`3bc8be53` renamed Mantra to Curated by searching for "Mantra", which is exactly why
it missed this: the tree carried a second, older brand -- Torch, from the rename before
that one -- in 130 references and two error messages. A user could still be told to go
and install a product that does not exist.
**`UserAgent` was the last thing publishing the old name, and the reason it could move
is the reason the `mantra/` tags could not.** Both were described the same way -- "it
goes on the wire" -- and that phrase hid two different facts. `CLIENT_NAME` has one
caller: it is the `client` tag on a published NIP-65 relay list, which is public
attribution and nothing else. It is not an input to a hash, so changing it cannot
re-derive a key, cannot orphan a room, and cannot make two clients disagree about a
threshold key; the worst it does is label future relay lists correctly. `TWEAK_TAG` and
`HOST_KEY_DERIVATION_TAG` are hashed, which is the whole difference, and they stay.
`APP_NAME` moved with it and has no callers at all -- dead, and left that way, because
deleting it is not this change.
**`TorchTheme` is `CuratedTheme`, at 130 references across 70 files.** It is the root
theme composable, so the rename reaches every screen, every `@Preview` body, the desktop
entry point and six jvmTest files that compose against it, plus 13 mentions in
`docs/material-design-conformance.md` that document it as a live API rather than as
history. Wide, but mechanical: it is one identifier with no string, table, path or hash
depending on its spelling, which is what makes it the opposite of the 12 Room entities
that kept theirs.
**Two error strings named the wrong product to a user.** "tell <name> to use Torch" and
"they may not have finished setting up Torch" both now say Curated, along with the
commented-out preview fixture beside the first, so the two cannot drift apart if it is
ever uncommented. They stay inline literals rather than moving to the catalogue: both are
set inside a `viewModelScope.launch`, where `stringResource` is not available, and of the
44 view models the five that reach `Res.string` all do it from `@Composable` functions
that merely live in view model files. Nothing here reaches the catalogue from view model
logic, so externalising these would introduce a `getString` pattern the tree does not
have -- worth doing, and worth doing as its own change rather than inside a brand fix.
**The rename caught its own prose, twice.** A comment in `HomeScreen` and a paragraph in
`material-design-conformance.md` each said "`TorchTheme` and two error strings still carry
the older brand" -- true when written, and turned by this very sed into "`CuratedTheme`
... carries the older brand", which contradicts itself. Both now state what is true. It is
worth noting because a global identifier rename edits the sentences that describe the
rename, and those are the ones nothing compiles or tests.
**Three references to "Torch" survive, and rewriting them would destroy information.**
`material-design-conformance.md:278` is the finding that the app once had three names for
one product; the note at :654 is the deferral this change carries out, so it has to say
what was deferred; and `NavigationRoutingTest`'s `"Introducing... Torch"` is documented as
"the screen the bug left people on" -- a test-only fixture no production path emits, whose
value is that it names the state the regression was about. No live code answers to Torch
or to Mantra any more; what is left of both is the record of what was.
1136 tests pass -- 733 in `:composeApp:jvmTest`, 403 in `:composeApp:testDebugUnitTest` --
`:composeApp:compileDebugKotlinAndroid` is clean, and `m3Audit` meets every budget.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>