feat: recover experimental Googlevideo media autonomously

Preserve operator-owned profiles while rechecking fresh media and bounded native relearning in the deployed daemon. Keep mixed safety failures fatal and expose active coverage separately from qualification. Document source-only acceptance and retain existing recovery budgets and privacy boundaries.

Co-authored-by: GPT-6 <gpt-6@openai.com>
This commit is contained in:
ZeusinaandGPT-6 Bot committed 2026-10-12 20:19:16 +03:00
1 parent 1a35cb3019
commit a60f62de22
14 files changed
+878 -41

No files matched your search

+6 -4
View File
@@ -6,18 +6,20 @@
### Googlevideo suffix RC — browser-verified experiment (2026-10-12)
- Source RC `eade051c2bc335b71b1616d697326ca4c8019d70`, immutable image `sha256:fdb60593e29d7c337a413a47aaddfbcaaeabdeb877042c58f506d4abdc98a13b`, isolated host SOCKS `127.0.0.1:11082`. Recorded journal: generation3, active candidates [1,2,3], scope `experimental_googlevideo_suffix`; `www.youtube.com` and `i.ytimg.com` profiles preserved. Older targets were unchanged. These are recorded field observations, not a new probe/deployment by this documentation commit; its SHA is not the tested image source.
- Initial source RC `eade051c2bc335b71b1616d697326ca4c8019d70`, immutable image `sha256:fdb60593e29d7c337a413a47aaddfbcaaeabdeb877042c58f506d4abdc98a13b`, isolated host SOCKS `127.0.0.1:11082`. Historical initial journal: generation3, active candidates [1,2,3], scope `experimental_googlevideo_suffix`; `www.youtube.com` and `i.ytimg.com` profiles preserved. Older targets were unchanged. These are recorded field observations, not current candidate IDs or a new probe/deployment by this source update; its SHA is not the tested image source.
- Fresh signed media on `rr16---sn-n8v7znsz.googlevideo.com` returned HTTP206, 65536 bytes, MP4 `ftyp`, TLS1.2. On **2026-10-12 the user confirmed browser videos work** with this live RC and accepted one shared profile for Googlevideo, including initial learning. Browser images also appeared, but that does not qualify every image route.
- One suffix profile is a **composite native circular A/E profile with two strategies and per-host learning**, not one uniform static strategy. Explicit operator experimental policy is durable across restart with hash/rollback guards; it is neither family-qualified nor `READY`, and no all-host cold 3/3 result is claimed.
- Service health already checks actual fresh image resources and media with both TLS protocols. Experimental recovery intentionally returns without automatically qualifying or replacing the operator-approved experiment: the remaining gap is autonomous experimental recovery/autoqualification, not missing probe implementation. Operator scope coverage and qualification are separate states.
- Source now wires autonomous experimental media recovery in either configured service mode. Healthy actual-resource probes short-circuit recovery; a validated media transfer failure permits one guarded restart of the same committed profiles and bounded native relearning in the persistent deployed daemon (10 attempts / 120s), followed by fresh signed media 3/3 on both TLS protocols. Learning stays in daemon RAM, not a temporary verification worker; expired inventory is refreshed, never rewritten or persisted. Missing inventory is degraded, not strategy absence; guard/infrastructure/cancellation and mixed failures remain fail-closed. No experimental candidate is replaced or promoted to qualified.
- Status distinguishes active-profile coverage (`covered=` / JSON `covered_groups`, `googlevideo_scope`) from qualified `groups` and `unresolved_groups`. Experimental Googlevideo remains unresolved for qualification even when operational media is healthy; a thumbnail profile does not mean the full asset inventory is qualified. Status itself performs no live health check. New recovery/classifier/status code is pending field acceptance; no new deployment, network test, family-qualified/`READY` or production claim is made. Generic production media handling, signed-URL/SS privacy, persisted minimum two-hour strategy-absence pause and the separate 90-minute whole-job budget are unchanged.
- Local source verification: fresh `go test ./... -count=1`, affected experimental/resource-classifier/status/monitor regression tests under `-race`, and `git diff --check` passed. These are unit/fixture checks only, not new field evidence.
Remaining acceptance gates (not asserted defects):
1. Representative fresh signed media evidence: at least three unique hosts across at least two roots/prefixes, post-warmup 3/3 for both TLS protocols plus actual media transfer. Do not impose uniform static/cold behavior on the accepted learning experiment or label it normally qualified without evidence.
2. Qualify the avatar route/profile: only the thumbnail image profile is currently active; browser-visible images do not close this gate.
3. Field-accept actual restart/recreation recovery after the all-Googlevideo suffix change, including renewed native learning; older exact-host lifecycle PASS does not cover this RC.
4. Resolve autonomous recovery/autoqualification policy for the approved experiment without silently replacing or promoting its operator-approved scope.
5. Reconcile service status with those results: distinguish operator-covered suffix from unresolved qualification/assets; no family-qualified/`READY` or blanket production claim yet.
4. Field-accept the new autonomous same-profile experimental recovery and resource-error classifier without replacing or promoting operator-approved scope. Automatic family qualification is not implemented.
5. Field-check the new covered-versus-qualified status fields against the live journal and asset inventory; no family-qualified/`READY` or blanket production claim yet.
### Earlier isolated-target acceptance (2026-10-11)