fix(selfhost): skip dimension-mismatched vectors in the SQLite vectorize adapter instead of silently truncating

[JSONbored]

Summary

RAG-audit hardening from epic #8757. cosineSimilarity's Math.min(a.length, b.length) silently scored mixed-width vectors over the shorter prefix — wrong similarity results if an operator ever swaps embedding models without a full reindex. Qdrant hard-fails this at the collection level and pgvector raises; the SQLite adapter was the only backend that silently computed garbage.

Changes

• src/selfhost/vectorize.ts — the query path skips any stored vector whose width differs from the query's (degrade-to-fewer-candidates, the RAG pipeline's established fail-safe posture) and emits one sqlite_vectorize_dimension_mismatch WARN per query (not per row) naming the query width and the deduped mismatched widths.
• cosineSimilarity itself is untouched (pure function, existing behavior pinned by its own tests).

Test plan

• npm run typecheck — clean
• 3 new tests: mismatch skipped + single warn with both widths; steady-state silence; width dedupe with sorted output and score ordering preserved — 10/10 unit + 15/15 contract suite
• Changed range: 0 uncovered statements/branches

Closes #8766

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	loopover-ui	c1572db	Commit Preview URL Branch Preview URL	Jul 26 2026, 06:00 AM

[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 92.26%. Comparing base (f4c946b) to head (c1572db).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #8792      +/-   ##
==========================================
+ Coverage   90.56%   92.26%   +1.70%     
==========================================
  Files          96      797     +701     
  Lines       22490    79563   +57073     
  Branches     3884    24111   +20227     
==========================================
+ Hits        20367    73408   +53041     
- Misses       1945     5090    +3145     
- Partials      178     1065     +887     

Flag	Coverage Δ
backend	92.93% <100.00%> (?)

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

Files with missing lines	Coverage Δ
src/selfhost/vectorize.ts	100.00% <100.00%> (ø)

... and 700 files with indirect coverage changes

[loopover-orb]

Tip

✅ LoopOver review result - approve/merge recommended

_{Review updated: 2026-07-26 06:09:57 UTC}

2 files · 1 AI reviewer · no blockers · readiness 100/100 · CI green · clean

✅ Suggested Action - Approve/Merge

• safe to merge

Review summary
The AI review returned non-blocking notes for this change but did not include a separate narrative summary. Review the nits below before deciding this PR.

Nits — 5 non-blocking

• src/selfhost/vectorize.ts:66-84 — the inline comment block is quite long (7 lines) for a single guard; consider trimming it since the PR description already carries most of this rationale.
• The external brief flags `console.warn` at vectorize.ts:83 as a 'debug leftover', but this is clearly an intentional structured operational log (JSON with an `event` field), not debug output left behind by accident — worth noting the flag is a false positive.
• No corresponding dimension-mismatch guard/log was added to pg-vectorize.ts, so if that backend has the same Math.min-style truncation risk it remains silent there — worth confirming whether pgvector's own DB-level raise already covers this (per the PR description it does).
• Consider whether the WARN should also fire on `upsert` when a caller inserts a vector whose width doesn't match existing rows for the same namespace, giving earlier visibility than waiting for the next query.
• The `mismatchedWidths` Set could be seeded lazily only when at least one mismatch is found (already done) — no change needed, just confirms current design is fine.

Decision drivers

• ✅ Code review — No blockers (1 reviewer)
• ✅ Gate result — Passing (No configured blocker found.)

Context & advisory signals — never blocks the verdict

Signal	Result	Evidence
Linked issue	✅ Linked	#8766
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 (1 linked issue).
Validation posture	✅ 25/25	PR body includes validation/test evidence.
Contributor workload	✅ 10/10	Author activity: 13 registered-repo PR(s), 13 merged, 275 issue(s).
Contributor context	✅ Confirmed Gittensor contributor	JSONbored; Gittensor profile; 13 PR(s), 275 issue(s).
Improvement	✅ Minor	risk: clean · value: minor

Linked issue satisfaction

Addressed
The diff adds a width check that skips mismatched vectors instead of scoring them, emits a single deduplicated WARN per query naming both widths, and adds tests covering skip behavior, silent steady-state, and dedupe/ordering as requested.

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, MDX, Shell, Solidity, JavaScript
• Official Gittensor activity: 13 PR(s), 275 issue(s).
• PR-specific overlap: none found.

Contributor next steps

• Start here: Treat this as maintainer-lane context rather than normal contributor-lane activity.

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.

🧪 Chat with LoopOver

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

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

Full command reference: https://loopover.ai/docs/loopover-commands

🧪 Experimental — new and may change.

_{🟩 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 LoopOver, a quiet PR intelligence layer for OSS maintainers.

• Re-run LoopOver review

[loopover-orb]

LoopOver approves — the gate is satisfied and CI is green.