TT-7643 fix: name a Phrase BT take for its step language so it cannot collide

[nabalone]

Problem

TT-7643 — with a Phrase Back Translation step per language (Sena, Hebrew, English), audio recorded in one language plays in another language's PBT step.

On the desktop app a take's audio is resolved by file name, not by id. useFetchMediaUrl calls dataPath(mediafile.attributes.audioUrl, PathType.MEDIA), which maps to <offlineData>/media/<basename(audioUrl)> and returns that local file if it exists — the mediafile id is never consulted. store/upload/actions.tsx also stages every online upload into that same folder, so the cache is populated as you record.

The upload name is passageDefaultFilename(...) + config.buildFilenamePostfix(unitIndex, sourceVersion). It carried the passage, the segment index and the source version — but not the step language. A team configures one PBT step per language and every one of them records the same segment of the same vernacular, so Hebrew segment 1 uploaded under exactly the name Sena segment 1 had already cached. First file cached wins, and that is Sena.

The naming is visible in a real cache:

~/transcriber/media/GEN001_001-010backtranslation1_v1_lofflin.ogg
                    GEN001_001-010backtranslation2_v1s1_lofflin.ogg
                    GEN001_001-010backtranslation3_v1s2_lofflin.ogg

One file per segment, no language anywhere in the name.

What this supersedes

The first two commits on this branch were aimed at component reuse, on the reading that the step instance carried its take across a step change. That is not what the ticket reports: the ticket video is the desktop layout, and PassageDetailGrids has had <Paper key={currentstep}> all along, so the step was already remounting. Those commits stay because they are correct on their own terms — the mobile route lacked that parity, and the segment hook's reset had to key on the language bucket as well as the audio (Devin's finding) — but they are hardening, not the fix. The row-level language filter from #490 is likewise sound: a CT test on the real step shows a Hebrew step never fetches a Sena take.

Changes

• guidedPhraseRecord/types.ts — buildFilenamePostfix takes the step bcp47 and appends it: backtranslation1_v1_seh vs backtranslation1_v1_he. Steps with no configured language (Careful Speech) keep the names they always had, so existing files are untouched.
• PassageDetailGuidedPhraseRecord.tsx — passes stepLanguageBcp47 into the postfix builder.
• cypress/support/pbtHarness.tsx — the fake server stamped every uploaded take English|en instead of the language that was posted, which hid whether the step language reached the upload at all.

Retained from the earlier commits (hardening, not the fix):

• routes/PassageDetail.tsx — key={currentstep} on the guided-record steps in MobileStep, matching desktop and the TT-7503 keyed transcription step.
• PassageDetailGuidedPhraseRecord.tsx — per-step reset gated on `${mediafileId}|${currentstep}`; claimRanRef released with it.
• carefulSpeech/useGuidedPhraseSegments.ts — resetForScope keys on the BT:<bcp47> bucket as well as the mediafile, so a step change is not rejected as a no-op.
• carefulSpeech/claimLegacyPhraseBt.ts — the legacy language claim is skipped once another BT:<bcp47> bucket has regions, instead of stamping every untagged take with whichever step opened first.

Test plan

Failing-before / passing-after, all against unfixed source:

• guidedPhraseRecord/types.test.ts — buildFilenamePostfix separates the languages of the same segment.
• PassageDetailGuidedPhraseRecord.stepScope.test.tsx — "names a take so it cannot collide with another language of the same segment": drives the real component through a Sena step then a Hebrew step and reads the filename it hands MediaRecord. Pre-fix both steps produce GEN001_014-019backtranslation1_v1_plan.
• PassageDetailPhraseBackTranslate.cy.tsx — "uploads a take under a name no other language can take": records a real take on a Hebrew step and asserts the posted original-file carries the language.

Also on this branch: the step-scope reset tests, useGuidedPhraseSegments.test.tsx (real hook, bucket change on the same audio), the "PBT language scoping" CT pair, and 3 new claimLegacyPhraseBt cases.

• npx jest (full renderer suite) → 1389 pass, 3 skipped; 1 pre-existing unrelated failure (EditReferenceDropdown.test.tsx, MUI Drawer module resolution) present on develop too.
• tsc --noEmit clean; Prettier clean on the touched files.
• Manual: on a team with a PBT step per language, record every segment in one language, then record them in another language's step, and confirm each step plays back its own take. Takes recorded before this change keep their colliding names — re-record to verify.

Known, not fixed here

Re-recording the same segment in the same language reuses the name too, so the earlier take's cached audio can play for the new one. Same defect class, outside this ticket — flagging rather than folding in, since a per-take unique token changes every filename users see in exports.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR fixes cross-language state/audio bleed when teams configure multiple Phrase Back Translation (PBT) guided-record steps (one per language) against the same vernacular audio. It ensures guided-record UI state and mounted takes are scoped to the current step/language, particularly on mobile where the route previously reused an unkeyed component instance.

Changes:

• Key mobile guided-record steps by currentstep so navigating between adjacent per-language steps remounts the step UI instance.
• Scope guided-record reset logic to ${mediafileId}|${currentstep} (not just mediafileId) and reset legacy-claim state per step.
• Prevent claimLegacyPhraseBt from stamping untagged takes with a language once multi-language PBT boundaries exist, and add unit + Cypress coverage to lock in behavior.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/renderer/src/routes/PassageDetail.tsx	Adds key={currentstep} to guided-record mobile steps to force remount between adjacent per-language steps.
src/renderer/src/components/PassageDetail/PassageDetailGuidedPhraseRecord.tsx	Scopes reset behavior to step+mediafile and releases per-step legacy-claim gating when step changes.
src/renderer/src/components/PassageDetail/carefulSpeech/useGuidedPhraseSegments.ts	Renames/reset logic to resetForScope and keys the boundary reset on (mediafileId, namedRegion) (language bucket).
src/renderer/src/components/PassageDetail/carefulSpeech/claimLegacyPhraseBt.ts	Skips language-claiming untagged takes when other-language boundary buckets are present with regions.
src/renderer/src/components/PassageDetail/PassageDetailGuidedPhraseRecord.stepScope.test.tsx	New test verifying a reused guided-record instance resets cleanly when currentstep changes.
src/renderer/src/components/PassageDetail/carefulSpeech/useGuidedPhraseSegments.test.tsx	New test verifying boundary reset re-reads when only the bucket/language changes on the same audio.
src/renderer/src/components/PassageDetail/carefulSpeech/claimLegacyPhraseBt.test.ts	Adds cases covering multi-language boundary presence and empty/own-bucket scenarios.
src/renderer/src/components/PassageDetail/PassageDetailCarefulSpeech.test.tsx	Updates mocks to the new resetForScope API.
src/renderer/src/components/PassageDetail/PassageDetailPhraseBackTranslate.cy.tsx	Adds CT coverage that only this step’s language take is fetched/played.
src/renderer/cypress/support/pbtHarness.tsx	Extends the PBT harness to seed per-language takes/step language and record fileurl fetch IDs for assertions.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.