M2 Workflow Embedding Checkpoint

A small clean-before-embed run completed the jobs-research workflow vector family.

Ali Mehdi Mukadam - co-authored with Codex - 2026-06-15

What changed

The workflow family moved through the clean-before-embed ladder: repaired corpus confirmation, deterministic semantic QA, dry-run candidate selection, small live Vertex ADC embedding, AIN-510, vector reconciliation, source-authority registry refresh, exposure scan, runtime readiness, AIN-506, and full validation.

1. Confirmed the repaired corpus exists for jobs_research_workflow.
2. Ran deterministic semantic QA over 50 workflow rows.
3. Dry-ran Gemini candidate selection with repaired chunks included.
4. Embedded the 56 missing workflow vectors.
5. Refreshed the receipts that make the vector snapshot authoritative locally.

Live embedding scope

The live call used Gemini Embedding 2 at 768 dimensions through Vertex ADC on project aina-495702. No Developer API key route was used.

Semantic QA proof

jobs_research_workflow semantic QA passed with 50 sampled rows, 50 passes, no failures, no raw JD key hits, and no legacy review gate hits. The adjacent jobs_research_tool plus workflow_tool_evidence family also passed QA; the dry run found no new candidates because all 83 chunks already had accepted vectors.

Locked boundaries

The run did not change runtime or release posture. Public runtime, real-user data, external writes, production telemetry, runtime embedding authority, batch promotion, and donor repo mutation all remain off.

Verification

uv run aina-data-engine --root /srv/aina/aina-data-engine-room production-embedding-semantic-qa --source-family jobs_research_tool --source-family workflow_tool_evidence --limit 50
uv run aina-data-engine --root /srv/aina/aina-data-engine-room production-embedding-semantic-qa --source-family jobs_research_workflow --limit 50
uv run aina-data-engine --root /srv/aina/aina-data-engine-room gemini-embedding-run --source-family jobs_research_workflow --include-repaired --max-new 100 --selection-mode progressive --workers 4 --timeout-seconds 60 --max-retries 4 --write-every 25 --allow-live-gemini --confirm-paid-api
uv run aina-data-engine --root /srv/aina/aina-data-engine-room ain-510-retrieval-promotion-gate
uv run aina-data-engine --root /srv/aina/aina-data-engine-room production-chunk-vector-reconciliation
uv run aina-data-engine --root /srv/aina/aina-data-engine-room source-authority-registry-v2
uv run aina-data-engine --root /srv/aina/aina-data-engine-room docs-frontmatter-check
uv run aina-data-engine --root /srv/aina/aina-data-engine-room artifact-exposure-scan
uv run aina-data-engine --root /srv/aina/aina-data-engine-room production-runtime-readiness
uv run aina-data-engine --root /srv/aina/aina-data-engine-room ain-506-p0-gate
uv run aina-data-engine --root /srv/aina/aina-data-engine-room validate

All passed after rerunning vector reconciliation serially. A first reconciliation attempt was run in parallel with AIN-510 and read the old vector count; the serialized rerun is the durable receipt.

Next work

Continue M2 with another bounded source family. Recommended order: workflow_seed / workflow_intelligence, then jobs_research_responsibility, and only then serviceable_title after stronger source-authority repair proof.

cd /srv/aina/aina-data-engine-room
git status --short --branch
git log -3 --oneline
uv run aina-data-engine --root /srv/aina/aina-data-engine-room validate
uv run aina-data-engine --root /srv/aina/aina-data-engine-room ain-510-retrieval-promotion-gate
uv run aina-data-engine --root /srv/aina/aina-data-engine-room production-chunk-vector-reconciliation