fix(ai-usage): record embeddings + self-host vision, fix advisory-routed model attribution

[JSONbored]

Summary

Phase B1 of the Grafana/AI-usage observability audit: the underlying ai_usage_events
data had real gaps that a dashboard consolidation (Phase B2, follow-up PR) would just
have visualized incorrectly otherwise.

• Embeddings were never recorded at all. src/review/rag.ts's inference calls
  (via src/review/adapters.ts's reviewInferenceAdapter) had zero ai_usage_events
  rows. Now every embedding call records under feature: "embeddings", using the real
  reported provider/model when available and falling back to the requested model id
  otherwise (e.g. the Workers AI binding, which reports no usage).
• Self-host/Ollama vision was never recorded. runVisualVisionForAdvisory and
  runScreenshotTableVisionForAdvisory only recorded their BYOK branches; the
  env.AI_VISION self-host branches recorded nothing. Both now record under
  visual_vision/screenshot_table_vision with a nullable providerKey distinguishing
  BYOK (byok:<provider>) from self-host (real reported model, or a static fallback
  label when the provider reports none).
• Advisory-routed calls had a stale/wrong model label. When ai-slop,
  ai-e2e-test-gen, linked-issue-satisfaction, and the two ai-summaries paths route
  through settings.advisoryAiRouting to env.AI_ADVISORY, the recorded model column
  was a hardcoded legacy Workers-AI id (or, for ai-summaries, always an empty string)
  instead of the real serving model. Fixed at the root in createOpenAiCompatibleAi
  (src/selfhost/ai.ts, from a prior commit already on this branch) so every caller now
  gets correct provider/model attribution; this PR updates the 4 consumer call sites to
  prefer that real value over their hardcoded fallback.

Test plan

• npm run typecheck — clean
• npm run test:ci — full local gate green (742+ test files / 14.7k+ tests, engine,
  workers, MCP, UI suites all pass)
• npm audit --audit-level=moderate — 0 vulnerabilities
• New/updated unit tests cover both branches of every changed conditional: real-model
  attribution vs. fallback-label, BYOK vs. self-host, success vs. no-usable-output,
  and a thrown-error/rethrow path for the new embeddings recording wrapper.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.35%. Comparing base (d3b9707) to head (44358d3).
⚠️ Report is 5 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #5247   +/-   ##
=======================================
  Coverage   94.34%   94.35%           
=======================================
  Files         473      473           
  Lines       39982    40017   +35     
  Branches    14576    14590   +14     
=======================================
+ Hits        37722    37757   +35     
  Misses       1585     1585           
  Partials      675      675           

Flag	Coverage Δ
shard-1	46.30% <32.78%> (-0.13%)	⬇️
shard-2	34.64% <47.54%> (+0.04%)	⬆️
shard-3	32.10% <39.34%> (-0.06%)	⬇️
shard-4	31.83% <19.67%> (-0.26%)	⬇️
shard-5	33.74% <26.22%> (+0.02%)	⬆️
shard-6	45.05% <4.91%> (+0.19%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/queue/processors.ts	95.63% <100.00%> (+0.01%)	⬆️
src/review/adapters.ts	96.42% <100.00%> (+1.98%)	⬆️
src/selfhost/ai.ts	98.12% <100.00%> (+0.04%)	⬆️
src/services/ai-e2e-test-gen.ts	100.00% <100.00%> (ø)
src/services/ai-slop.ts	93.42% <100.00%> (ø)
src/services/ai-summaries.ts	99.16% <100.00%> (+0.01%)	⬆️
src/services/linked-issue-satisfaction-run.ts	100.00% <100.00%> (ø)

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[loopover-orb]

Caution

🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

🛑 Gittensory review result - fixes required

_{Review updated: 2026-07-12 11:39:14 UTC}

15 files · 1 AI reviewer · 2 blockers · readiness 93/100 · CI failing · unstable

🛑 Suggested Action - Manual Review

• No linked issue detected — If this PR is intended to solve an issue, link it explicitly in the PR body.
• Maintainer requires a linked issue — Link the relevant issue (for example Closes #123) before opening the PR.

Review summary
This PR fixes three real gaps in ai_usage_events: embeddings were never recorded (now wired through a new reviewInferenceAdapter(env, ai) that records a usage row on both success and failure), self-host/Ollama vision calls were silently dropped (runSelfHostVisualVision now returns {text, usage} and both visual_vision/screenshot_table_vision recorders accept a nullable providerKey to distinguish byok:<provider> from a real/fallback self-host label), and four advisory-routed callers (ai-slop, ai-e2e-test-gen, linked-issue-satisfaction, both ai-summaries paths) now prefer usage?.model over their stale hardcoded/empty-string fallback. The wiring is traced correctly end-to-end (createOpenAiCompatibleAi now returns usage with provider/model on both embedding and chat paths via the new buildAiUsage helper), and each behavior change (real-model path, fallback-label path, error path) has a matching new test with concrete D1 row assertions. The one loose end is that the self-host branch in runScreenshotTableVisionForAdvisory records usage once immediately (with `null`/generic detail) before the shared `if (visionText)` parse step, mirroring the pre-existing BYOK branch's structure rather than introducing a new bug.

Blockers

• PR description: this PR does not close or clearly link an eligible open issue for the ai-usage observability audit, so I need you to add the maintainer-authorized issue reference or explain why this is maintainer-authored before merge.

Nits — 5 non-blocking

• The PR description doesn't link to a specific tracked issue number for this Phase B1 audit work — worth confirming this fix chain is maintainer-authorized/tracked even though it reads as legitimate internal follow-up rather than scope creep.
• codecov/patch failed at 93.33% vs the 99% target (per CI status) — worth identifying which changed branches (e.g. the `usage?.model ?? fallback` nullish arms, or an error path in the new reviewInferenceAdapter) are uncovered before merge.
• `SELF_HOST_VISION_MODEL_FALLBACK = "ollama:visual-vision"` is hardcoded in src/queue/processors.ts even though the binding could theoretically be a non-Ollama self-host provider; consider naming it more generically (e.g. `SELF_HOST_VISION_FALLBACK_LABEL`) since it's just a label, not a functional issue.
• In src/selfhost/ai.ts's chat `run`, `provider: opts.providerName` unconditionally overrides any `usage.provider` that `extractCliUsage` might already contain — reads correctly per the added comment, but worth a one-line note that this assumes extractCliUsage never surfaces its own provider field on an HTTP JSON response.
• Double check whether `reviewInferenceAdapter`'s `run` in src/review/adapters.ts is ever invoked for anything other than embedding batches from rag.ts — if any non-embedding call path exists, it would now be mis-recorded under the `embeddings` feature.

Why this is blocked

• No linked issue detected — If this PR is intended to solve an issue, link it explicitly in the PR body.
• Maintainer requires a linked issue — Link the relevant issue (for example Closes #123) before opening the PR.

CI checks failing

• codecov/patch — 93.33% of diff hit (target 99.00%)

Signal	Result	Evidence
Code review	❌ 2 blockers	1 reviewer
Linked issue	⚠️ Missing	No linked issue or no-issue rationale found.
Related work	✅ No active overlap found	No same-issue or scoped active PR overlap found.
Change scope	✅ 20/20	Low review scope from cached public metadata (no linked issue context).
Validation posture	✅ 25/25	PR body includes validation/test evidence.
Contributor workload	✅ 10/10	Author activity: 44 registered-repo PR(s), 36 merged, 470 issue(s).
Contributor context	✅ Confirmed Gittensor contributor	JSONbored; Gittensor profile; 44 PR(s), 470 issue(s).
Gate result	❌ Blocking	Repo-configured hard blocker found.
Improvement	✅ Minor	risk: clean · value: minor — Code changes are accompanied by test evidence.

Review context

• Author: JSONbored
• Role context: owner (maintainer lane)
• Public audience mode: oss maintainer
• Lane context: Repository is configured for direct PR review.
• Public profile languages: Python, TypeScript, Ruby, Go, JavaScript, MDX, Shell, Solidity
• Official Gittensor activity: 44 PR(s), 470 issue(s).
• PR-specific overlap: none found.

Contributor next steps

• Treat this as maintainer-lane context rather than normal contributor-lane activity.
• Explain no-issue PR.
• Link the issue being solved, or explicitly explain why this is a no-issue PR.

Signal definitions

• Related work = same linked issue, overlapping active PRs, or title/path similarity.
• Change scope = cached public metadata such as size labels, draft state, and review-burden hints.
• Validation posture = whether the PR provides enough public validation/test evidence for maintainer review.
• Contributor workload = public contributor activity and cleanup pressure, not a repo-wide quality failure.
• Contributor context = public GitHub/Gittensor identity context; non-Gittensor status is not a blocker.

[BETA] Chat with Gittensory

Ask Gittensory a question about this PR directly in a comment — grounded only in the same cached, public-safe facts shown above, never a new claim.

• @gittensory ask <question> answers contribution-quality Q&A with source citations and freshness.
• @gittensory chat <question> answers in natural prose from cached decision-pack facts via local inference (maintainer/collaborator; read-only).
• A plain-language @gittensory mention with a real question is routed to the closest matching read-only command automatically -- no exact syntax required.

Full command reference: https://gittensory.aethereal.dev/docs/gittensory-commands

_{🟩 Safe / merged · 🟦 Advisory · 🟨 Held for review · 🟥 Blocked / closed}

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

• Re-run Gittensory review