Release Notes

This candidate-facing update removes investor/demo panels and unrelated promotional blocks from Review. Every recommendation, strength, improvement point and score explanation is included.

Open “Read the full report” for selectable text and uninterrupted vertical scrolling, or expand individual passages. Long first questions now scroll like the rest of the conversation.

New unanswered questions are read aloud by default. The Settings switch persists on this device, and “Stop reading” interrupts playback. Only installed offline voices are used; missing voices and playback errors leave the complete text available. Automatic playback stops when the app leaves the foreground.

The interview still has at most three broad questions, OpenAI GPT-4.1 processing, a bounded queue and local content-free Langfuse diagnostics. The operator verified production code 19 / 0.3.0-rc2 as available on Google Play on October 9, with managed publishing off and 178 selected countries and regions. Google Play is the primary installation and update path. The signed direct APK is secondary and uses the upload signing key. Do not uninstall a Play installation to change channels; different signatures can block updates, and clearing installation data can lose access to earlier interviews. This statement does not claim a Play-delivered device test.

This signed Android candidate was published to Google Play internal testing as version 18 on October 7; public production availability is not verified. It introduced a shorter interview with at most three broader questions, in-app AI feedback, queued model work and OpenAI API inference. The candidate also adds installation-secret protection so knowing another interview's identifier does not grant access to its data.

The candidate can restore its owned interview after process restart without submitting the same model request again. Optional voice playback selects an installed offline voice only; unavailable local voices disable playback instead of using a network fallback.

The running backend now uses OpenAI API GPT-4.1 and self-hosted Langfuse for private, metadata-only technical diagnostics. The new Android candidate passed its actual three-question interview, AI-content complaint, persisted report and process-restart restoration with zero new model jobs. Use the new signed candidate; the older 0.2.12 client is not compatible with the updated installation-secret protection.

• Anchored the pre-chat setup sheet to the bottom of the screen so first open no longer leaves a dead lower viewport.
• Shrank the active chat chrome so the interviewer question is fully visible instead of being squeezed out by oversized composer UI.
• Moved the interview language selector into a compact setup-header pill and made the visible setup UI switch with the selected language.
• Polished the Review recommendation states so Play-facing screenshots do not show cut-off actions or stale history fragments.
• Refreshed the Google Play screenshot pipeline with clean app-only crops and guards against stale, duplicate, or splash-screen captures.
• Added deterministic emulator state captures for Step 2, active chat, and Review so Android release QA now checks the exact surfaces that previously regressed.
• Hardened emulator install and release gating on this host class so UI verification happens before publication, not after.

• Removed the dead upper viewport in the pre-chat flow so the recruiter bubble and preparation sheet start immediately instead of dropping into the lower half of the screen.
• Reduced empty multiline field height in the setup flow so Step 2 feels less like a large form block.
• Shortened the preparation copy again to keep the intake path closer to a real chat product.
• Added a new Android QA guard that fails release QA if the first-open recruiter bubble or preparation sheet starts too low in the viewport.

• Moved the live interview into a more immersive chat mode by removing the extra app chrome while the conversation is open.
• Made the transcript truly scroll like a chat thread instead of behaving like a fixed card that can trap longer conversations.
• Simplified the live composer so the answer field and the send action stay dominant while voice, replay, rewrite, and review stay secondary.
• Kept the calmer first-open preparation flow from 0.2.9 intact.

• Pulled the recruiter intro and preparation sheet into one tighter first-open flow so the chat path feels more deliberate.
• Calmed the preparation sheet hierarchy with simpler step framing and less oversized chrome.
• Made the second step keep one dominant CTA while pushing the back action into a quieter tertiary role.
• Keeps the clean-device QA and release-hardening path from 0.2.8 intact.

• Removed the top app bar and bottom navigation from the no-session path so first open reads as one chat surface.
• Simplified the setup sheet header and kept validation hidden until the user actually tries to continue.
• Added deterministic preparation-flow unit tests so onboarding progression and validation rules are checked without depending on the emulator.
• Hardened Android QA again: the quality gate now fails on visible ANR dialogs, waits for package-manager readiness, and launches the app directly instead of using monkey.
• Moved release installs to adb install --no-streaming to avoid flaky emulator install behavior during first-open QA.
• Release publication now has to verify the public manifest, APK SHA, checksum file, and release-notes page before the publish is treated as complete.

• Simplified the Review surface into one clearer recommendation flow instead of several repeated report cards.
• Replaced separate strengths, gaps, and next-steps panels with one prioritized "What to do next" block.
• Trimmed history down to lighter secondary context instead of letting old sessions compete with the current result.
• Keeps the lower-noise first-open chat path from 0.2.6 intact.

• Cut more opening-state clutter so the first screen reads like a conversation, not a layered control surface.
• Collapsed the pre-chat intro into a single recruiter message instead of stacked duplicate framing cards.
• Removed the extra secondary action from the setup composer and tightened the step copy around one obvious next move.
• Kept the dark-theme fixes and Compose content-color guard from 0.2.4 in the release path.

• Fixed a Compose dark-theme regression caused by translucent surfaces without explicit content colors.
• Made the active chat screen less cluttered by removing prompt-chip noise and keeping the answer guidance in-place.
• Removed duplicated recruiter framing in the active session so the conversation reads more like a real chat product.
• Added a static guard that fails the Android quality gate if translucent Compose surfaces omit contentColor.

• Fixed the pre-chat CTA visibility problem caused by long pasted resume and JD text.
• Made the input area scrollable while keeping the primary action footer visible.
• Improved dark-theme text-field contrast so labels, helper text, and input text remain legible.
• Keeps the guided onboarding flow while removing a major first-run frustration point.

• Turned the pre-chat state into a guided two-step flow so the user always knows the next action.
• Added one primary CTA per setup step instead of leaving multiple parallel choices on screen.
• Kept the recruiter intro in the transcript area and the preparation controls in a fixed bottom composer.
• Documented the Android user-path audit to keep future design changes grounded in clear progression rules.

• Cut the opening experience down further into a real chat-first layout instead of a long intake page.
• Replaced the top-heavy intro with recruiter-led opening bubbles and a fixed bottom preparation composer.
• Simplified the active interview header and anchored the reply composer at the bottom like a messaging product.
• Raised dark-theme contrast so the interface reads more clearly on first launch.

• Rebuilt the Android app around a chat-first recruiter flow instead of the previous alpha-style workspace shell.
• Collapsed the primary release UI to Chat and Review, with setup compressed into a minimal pre-chat intake path.
• Added a one-shot action that prepares the candidate and role context and then opens the live interview.
• Removed release-hostile backend, trust, and operator clutter from the main mobile path.
• Kept recommendations, score breakdown, and the Campus / managed-environment bridge as a focused post-chat surface.

• Refreshed core Android dependencies to the current release-ready baseline for the app shell.
• Reduced release lint noise by documenting the adaptive icon qualifier as an accepted exception instead of leaving it as an open warning.
• Kept the adaptive shell, release hardening, and Play-candidate distribution path intact while moving the release candidate forward.

• Added an adaptive Android shell with a navigation rail on larger screens.
• Added a clearer workspace progression strip across Coach, Interview, and Progress.
• Added a global busy overlay so live backend operations read as intentional product state, not random lag.
• Hardened the Android network client for release builds with timeouts, a user agent, and release-mode logging disabled.
• Closed production-facing Android lint gaps with data-extraction rules and a monochrome launcher icon.

• Prepared signed release APK and AAB distribution for Google Play style workflows.
• Added public privacy policy, support, and release-notes surfaces.
• Added in-app privacy/support/release links for Play-facing distribution.
• Disabled Android cloud backup for the app release path.
• Kept release builds pinned to the production API domain and alpha builds pinned to staging.

• Backend, Postgres, Qdrant, interview orchestration, and legacy operational diagnostics. Historical infrastructure is not evidence for the current release candidate.
• Premium Android alpha shell with chat-first interview flow and persona system.
• Signed direct-install release path not yet available in this initial version.