to a line that cannot say what it meant -- every proposal with its own state, and the reader picks -- rather than a guess dressed as an answer. That empties `FrostSigningRoute.sessionId` of its reason to be optional, so it is required, and `liveSessionForChatRoom` goes with it from the interface, the implementation and the no-op. `FrostSigningViewModel` loses its resolution step and the "This group is not signing anything right now" error underneath it -- which was never the right thing to say to somebody who had just tapped a line about a specific session. The transcript keeps doing the one job it is good at: showing a proposal as it happens, and saying whether it is still asking something of you. What it stops doing is standing in for a list of them. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 12:43:00 +02:00
Kgothatso Ngako	4de87edf12	fix: tell one proposal's transcript lines from another's ... A room with two proposals open showed "Review" on both, and then dropped it from both the moment either one was decided. The second proposal was still waiting on the reader, still had a decision in it, and had nowhere left to be reached from. Two proposals at once is not a corner case any more: a chapter and the translation scaffolding beside it are proposed as separate sessions, on purpose, and they run at the same time. Both write the same line types into the same stretch of transcript. `answeredRequests` matched a request against any later line of the fulfilling type, and `settledRequests` against any later ending. That reads a room signing one thing at a time exactly right -- the nonce after the request is the answer to it, because there is nothing else it could be an answer to -- and a room signing two things at once exactly wrong. Nothing else on the row could separate them: same type, same room, same minute, and `ChatMessage` carried no session. So the session goes on the row. `ChatMessage.frostSigningSessionId` is nullable, added as schema v11 through `AutoMigration(10, 11)`, and stamped by `FrostSigningManager.announce` -- the one place every FROST line is written, so there is no line that can be forgotten. Both rules read it when both rows have one and fall back to the clock when either does not. The fallback is not a compromise, it is the right reading of the rows it applies to. A line written before this column has no session and never will, and the rooms that wrote those lines could not run two sessions at once, so the clock is the whole truth there. A ceremony line falls back too and always will: a room runs one ritual at a time, and a DKG step is either taken or still waited on. **This reverses a call `FrostSigningRoute` argued for.** Its note said a chat row carrying a session id was "a poor trade for a lookup the screen can do". That was right when the lookup could only be wrong about which of one session it meant. The batch work made two sessions ordinary, and the lookup and the rules both became guesses at the same moment. A column on the table every message uses is the cost; two proposals, one of them unreachable, was the alternative. **Tests.** Three in TranscriptRequestStateTest for what the column buys: a nonce answers its own session's request and not the other's, one session completing settles nothing in the other, and a line naming no session is still read by the clock. TranslationBatchProposalJvmTest proves the other half against a real two-session proposal -- every FROST line the manager writes names its own session, and neither session's lines are attributed to the other. The rule is tested on rows and the stamping is tested on a database, because a rule that is right about rows nothing writes correctly is worth nothing. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 12:41:01 +02:00
Kgothatso Ngako	2e133dd337	feat: sign a chapter and every chunk of it in one session ... A chapter proposal now carries the chapter and a chunk per paragraph, and the group signs the lot at once. Every row a member ends up with is signed: a translation is of a chunk, and a chunk that carries the group's signature over its own words can be checked by anybody holding it, rather than only by re-deriving it from the chapter it came out of. This replaces the derivation two commits ago, which split the chunks out of the signed chapter's text on each device and left them as rumors. That was the right shape when a chunk could only have its own signature by having its own quorum. Batch signing removed that, and this is the other side of the trade `MantraChunk.chunksOf` was weighed against. **A batch whose items name each other.** A chunk carries its chapter's id, and that id is a hash over the group's key at the room's derivation path -- neither resolved until the proposal runs. A caller computing it would be recomputing `signingPath`, the one input in this protocol that must never come from a proposer, since the path decides which key the group signs as. So `proposeSigningBatch` gains a second form: a `lead` template, and a `dependents` builder handed the lead *after* it is authored, returning the events that reference it. Every id still comes out of `unsignedEventOf`, which makes an item naming a chapter nobody signed something that cannot be built rather than something to be tested for. `AddChapterViewModel` passes `ChunkEvent::splitOf` and nothing else. The lead is item 0. Items apply in `itemIndex` order and a chunk row whose chapter does not exist yet is a foreign key violation, so what is referenced is signed first as well as named first. **The cost, in front of whoever is typing.** `MAX_BATCH_SIZE` is 64 and the chapter takes one place, so a chapter is capped at 63 paragraphs and a longer one has to be split in two. That is a real limit on real prose. The form counts chunks against the cap as the text is typed, colours the count when it is past, says what to do about it, and will not propose -- because the alternative is an IllegalArgumentException after the fact. The manager still refuses independently; the screen is not what enforces it. **What went away.** `MantraChunk.chunksOf` and the derivation it did inside `ChatMessage.applyInnerEvent`. Chunks arrive as their own signed events now and go through the `ChunkEvent.KIND` branch that was always there. `ChapterEvent` still carries the whole text beside chunks that hold the same words: chunk boundaries are a decision about how to divide the work, and a chapter that kept only the pieces could never be divided differently again. **Tests.** `ChapterChunkSplitTest` covers the split as a pure function -- what each chunk names, counts and carries. `SignedChapterTest` signs a real batch, one FROST instance per item, and checks every chunk row is authored by the room and carries a signature over its own id. `ChapterBatchProposalJvmTest` runs the real proposal against a real database, which is where the sharp edge is: item order, the chunks naming the chapter as the group will author it, and both ends of the cap -- 63 paragraphs proposes, 64 is refused and leaves no session behind. Checked against broken implementations: putting the lead last, naming the wrong chapter, and stamping the chunks off the clock are each caught, in both suites. `jvmTest` runs on linux again as of the merge, which is what made the database-backed test possible. Dropped a nonce-reuse test that was in the first draft of this: it asserted over its own fixture, and `FrostSigningRoundTest` and `SignedGroupKeyStateTest` already hold the manager to giving every item its own nonce. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 10:34:12 +02:00
Kgothatso Ngako	e081d14f37	refactor: weigh the chapter's chunks against batch signing, and keep deriving ... Batch signing landed on mantra while this branch was open, and it makes the argument this change was built on obsolete as written. MantraChunk.chunksOf said the chunks "cannot be events proposed on their own -- that would cost a quorum per paragraph". They can now: proposeSigningBatch would carry the chapter and a chunk per paragraph through one quorum, and every row would hold a signature of its own. Weighed and declined, and the KDoc now says so rather than leaning on a reason that stopped being true. MAX_BATCH_SIZE is 64, which caps a batched chapter at 63 paragraphs and fails an ordinary one outright; the text would go on the wire twice, whole on the chapter and again split across the chunks, and the proposal is the term that cap is sized against; and all-or-nothing over k items would make a long chapter less likely to be signed than a short one, for no reason a member could see. The signature it would buy is redundant besides -- the appendix rejects the manifest shape because an item then needs a lookup to be checked, and here that lookup is a foreign key: a chunk is a pure function of its chapter and cannot be stored without it. docs/frost-batch-signing.md records this under the slot Phase 7 leaves open -- "deciding *what* to batch" -- because the next caller will reach for the same shape. The rule it leaves behind: batch siblings, not derivations. Events that could each have been authored separately are worth a batch; events that are a function of another event in the same batch are worth deriving instead. **The merge.** Only SignedChapterTest broke: the five per-item columns moved off FrostSigningSession onto FrostSigningItem, so it builds an item and calls signedEvent(item, sig), which is how SignedArtifactTest was ported in the same commit. Nothing in the flow itself moved -- proposeSigning kept its signature as the one-event form, and complete() applies each signed event through ChatMessage.applyInnerEvent, so the chapter's chunk derivation works the same whether the chapter arrives alone or as one item of somebody else's batch. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 10:13:09 +02:00
Kgothatso Ngako	1448ed5ad8	docs(frost): record batch signing as built, and what rollout needs ... Phase 7 of docs/frost-batch-signing.md, which is the phase with no code in it. Nothing needs a feature flag. k=1 is the entire behaviour of the app as shipped -- no caller batches anything yet -- and at k=1 every message is byte-identical to the app before Phase 1: encodeProposal returns the bare event object, joinPayload of one value is that value, and every plural branch in the transcript is only taken above one. The doc now tabulates that rather than asserting it in prose, since it is the claim the whole rollout rests on. The one rollout constraint stands: before a caller batches, the group has to be on a build that understands array proposals. There is no negotiation for it and adding one is not worth it -- an old device refuses an array proposal outright, so the failure mode is a batch that never reaches threshold and is abandoned, visible in the transcript and costing a retry. Also records what is left, which is nothing in the protocol: deciding what to batch is a product question, bounded only by "a batch is only as available as its worst item" and "GroupKeyStateManager.propose must never batch". The phases are kept as written rather than rewritten into a description of the result -- the code reads better against the argument it came from -- with the two places the implementation chose differently (itemIndex over index, DROP COLUMN over a table rebuild) marked in their own sections. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 05:00:21 +02:00
Kgothatso Ngako	2309879153	test(frost): cover the batch's failure modes and its crypto without a database ... Phase 6 of docs/frost-batch-signing.md. 361 jvmTest and 227 testDebugUnitTest pass. ## Inbound path (SignedGroupKeyStateTest) Both drive the manager with a hand-built inner event rather than one the other device queued, which is the only way to be a faulty or dishonest member in this harness. - A one-value nonce offered for a three-item batch does not count towards the threshold: the coordinator never reaches a signer set. The length check is all that stands between a batch and a signer whose contribution lines up against the wrong messages, so truncating or padding would produce partial signatures aggregated against events nobody agreed to. The test then pumps the real nonce and the batch completes -- it is a stall, not damage, which is FrostSignerMessage's composite key doing its job. - A second proposal under the session's own id changes neither its event ids nor its seeds. Every seed is already committed to its item's message; a different batch under the same id would have those seeds produce a second partial signature over a second message, which is how a share is extracted. ## Real FROST, no database (FrostSigningRoundTest) - A k=3 batch from one signer set, all three verifying against the room's key -- the manager's shape with the database taken out of the way. - Item 0's signature does not verify against item 1. Signing three events in lockstep must not make any of them interchangeable. - Both halves of the no-shared-nonce property, because either alone is enough to be relied on by accident: SecretNonce.generate mixes the message in, so one seed under two messages already gives two nonces -- and the manager mints distinct seeds regardless. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 04:58:48 +02:00
Kgothatso Ngako	59c34263b3	feat(frost): let one signing session carry a batch of events ... Phase 3 of docs/frost-batch-signing.md. A session can now be proposed over several events, and the whole batch is signed in one round of four group events with one approval. 356 jvmTest and 224 testDebugUnitTest pass. ## The wire, and the compatibility rule that shapes it FrostSigningEvents.encodeProposal serialises a batch of one as the bare event object it always was, and only a genuine batch as a JSON array. That is not tidiness. A build predating this reads an array with Event.fromJsonOrNull, gets null, and drops the proposal -- so an old device refuses a batch outright rather than signing part of one, while single signing keeps working right through a mixed-version rollout. Emitting an array unconditionally would break every one-event session for those devices and buy nothing. decodeProposal accepts both forms permanently: proposals in the old shape do not stop arriving because this build stopped writing them. It is all-or-nothing -- an array with one unreadable element is refused rather than silently shortened, because the batch's length is what every later payload is checked against, and a proposal that quietly lost an event would have every signer's contribution rejected for being the wrong size: a stall with nothing to blame. ## MAX_BATCH_SIZE, checked twice 64, enforced in proposeSigningBatch and again, independently, in acceptProposal. The second check is the one that matters. A proposal is the only place in this protocol where a remote party decides how much work everyone else does -- k native key generations, k signatures, and a group event carrying k payloads, from a single message -- and until batching that was bounded only by never being more than one. ## acceptProposal over a list Each element is rebuilt from its own fields under this device's own reading of the room's path and checked against the id it claims, exactly as before but per item, and the whole proposal is dropped if any one fails. The write-once rule widens from "the event this session signs" to "the ordered list of events this session signs": a second proposal under the same id whose list differs anywhere is logged and ignored. ## The API FrostSigningManager.proposeSigningBatch(events: List<EventTemplate<*>>) is public here rather than in Phase 4, because without it there is no way to produce a k>1 session and everything above would ship untested. proposeSigning keeps its signature as the one-event form, so no caller moves. Each template carries its own createdAt. ## Tests - FrostProposalCodecTest (new, commonTest): a batch of one is byte-for-byte the old JSON object -- the assertion that stands in for the old build nobody can run here -- plus order preservation, old-form decoding, and refusal of empty, malformed and partly-unreadable arrays. - SignedGroupKeyStateTest: a k=3 batch between two devices over two databases. Three signatures verifying against the room, three dialects applied on both devices in order, five messages from the coordinator and two from the other signer, and one approval line rather than three. - The negative test that matters: no two items of a batch share an aggregated nonce or a seed, and the two devices' seeds do not intersect. Every positive test still passes if two items share a nonce -- the signatures verify fine; what sharing costs is the secret share. - The cap is refused when proposed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 04:48:24 +02:00
Kgothatso Ngako	935a8fe37a	refactor(frost): run a signing session as k FROST instances in lockstep ... Phase 2 of docs/frost-batch-signing.md. Pure refactor: proposals still carry one event, the wire is byte-identical, and every test passes unchanged -- 344 jvmTest and 217 testDebugUnitTest, none of them edited in this commit. advance() now loops over FrostSigningItem rows rather than reading the first one. One nonce per item, one aggregate per item, one Session.create per item, one partial signature per item, one signature per item. The signer set, the public shares, the tweak cache and the approval stay shared, because they are the terms that do not enter e = H(R‖P‖m). The coordinator's aggregation is the place where that distinction bites: it builds one AggregatedNonce per item, each from that item's nonce from each chosen signer. Reusing one across two items would be reusing R across two messages. ## The payload codec, early joinPayload/splitPayload land here rather than with the wire change, because at a batch of one a comma join is the identity -- the payload is the bare value it has always been. That leaves Phase 3 to the proposal encoding alone. splitPayload is strict: a payload that is not exactly the batch's length is dropped rather than truncated or padded. It runs in orderedNonces, orderedPartialSignatures and splitForSession -- never in record(), which stores payloads without parsing them so that a nonce can arrive before the proposal that would give it a length to check against. ## Two short-circuits, and one trap in the first advance() runs on every arriving message, so at a batch of k it was k native key generations, k Session.creates and k signs each time, usually to discover there was nothing left to do. - Nonces are generated by `lazy`. The obvious version -- a guard computing `ownNonce == null || (isSigner() && ownPartial == null)` -- is wrong, and wrong in a way that reads fine and fails every signing test: the coordinator settles the signer set further down the same pass, so isSigner() at the top is false on exactly the pass where the coordinator goes on to sign, and the nonces are never generated. Reproduced as IndexOutOfBounds before switching to lazy, which has no prediction to make. - A device that is neither signing nor aggregating leaves before building any FROST session, rather than building k of them to do nothing with. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 04:40:22 +02:00
Kgothatso Ngako	dff41d417d	feat(frost): move a signing session's per-event columns onto FrostSigningItem ... Phase 1 of docs/frost-batch-signing.md, which is added here as the plan the next phases follow. Schema only: a session still signs exactly one event, the wire is byte-identical, and every existing test passes on the moved columns. ## What moved, and why it had to A batch of k events is k independent FROST instances sharing a signer set, not one signature over k messages. That is forced rather than chosen: a Schnorr partial signature is `s = k + e·x` with `e = H(R‖P‖m)`, so two messages under one nonce R give two equations in one unknown and the secret share falls out. So the five columns that enter that equation -- unsignedEventJson, eventId, nonceRandom, aggregatedNonce, signature -- move to a child table keyed (sessionId, itemIndex). What stays on FrostSigningSession is everything outside it: the ceremony, the threshold, the derivation path, the signer set, and the one approval. itemIndex is protocol rather than presentation -- nonces and partial signatures are joined positionally against it -- so getItems() orders by it and nothing re-sorts. Spelled itemIndex rather than index to keep hand-written queries free of backticks. No itemCount column. The count is a COUNT(*), for the same reason signerIds is derived from the ceremony's participant order rather than stored: a denormalised count is one more thing that can disagree with the rows. ## Migration 9 -> 10 Manual, not auto: Room can create the table and drop the columns but cannot copy between them, and the copy is the whole point. A session in flight at upgrade holds its nonce seed and the aggregate it is already signing against, and neither can be regenerated -- losing either makes the next pass derive a different nonce for the same message and publish a second partial signature over it, which is the extraction case. Both are copied verbatim into item 0, so an in-flight session resumes as though nothing happened. Removing the columns uses ALTER TABLE DROP COLUMN rather than the usual create-copy-drop-rename rebuild. FrostSignerMessage and FrostSigningItem both reference FrostSigningSession(id) ON DELETE CASCADE, and DROP TABLE fires cascades -- with foreign keys enforced the rebuild would delete every signer message and every item just written. Whether it does depends on Room disabling foreign keys around migrations, which is not worth depending on when DROP COLUMN cannot go wrong. It needs SQLite 3.35 and unindexed, unconstrained columns; these five qualify, and getRoomDatabase pins BundledSQLiteDriver on every platform. ## Invariants established here for the phases that follow - signerIds and every item's aggregatedNonce are one write-once unit, applied by applyAggregate() -- items first in one transaction, then the session, so "some items aggregated" is unreachable and signerIds != null stays the gate. - Signatures likewise, via applySignatures(); isSigned() counts rows instead of reading a flag. - complete() verifies every signature before applying any event, so a batch is all-or-nothing rather than half-filed. - itemsOver() gives each item its own 32 bytes of seed. Independent seeds mean an off-by-one in index handling produces a session that fails to aggregate rather than one that signs two messages under a single nonce. signedEvent() and isAwaitingApproval() now take the item(s) rather than the session, which propagates to the repository, the view model and the screen. advance() reads items.first() and Phase 2 turns that into a loop. ## Tests - FrostSigningSessionDaoJvmTest: index ordering, single-item read, upsert replacing rather than accumulating, signed-item counting, cascade delete. - FrostSigningItemMigrationJvmTest (new): the backfill against a real v9 database, asserting the seed and aggregate values survive -- not merely that a row appeared -- plus the exact column lists Room will check at open time. - 338 jvmTest and 217 testDebugUnitTest pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 04:33:46 +02:00
Kgothatso Ngako	af81933ab4	Merge branch 'mantra' into claude/room-db-testing-setup-b053cd ... Brings the branch up to date with the 40 commits mantra gained while the jvm target was being built, so that merging the other way is a fast-forward. One conflict, in docs/README.md, where both sides added rows to the index table. Kept both, and gave the jvm-target note a clause in the closing prose since it is the one document there that is not about the protocol. One thing the auto-merge could not have caught. `9250991` added NostrEventDao.getMarmotGroupNostrEventsByChatRoomId as a blocking query, which android accepts and which Room refuses to generate for any other target -- so the merged tree failed :composeApp:compileKotlinJvm with the same "Only suspend functions are allowed in DAOs declared in source sets targeting non-Android platforms" that phase 4 dealt with 58 times. Made suspend; its only caller, NostrDao.reindexMarmotGroupEvents, was already suspend, so again no cascade. That is now a standing cost of this branch rather than a one-off: any DAO method added on mantra while this is outstanding will break the jvm build on merge. It is a one-word fix each time, and the compiler names the line. Verified on the merged tree: :composeApp:compileKotlinJvm and :composeApp:compileDebugKotlinAndroid green, :composeApp:testDebugUnitTest 208 passing, :composeApp:jvmTest 214 passing -- both test tasks re-run from scratch rather than taken from the cache. The jvm figure is larger than the android one because jvmTest inherits commonTest, so declaring the target quietly gained the whole shared suite a second execution environment. That is worth knowing independently of whether desktop ever ships: the same tests now run on the host, without an emulator. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 02:01:35 +02:00
Kgothatso Ngako	adf1f03817	feat: a desktop entry point, and the first code here that runs ... Phase 5. `press.mantra.desktop.MainKt` has been named by the compose.desktop block since before this work started and did not exist; now it does, and `./gradlew :composeApp:run` opens a window. **The window opens onto a passphrase gate, not onto the app.** That is phase 3 landing here rather than there, and it was not in the plan. keyStoreEncryption(keyName, plainText) takes no secret, because on android the OS keystore serves keys without asking anybody anything -- so a passphrase scheme needs an unlock the expect signature cannot express. MainKt calls JvmKeyStore.unlock before MantraApp is composed, off the ui thread, because Argon2id at 64 MiB is deliberately slow enough to stop the window painting. The gate says on its face that this build is not for real funds. One application directory is handed to both the mantra and the phoenix context, so a single install keeps a single place on disk rather than two named after different projects. **MantraDatabaseJvmTest is the part worth keeping.** Running the app proves the window paints; it proves nothing about Room, because the gate stops before anything touches the database. Six tests now open it: the schema is created, a profile survives a write and a read, upsert replaces rather than duplicates, the @Transaction relation query behind findChatRoomById reads back, a soft-deleted room stops being found, and the on-disk builder writes under the context directory rather than java.io.tmpdir. This is the first time this database has been opened anywhere but android, and it covers exactly what the compiler cannot see -- that Room's ksp output for this target is usable, that the *host* SQLite native loads where the android artifact's would not, and that the 58 queries forced from blocking to suspend still return what they stored. Both of that test's first drafts were wrong in ways worth keeping the scars of. Every write failed with SQLite error 787 because Profile has a foreign key onto NostrEvent and the test never created the parent row -- which is evidence rather than an annoyance, since a schema whose constraints were quietly off would have let all of it pass. And Kind is a typealias for Int, not a constructor. Window sizing is 480x900: a starting size that does not immediately misrepresent layouts only ever exercised at phone widths, not a considered desktop layout. That, along with back handling and any ui offering an nfc affordance, is the shakeout this phase names and does not do. Verified, all five green: :composeApp:compileKotlinJvm, :composeApp:compileDebugKotlinAndroid, :composeApp:testDebugUnitTest (52), :composeApp:jvmTest (6), and the fork's :library:jvmTest (97). Not verified: nothing past the gate. No seed has been written, no business started, no relay contacted. A gradle `run` killed with SIGTERM reports BUILD FAILED with exit value 143 -- that is the signal, not the app. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 01:34:10 +02:00
Kgothatso Ngako	bf4041a5b2	feat: mantra compiles for the jvm ... Phase 4. Declares jvm(), implements all 16 expects, and bumps the submodule to the fork branch carrying phases 1-3. :composeApp:compileKotlinJvm is green. **The actuals were the small half. Room was the blocker.** The first jvm compile failed with 58 copies of "Only suspend functions are allowed in DAOs declared in source sets targeting non-Android platforms". Room permits blocking query methods on android and nowhere else, so every @Dao function that was neither suspend nor Flow-returning had to change -- 58 of them across 24 files. KSP reports these in alphabetical batches, so the count shrinks in stages and looks bottomless; scanning the dao package directly for abstract funs with no suspend and no Flow return finds them all at once. It stops there, which is the only reason this is a 58-line change rather than a refactor. Every one of the 15 call sites outside the dao package was already inside a suspend function -- the repositories were written that way throughout -- so nothing needed rewriting. One private helper, DatabaseNostrRepository.matchNegentropicNostrEvents, had to become suspend, and its single caller was already suspend, so the cascade terminated immediately. Zero call-site edits. **The cost lands on android, not on the jvm.** A blocking DAO method runs on its caller's thread; a suspend one is dispatched to the query coroutine context, which getRoomDatabase sets to Dispatchers.IO. That is the better behaviour -- it is what stops a query running on the main thread -- but it is a real change to the shipping platform, made for a target that does not run yet. Hence the unit tests below rather than a compile alone. **BusinessManager was not an expect**, so nothing warned about it. It is now ported to the fork's jvmMain (05ce7eb); Phoenix.jvm.kt and NavigationViewModel.jvm.kt are otherwise the ios actuals with one changed import, since those files use no ios API. **schedulePlatformLogic schedules nothing, and logs that it does not.** Android starts two WorkManager jobs here, one of which is ChannelsWatcher -- it wakes periodically to notice a channel force-closed while the app was shut. A desktop application has no process once its window closes, so there is nothing to wake, and running the watcher in-process would be strictly worse than not running it: it would only fire while the app was already open and watching. The exposure is real and belongs in release notes rather than a comment -- a desktop wallet left closed past a force-close deadline does not notice. Smaller calls. PlatformContext carries an application directory, since there is no Context to read one from, and PlatformDatabaseBuilder puts aux.db under it rather than in java.io.tmpdir, which is what the abandoned Aux implementation did behind a TODO and which most systems clear on reboot. themeColorScheme ignores dynamicColor, which means Material You and has no desktop counterpart. AppVersion reads the jar manifest that compose.desktop writes, falling back when running from a class directory. Verified: :composeApp:compileKotlinJvm green, :composeApp:compileDebugKotlinAndroid green, and :composeApp:testDebugUnitTest 52 passing -- the one that matters, since this commit changes shared code every android query path goes through. Not verified: nothing has run. No jvm entry point exists yet, so the database has never been opened on this platform and no business has been started. That is phase 5, which also has to unlock JvmKeyStore before the wallet starts -- a passphrase prompt, not just a window. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 01:23:42 +02:00
Kgothatso Ngako	670f87a609	docs: record what phase 3 turned out to require ... Phase 3 is implemented in the fork on claude/jvm-target-actuals (ce49657). The security analysis in the plan held up; three practical constraints around it did not appear until the code was written. **A passphrase-derived KEK is not a drop-in.** The plan treated the choice between a passphrase, an OS keychain and a key file as the whole decision. But keyStoreEncryption(keyName, plainText) takes no context and no secret -- on android the OS holds the key, so none is needed -- which means any passphrase scheme needs an out-of-band unlock the expect cannot express. That is a change to application startup, not just to the actual, so it is now called out against phase 5: the desktop entry point has to prompt and unlock before the wallet starts. **The iv must be 16 bytes.** EncryptedSeed.V2.serialize in commonMain throws on anything else, which rules out a conventional 96-bit GCM nonce -- worth knowing before designing around one. It turns out to help: with randomly generated nonces the risk is a repeat under one key, and 128 bits makes that vanishingly unlikely where 96 merely makes it unlikely. **Argon2id costs a dependency.** The jdk has PBKDF2 and no memory-hard KDF at all, so it means bouncycastle. Recorded with the reason to pay it: if the build is dev-only because it lacks hardware backing, weakening the KDF too gets the trade backwards. Also recorded: wrap a per-name data key under the KEK rather than encrypting the seed with it directly, so a passphrase change rewraps 32 bytes; throw java.security.KeyStoreException when locked, since the graceful* wrappers already map it to DecryptSeedResult.Failure.KeyStoreFailure; and a verification section naming the properties that fail quietly, plus the two limits worth writing down rather than fixing -- the first unlock on a new store accepts any passphrase, and zeroing the key is best effort. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 01:13:07 +02:00
Kgothatso Ngako	506dfea802	docs: correct phase 2 against what the jdbc drivers actually needed ... Phase 2 is implemented and verified in the fork on claude/jvm-target-actuals (d1a82ea). Three corrections and one omission. **Schema handling does not need hand-rolling.** The plan said "you call Schema.create(driver) and the migration path explicitly, and you have to track the applied version yourself". SQLDelight 2.x ships a factory function that shadows the constructor -- JdbcSqliteDriver(url, properties, schema, migrateEmptySchema, vararg callbacks) -- which does all three, user_version included. The same-named constructor does none of it, which is the trap worth naming rather than the work that was budgeted for. **Foreign keys were the actual work, and the plan never mentioned them.** Off by default in SQLite, and the pragma is per connection while JdbcSqliteDriver opens one per thread, so it has to go through the connection Properties rather than be issued once against the driver. Recorded along with why that needs no compile dependency on org.xerial:sqlite-jdbc, which arrives at runtime scope only. **commonTest has an expect too.** The 23 counted at the top of this document are commonMain's. Declaring jvm() also creates jvmTest, which inherits commonTest, so `connect` in ElectrumServersTest blocks every jvm test from compiling. Noted along with the reason not to stub it empty the way ios does: the class is @Ignore'd everywhere, so an empty body looks harmless right up until somebody removes the @Ignore and connect_to_mainnet_servers starts passing without connecting to anything. **Phase 2 is the first phase that can be run, and the plan told you not to bother.** It said "none of this is exercisable until Phase 4. Write the SQLDelight schema-creation path against a scratch main() if you want feedback sooner." That was wrong twice: library/src/jvmTest/ already exists, and the two properties worth checking are exactly the ones a compiler cannot see. Schema creation and the foreign-key pragma both fail silently in production -- a missing table only shows up at first query, and foreign keys being off means cascading deletes quietly do not happen. The phase now carries a real exit condition, and DbFactoryJvmTest meets it with five passing tests. Recorded with it: the two KeyStoreFunctions actuals have to exist before phase 3 decides anything, because nothing jvm compiles without them, and they should throw rather than do something plausible. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 01:04:02 +02:00
Kgothatso Ngako	0757e50dc5	docs: correct the jvm plan against what phase 1 actually did ... Phase 1 is implemented and verified in the lightning-kmp-app fork on claude/jvm-target-actuals (27a0054). Four things in the plan were wrong, and doing the work is what surfaced them. **jvm() belongs at the start of phase 1, not phase 4 -- for the library.** The plan said leave it off in both builds until phase 4. That is right for mantra and wrong for the fork: library/src/jvmMain/ is an orphan source set until the library declares the target, so phases 1-3 would all have been written blind. Declared first, `:library:compileKotlinJvm` names the remaining expects, and that list beats grepping for `expect ` -- it shrinks by exactly what you implement and cannot drift from the truth. The build stays red across phases 1-3 by design. That checklist is now recorded as the phase 1 exit condition: exactly eight expects should remain, and exactly which eight. Anything else means something in the phase is wrong. **Phase 3 is two decisions, not four.** gracefulSingleSeedDecryption and gracefulMultiSeedDecryption are pure exception mapping into a DecryptSeedResult, and the exception they branch on is java.security.KeyStoreException -- a plain JCA type that exists on the jvm. Both are near-copies of the android actuals and need nothing settled first, so they move alongside phase 2. Only keyStoreEncryption and keyStoreDecryption are the security decision, and that part of the analysis stands. **The Fibonacci template must not be deleted.** The plan said to drop it "assuming nothing references them". Things do: generateFibi is exercised by template tests in commonTest, androidHostTest, iosTest, jvmTest and linuxX64Test, and JvmFibiTest asserts a value that depends on precisely the two properties fibiprops.jvm.kt defines. That file already satisfies two of the 25 expects, which is why the count was 23 missing rather than 25. Removing the template is five test files plus four fibiprops.* actuals, and it is a separate cleanup. **Phase 1 is fifteen actuals, not fourteen**, and two of them are not copies of android -- platformElectrumRegtestConf (10.0.2.2 is the emulator's alias for the host loopback; a jvm process is already on the host) and phoenixLogWriters (android routes kermit into slf4j because android tooling reads that back). Also recorded, because it cost time: a worktree cannot run gradle at all until the submodules are checked out *and* local.properties exists at five levels. Neither is version controlled, so a fresh worktree has neither, and the failure surfaces four builds down at :...:secp256k1-kmp:jni:android as "SDK location not found" rather than anywhere obviously related. Both builds were run: `:library:compileKotlinJvm` fails only on the known eight, and `:composeApp:compileDebugKotlinAndroid` still passes with the library's jvm target declared -- the check that matters, since a new variant must not change how the android target resolves the library. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 00:53:51 +02:00
Kgothatso Ngako	2aaa7b99a6	build: phase 0 of the jvm target -- clear the ground, correct the plan ... First phase of docs/jvm-target.md. Nothing here turns the target on; it removes what would break the moment it is turned on, and stages the two catalog entries that cannot be derived automatically. Two of the four steps as written in the doc turned out to be wrong, and implementing them is how that surfaced -- both are corrected in the doc in this commit. **Deleted the stale jvmMain tree.** Six files under composeApp/src/jvmMain/kotlin/ac/cord/auxiliary/ survived from the Aux project this codebase grew out of. They have gone unnoticed because `jvmMain` is currently an orphan source set -- the accessor creates it, no target compiles it -- so the wrong package, the Room 2 imports (androidx.room, not androidx.room3), and the references to a long-gone AuxDatabase and AuxGlobal have never had to resolve. They would all become compile errors in phase 4. They are not lost: they are the closest thing to a skeleton for five of the six platform actuals phase 4 needs, and main.kt is a reasonable starting shape for the phase 5 desktop entry point. `git show HEAD~1` has them. **Added two catalog entries, not four.** sqlite-bundled-jvm and sqldelight-sqlite-driver. Both earn their place by being unreachable otherwise: sqlite-bundled-jvm has to be named explicitly because variant-aware resolution hands the *android* artifact to anything running on the host, and sqldelight-sqlite-driver is the jvm counterpart to the android-driver and native-driver entries already there. The doc also listed room3-runtime-jvm and sqldelight-jdbc-driver. Neither is right. Once jvm() exists, commonMain's existing androidx-room3-runtime resolves to the -jvm variant on its own, so an explicit entry is redundant and would drift. And the SQLDelight drivers phase 2 needs are for DbFactory, which lives in lightning-kmp-app -- a separate gradle build with its own version catalog, where an entry here is simply not visible. **kspJvm cannot be wired yet, and the build file already said so.** The doc's phase 0 told you to uncomment composeApp/build.gradle.kts:194. It contradicted its own phase 4, which is where jvm() gets turned on. The comment three lines above it states the rule: These configurations only exist when the ios targets are declared, which the kotlin block above does only on a mac. The same holds for kspJvm -- `dependencies { add("kspJvm", ...) }` throws UnknownConfigurationException until a jvm() target creates the configuration. So it moves into phase 4, into the same edit that declares the target. composeApp/build.gradle.kts is deliberately untouched by this commit. **Also documented: gradle does not run in a worktree here at all** until the submodule is checked out, which worktrees do not do automatically. `lightning-kmp-app/` is empty and configuration fails with "Project with path ':library' not found in build ':lightning-kmp-app'". Recorded in the phase 0 verification section along with the caveat that a linked worktree shares .git/modules/ with the main checkout, so both trees end up on one submodule git dir. **Not verified by a build.** For that reason. The deletion is an orphan source set and the additions are unreferenced catalog lines, so neither can change a build's outcome -- but that is an argument, not a green check, and it is the second commit in a row on this branch that has not compiled anything. Phase 4 is the first phase that genuinely cannot be done without a working gradle invocation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 00:44:10 +02:00
Kgothatso Ngako	65e4a3acc0	fix: seal the Welcome, the one gift wrap an MLS room must publish ... No invite to a Marmot room has been delivered since 1700e6d. The Welcome was built, hashed, queued and logged exactly as before -- and then refused one step short of the relay, by a guard that had no idea it was looking at one. **The guard.** 1700e6d ("send a direct message into the group, wrapped for one member") added a backstop at the top of sealGiftWrapPayload: val chatRoom = database.chatRoomDao().findChatRoomById(giftWrapPayload.chatRoomId) if (chatRoom?.chatRoom?.mlsGroupState != null) { log; return } Its reasoning is sound and still is: a Marmot direct message is a genuine, correctly signed NIP-59 gift wrap, indistinguishable from one this path would be right to publish, and the only thing keeping it off a relay is that it never becomes a GiftWrapPayload row. Refusing at the seal as well means a future caller cannot walk one onto a relay by accident. **Why it caught the Welcome.** MarmotOutboundDao.deliveryWelcome writes a GiftWrapPayload with chatRoomId = nostrGroupId -- the MLS room's own id, which by construction has mlsGroupState set. Every Welcome therefore matched a refusal keyed on mlsGroupState alone. There is no Welcome that does not: the tag identifying the room is the whole point of the event. The kind:444 arm further down -- the one that wraps only for the participant who published the referenced key package -- became unreachable, which is why nothing in the logs said "welcome" at all. **Why it went unnoticed.** The room reaches the correct state on the inviter's side whether or not the Welcome goes out: addMember advances the epoch, the group state is persisted, the Participant row exists, and "Invited X to chat" is written to the transcript. From the coordinator's side an invitee who never heard anything is indistinguishable from one who joined -- recorded as a known gap in docs/shared-key-ceremony.md, and this is what was behind it. **The second-order damage.** observeUnsealedGiftWrapPayloads is SELECT * FROM GiftWrapPayload WHERE publicKey = :p AND giftWrapSealId IS NULL collected as a Flow<GiftWrapPayload?> -- one row at a time. giftWrapSealId is only ever set inside persistAndBroadcastGiftWrap, which the guard returns before reaching, so a refused payload stays unsealed forever and sits at the head of that queue. The first Welcome a user queued blocked every gift wrap behind it, in every room, for the life of the install. **The fix.** Kind first, room second. MIP-02 addresses kind:444 to someone who is not yet in the group and holds no key to read a kind:445 -- a relay-borne gift wrap is the only way to reach them, and deliveryWelcome queues one on purpose. Every other kind is refused exactly as before. sendChatMessage already branches on mlsGroupState before writing a payload, so an MLS room's messages never arrive here anyway; the guard stays as the backstop it was meant to be. docs/marmot-direct-messages.md claimed "An MLS room should never produce a NIP-17 gift wrap for any reason". That premise is what made the guard look complete, so it is corrected rather than merely amended, with the test to apply when adding a kind to the exemption: can its recipient read a kind:445? If so, it does not belong on this path. **Not fixed, deliberately.** A refusal still leaves the payload unsealed and head-blocking. That is now unreachable -- nothing else can queue a payload against an MLS room -- but it remains a trap for whatever gets refused next. Marking a payload refused needs a state the schema does not have, so it is left for its own change rather than smuggled in here. Verified: :composeApp:compileDebugKotlinAndroid succeeds. No test covers this -- DatabaseChatRepository is Room-backed, and Room-backed code has no unit test harness in this project. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 00:40:31 +02:00
Kgothatso Ngako	5abc37e463	docs: scope the jvm target, and separate it from testing the daos ... Two questions arrived together -- whether Room's own testing guidance applies to this project, and what desktop support would cost -- and they turned out to have opposite answers. Both are now in docs/jvm-target.md, phased, with the blocking work separated from the mechanical work. **The expensive part is already done.** The four-deep native chain -- secp256k1 -> bitcoin-kmp -> lightning-kmp -> lightning-kmp-app -- already builds for JVM, on every android build we do. The comment at composeApp/build.gradle.kts:50 records the mechanism without drawing the conclusion: lightning-kmp-core publishes no android variant, so our android target resolves it to the *jvm* one, which pulls secp256k1-kmp-jni-jvm desktop natives, which is exactly why the build has to name the android artifact by hand. Read the other way round, every JVM artifact in the chain is already compiled from source by the composite build. A jvm target adds no cinterop, no C compilation and no new native constraints. That was the part worth being afraid of, and it is finished. **The blocker is one level down, and smaller than it looks.** lightning-kmp-app/library declares 25 expects and implements them across 35 androidMain files. Its jvmMain holds exactly one: fibiprops.jvm.kt, the Kotlin multiplatform library template's Fibonacci boilerplate, satisfying two of the 25 -- both of them the template's own. So 23 actuals are missing, which is why jvm() is commented out there (library/build.gradle.kts:18), which is why it is commented out here (composeApp/build.gradle.kts:46). Mantra cannot declare the target until the fork does. Six phases, ordered by that dependency. 0 build config; 1 the fourteen mechanical phoenix actuals; 2 the three SQLDelight JDBC drivers and NetworkMonitor; 3 key storage; 4 mantra's own sixteen expects; 5 the desktop entry point. 1-3 are independent and parallelisable, 4 is where the compiler finally checks the whole thing. Roughly a week to a launchable build. **Phase 3 has no day estimate, deliberately.** keyStoreEncryption / keyStoreDecryption and their two graceful* wrappers delegate on android to KeystoreHelper.kt -- 116 lines against AndroidKeyStore, StrongBox attempted first and fallen back from, key material never leaving hardware. Desktop JVM has no equivalent, so this is a decision rather than a port, and the doc gives the three real options against what each actually protects. A fixed-key JCEKS file is named there as a liability rather than a stopgap: this is wallet seed material, and it lands on top of the plaintext-key finding already open against this codebase. Recommended sequencing is a passphrase-derived KEK with the desktop build marked unsuitable for real funds, so phases 4 and 5 can proceed without the security question being quietly treated as answered. Two inherited mistakes are called out rather than carried forward. The old Aux jvmMain put the database in java.io.tmpdir behind a TODO -- the doc says not to inherit that in either phase that touches it. And schedulePlatformLogic goes through WorkManager on android with no desktop counterpart, so the doc asks for an explicit choice between a no-op and an in-process coroutine, written down. **The DAO answer is an appendix, because it is the opposite answer.** None of the above is needed to test the DAOs, and burying that would have been misleading. room3-runtime-android:3.0.1 already exposes the no-Context inMemoryDatabaseBuilder(Function0<T>) overload, and MantraDatabaseConstructor already supplies what it needs, so Room's recommended host-machine form compiles in commonTest and runs under testDebugUnitTest today. The one trap is native and is the secp256k1 problem mirrored: sqlite-bundled-android ships only android-ABI .so under jni/, so a local unit test's JVM cannot load it and BundledSQLiteDriver fails at construction; sqlite-bundled-jvm on the androidUnitTest classpath is the fix. Robolectric neither helps nor is needed -- it cannot load android .so on the host either. Everything structural here was checked against the artifacts rather than recalled: the Room builder overloads by javap on room3-runtime-android, the two sqlite-bundled native layouts by unzipping both, and the availability of room3-runtime-jvm, room3-testing, quartz-jvm and the two SQLDelight drivers by request against the repositories this build actually resolves from. The absence of android.* and java.* imports in commonMain, and of any NFC reference from it, was likewise grepped rather than assumed. **Not verified: anything that requires compiling.** No jvm target was turned on, nothing was built, and the day estimates are estimates. Phase 4 is where dependency-substitution surprises would surface if there are any, and it is precisely the phase nothing here exercises. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-06 00:38:11 +02:00
Kgothatso Ngako	39eac61838	Merge branch 'mantra' into claude/long-running-chat-sync-8983dc ... mantra had moved on ~30 commits, several of them in exactly this area — and it turns out both branches independently found the same bug and drew the same conclusion about the same filter. **The overlap.** f38a5f1 fixed the three kind:1059 filters that named the wrong pubkey, including the two `authors=[userPublicKey]` requests in NostrDao that could never match a wrap signed by a throwaway key. This branch deleted those same two blocks, inverting the same `if` to the `== null` case, for the same reason. The code merged to the same shape; only the comments conflicted, and they are combined. **Nip17Filters wins, and the live subscription now defers to it.** ad3304a extracted the inbox filter to one definition precisely because it had been wrong in three call sites, with the no-`since` reasoning this branch arrived at separately. Keeping a fourth copy inside LiveSubscriptionManager would recreate the problem that commit exists to solve, so: - queueCatchUpSynchronization now calls Nip17Filters.inbox() instead of building an identical SynchronizationFilter with its own limit constant, - Nip17Filters gains liveInbox(), the same shape as a quartz Filter for a REQ rather than a SynchronizationFilter for the queue, and giftWrapFilter() defers to it. Two types for one filter is not duplication worth removing — the queue stores one and hashes it for computeId, a live subscription puts the other on the wire — but they belong side by side, because drift here means one of them quietly stops matching mail. **ChatMessageListViewModel keeps this branch's resolution.** mantra had it refresh our own inbox on open (Nip17Filters.inbox on our DM relays, purpose "chat"); this branch removed that call entirely. Both were right when written, and the merge is where the second becomes true: LiveSubscriptionManager holds exactly that filter open on exactly those relays for the whole account and reconciles it on every foreground, so opening a chat has nothing left to ask for. The redundancy is now recorded in the comment where the branch used to be, so it reads as superseded rather than dropped. Discovery — the kind-10050 lookup for a participant we cannot yet address — is untouched, and the purpose is no longer a conditional now that only one case reaches it. The commonTest coroutines-test dependency arrived on both sides; the comment gives both reasons. Verified: 154 tests pass, both branches' suites included — Nip17FiltersTest and the marmot direct-message suites alongside this branch's 46. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-05 23:58:40 +02:00
Kgothatso Ngako	bcdfd2ec94	Merge branch 'mantra' into claude/marmot-direct-message-type-7a0473 ... Twenty-two commits had landed on mantra since this branch left it, several of them in the same files. Merged this way round so mantra stayed untouched until the result compiled and its tests passed. The migration had to be renumbered, and this is the conflict that mattered. mantra is at database version 7 and already has its own 5.json -- for MarmotInnerEvent.payloadEventId, nothing to do with direct messages. This branch had also written a 5.json, for a different schema. Resolved by restoring mantra's 5.json untouched and moving the direct message columns to an AutoMigration(7, 8) with a regenerated 8.json. Taking either 5.json over the other would have left every device validating a migration chain against a schema it was never built from; keeping version = 5 would have made a v7 install refuse to open at all. The regenerated 8.json is two ADD COLUMNs and nothing else, same as before. fromGroupEventResult was restructured on mantra: the kind switch moved into applyInnerEvent, and a SubmissionEvent envelope now wraps nip30303 payloads. Took that structure and re-applied the direct message branch ahead of it rather than inside it -- a gift wrap is not a nip30303 payload to apply, and what happens to it depends only on whether this device's key opens it, so it does not belong in a function about applying submissions. The isUserMessage fix was re-applied to the eight call sites mantra's version has, up from the six it had here. ChatMessageListViewModel and ChatRoomMessagingScreen took mantra's versions with the composer state, the two renderings and the reply action layered back on. docs/README.md keeps both new rows and mantra's closing note about the skipped-keys document. 108 tests pass, up from 50 here and 83 on mantra. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-05 23:45:08 +02:00
Kgothatso Ngako	c8cbd936f1	docs: record where the sync's safety net is, and where it is not ... Two updates after the test pass. long-running-sync.md gains a section naming what each test file pins and, more usefully, the three things they cannot reach: NostrSocketClientImpl's reconnect loop and ordered inbound (exercised only through their extracted arithmetic — covering them wants a fake WebSocketSession), everything downstream of saveNostrEvent (Room-backed, and there is no sqlite driver on the JVM test classpath), and the app on a device. The manual checks stay the manual checks. It also records that the tests were verified by mutation rather than by passing, so the next person knows the assertions were confirmed to bite. dead-code.md's line references are refreshed — the testability seams shifted most of them — and it now says which commit they were correct at and to confirm with the grep rather than trusting them. One entry added: the DefaultNostrSocketClientFactory overload taking an explicit HttpClient has no caller now that everything goes through the interface method. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-05 23:32:53 +02:00
Kgothatso Ngako	c24cbed390	docs: record what the coverage work found, and what it left uncovered ... Three additions. The decision the inbound path makes now has a name and a home -- MarmotDirectMessage.classify -- and the doc says why it is separate from the filing of it: only the filing needs a database, so splitting them is what lets the check that replaces MIP-03 be tested at all. A security property found while writing those tests, which I had asserted backwards. Relabelling a seal with another member's pubkey does not get as far as the signature check: NIP-44 derives the conversation key from the pubkey being claimed, so a relabelled seal is undecryptable by the person it was encrypted for. The label is bound to the key rather than asserted alongside it, and the outcome is a message the recipient genuinely cannot read. verify() catches the narrower case of a seal altered after signing in a way that survives decryption. An honest list of what has no automated test and why -- the recipient validation and the outbound id lookup (both need a database), the two transcript renderings (no Compose UI test dependency in this project), and anything touching a real MlsGroup. Better written down than rediscovered by someone assuming a green suite means the path is covered. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-05 23:18:19 +02:00
Kgothatso Ngako	d110737f9a	fix: keep a room's MlsGroup alive so a late message can still be read ... Two events published in the same second reliably lose one of them. The receiver stores the kind:445 and produces nothing from it -- no inner event, no chat line, no error anybody sees, because MarmotGroupEvent is written before the message is decrypted and so survives while everything downstream silently does not. Observed as a FROST signing session that never started on the receiver: proposeSigning publishes the proposal and then the proposer's own nonce, the relay handed them back in the other order, and the proposal was dropped. The nonce is still sitting there filed against a session that will never exist. The same bug ate a dialect earlier, which then took out the artifact referencing it via a foreign key. MLS is specified to tolerate this. RFC 9420 says a receiver that gets generation N+1 before N keeps the intermediate keys so the older message can still be read, and quartz's SecretTree does exactly that, in a private skippedKeys map. What it does not do is persist it: exportSenderStates() returns the ratchet positions only, so saveState() drops the cache. NostrDao rebuilt the group from stored state for every inbound event, so the cache was empty every single time, and generation N arriving after N+1 failed `require(generation >= applicationGeneration)` and was swallowed. Terminal -- the key is derived from a ratchet that has moved past it, and nothing asks the sender to resend. This keeps the instance alive instead. MlsGroupCache holds one MlsGroup per room, and the inbound path goes through it, so skippedKeys survives from one message to the next. That covers the case that actually bites -- a burst arriving in one sync, decrypted one after another against the same tree -- which is what every bursty flow needs: proposeRitual sends two, addArtifact sends two, and addChapter sends one per paragraph plus one, of which only the ones arriving in ascending generation order survived. Reuse is conditional on the stored state still being exactly what the cache last wrote. Sending a message advances the sender ratchet and saves; so does adding a member. When that happens the cache rebuilds rather than carrying on from a group that has been overtaken -- which is what keeps this from being worse than no cache at all: the fallback is always the old behaviour, never a diverged ratchet. One lock per room, not one overall, because the group is mutable and decryption advances it: two events for the same room decrypted at once would corrupt the tree, and a busy room should not hold up a quiet one. **This is a mitigation, not the fix.** It does not survive a restart, and it does not survive another writer, so a long enough reorder still loses the message. The fix belongs in quartz -- carry skippedKeys through saveState/restore -- and quartz is a mavenCentral binary, not a fork, so it cannot be made here. docs/mls-skipped-keys.md has the analysis, the patch, the migration constraint on the persisted state format, and the three ways to actually land it. Not verified end to end: the proposal that exposed this cannot be recovered, since its generation is already past, so confirming the fix needs a fresh burst. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-05 23:09:53 +02:00
Kgothatso Ngako	635cef9311	docs: write down how a direct message travels, and what it costs ... The reasoning behind this is not recoverable from the code, which is the bar docs/README.md sets for having a document at all. Three things in particular would otherwise have to be rediscovered by whoever changes this next, and two of them are traps. Why the wrap uses a throwaway key rather than the sender's own -- and what that does not buy. It does not hide the sender from the group: MLS authenticates every application message to a leaf, so the identity is there regardless. What it costs is a carve-out in MIP-03's pubkey check and the sender's ability to ever read their own messages back. Why the check that carve-out removes is not a hole. The authorship claim moves from the wrap's plaintext pubkey to the seal's verified signature, bound to the MLS leaf that sent it -- strictly harder to forge than what it replaced. The one query that would broadcast one of these. What this builds is a genuine, correctly signed NIP-59 gift wrap, indistinguishable from what the NIP-17 path would be right to publish, and the only thing keeping it off a relay is that it never becomes a GiftWrapPayload. Written against what shipped rather than what was planned, so it records two deviations. senderIdentity is resolved in NostrDao rather than added to GroupEventResult.ApplicationMessage, because quartz is a binary dependency here and the local checkout is a reference copy, not a build input. And a failed validation drops the message and logs rather than throwing, because the caller is inside storeNostrEvent's transaction. The unbuilt parts are listed as absences rather than left implied: there is no member picker, so a private message can only be a reply to one somebody already sent, and nothing in the UI yet tells a user in words that the group can see who they messaged. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-05 19:10:22 +02:00
Kgothatso Ngako	c8c962e4f4	docs: inventory the unreferenced code in the sync and relay stack ... Found while building the long-running sync. One item was orphaned by that change; the rest was already dead and only became visible because the subsystem was being read closely. Written down rather than deleted because several pieces are one decision away from being wanted, and those decisions are not the sync change's to make. Every claim is "this identifier appears exactly once in composeApp/src, at its own declaration", with the two things that method cannot see called out: Room DAO methods are reached through generated code, and Compose entry points can be invoked without a textual reference. The DAO cluster is flagged as the least certain for exactly that reason. Three findings are more than leftovers: - RelaysSocketManager.userRelays is a field nothing ever writes. The `userRelays` inside observeRelays is a different, shadowing local, so the single-argument publishEvent always takes its FALLBACK_RELAYS branch and the user's own relay list is never used for publishing. That is a bug wearing dead code's clothes, and the fix is to populate the field, not to delete it. - NostrPublisherRepository is entirely unreferenced, and it is the only consumer of CachingImportRepository.importEvents. RelayPool and RelaysSocketManager each take a cachingImportRepository parameter they store and never dereference, satisfied by NO_OP_CACHING_IMPORT_REPOSITORY — so the whole seam is a parameter passed from nowhere to nothing. Removing the publisher lets the interface and both parameters go with it. - sendAUTH is unused because NIP-42 is unimplemented, not because it is surplus. AuthMessage is parsed and dropped, so a relay answering CLOSED with auth-required is retried forever and can never succeed. Deleting sendAUTH means deciding against authenticated relays; that is worth doing on purpose or not at all. sendCOUNT and CountMessage are a similar matched pair — both go or neither, since a CountMessage cannot arrive if nothing sends a COUNT. isRecommendedRelay on the two request entities is separated out as its own risk class: never written, never read, but a Room column, so it wants a migration rather than a delete. Ends with an order to do it in, cheapest and least risky first. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-05 16:28:21 +02:00