Record live streaming and final reconciliation checks

This commit is contained in:
pewdiepie-archdaemon
2026-09-17 22:12:44 +00:00
parent f576406515
commit 1d8f73b9a0
+4
View File
@@ -151,6 +151,10 @@ Fixed fictional evidence diagnostic `reports/fixed-search-evidence-20260917T2154
The 22:01 live replay is terminal (four mechanically valid conversations, not four quality passes). Casual release lookup now receives CONTENT 1–5 instead of only 1–2; the evidence-preservation fix is exercised in production. Polished lookup quotes a date present in its retrieved release index and provides a link, but casual lookup still invents a different date not supported by its retrieved older-release pages. Both runs take roughly 15–16 seconds. Thus source preservation is validated; consistency, follow-up verification and claim grounding remain unresolved. No overall quality promotion.
Streaming repair `7cdd7886`: user session 6067439a-0e23-4c8c-8c1f-410f3b3acf94 exposed that search/citation requests deliberately buffered every text chunk until final completion. Removed this buffering while retaining canonical final replacement after completion checks. A failing-before/passing-after regression asserts first delta delivery before upstream completion; recovery tests now require visible drafts but clean canonical replacement. Runtime/routing and browser-rendering suites: 1,265 passed. Deployed on 7011. Live news replay `reports/clean-v3-search-quality-2026-09-17T22-11-07-276Z.json` emitted 100 text deltas, no runtime error, 38.7 seconds; it hit the round limit and appended the limit notice, so this validates streamed transport, not research quality or clean completed-answer reconciliation.
Completed-answer streaming replay `reports/clean-v3-search-quality-2026-09-17T22-12-10-396Z.json`: 36 text deltas followed by one canonical final response, no runtime error, 12.1 seconds. This exercises both progressive delivery and final reconciliation in the live UI request path.
1. Finish and manually audit all 16 conversations; inspect claim/source alignment, request completion, follow-up referents, and latency.
2. Distinguish provider emptiness from model query drift and unsupported synthesis. Do not label every weak answer a routing defect.
3. Preserve explicit user source constraints even when model queries omit them; do not infer official provenance from URL appearance.