GLMCHAT V2.2.1 — NEW CHAT HANDOVER
Prepared: 29 July 2026 12:37:09 UTC

AUTHORITATIVE CONTINUATION BASELINE
Use the provider-aligned complete handover ZIP generated from this folder. It supersedes the seven-defect-fixed baseline for future GLMCHAT work.

COMPLETED LOCALLY
- Seven prior audit defects remain repaired.
- Default Together text/coding model aligned to zai-org/GLM-5.1 through forward migration 014.
- Historical checksummed migrations 001 through 013 remain byte-identical to the protected baseline.
- GLM hybrid reasoning payloads, preserved reasoning and streamed reasoning fields aligned.
- Text and vision pricing separated; current catalogue metadata and Kokoro pricing aligned.
- Online/offline settings schema extended for vision pricing.
- Service-worker cache generation advanced.
- Provider-specific regression tests and migration preservation checks added.

CURRENT VALIDATION
- Deterministic build: passed.
- Type-check: passed.
- Lint: passed.
- PHP static contracts: 66 passed.
- Node tests: 16 passed.
- Migration, rollback, privacy/reset and HTTP preflight contracts: passed.
- Package/PWA/security/SQLite validation: 295 checks passed before final handover record generation.
- Target-host runner: 15 passed; sqlite3, curl, mbstring and zip unavailable locally.

DO NOT CLAIM COMPLETE
The following remain external acceptance gates:
- compliant LiteSpeed/PHP host with all required extensions;
- live Together GLM-5.1, reasoning, streaming, cancellation, tools, Qwen vision and TTS;
- private production-data migration, failed-upgrade and rollback tests;
- physical Android Chrome, Samsung Internet, TalkBack and 200% text tests;
- formal React 16 and PIN/offline-credential risk decisions;
- load, contention, low-connectivity, storage-pressure and hostile-upload acceptance.

PROTECTED SCOPE
Preserve all existing features, user data, routes, storage structures, authentication, offline/PWA behaviour, deployment architecture and light mobile-first interface. Do not modify historical migrations 001 through 013.