| name | evidence-verification-pipeline |
| description | Canonical entry point for the full Gaia evidence verification pipeline. Invoke as /evidence-verification-pipeline or /ev-pipeline — both names trigger this skill. Use whenever you need to run all four evidence phases end-to-end: collecting raw evidence from the data lake, verifying live GitHub star counts, running adversarial auditing for noise and URL errors, and checking link health via Firecrawl. Also use when someone says "run the evidence pipeline", "verify the evidence lake", "audit evidence", "run ev-pipeline", "full evidence check", "pipeline run", "prepare evidence for ingestion", or "refresh the data lake". This is the pre-ingestion step — run it before importing any evidence into the registry.
|
Evidence Verification Pipeline
This is the canonical orchestrator for the Gaia evidence data lake. Both /ev-pipeline and /evidence-verification-pipeline invoke this same skill — they are aliases of each other with identical behavior.
When to run this: Before any evidence gets imported into the registry. The pipeline validates the raw evidence/ lake — it does not touch registry/nodes/ or any canonical registry data. Running it late (after ingestion) is too late to catch URL errors or star-count drift.
graph TD
A[Phase 1: ev-collection] -->|Compile Index| B[Phase 2: ev-star-verification]
B -->|Partition Tiers| C[Phase 3: ev-adversarial-audit]
C -->|Audit Contexts| D[Phase 4: ev-link-validation]
D -->|Validate Statuses| E[Master Source Report & Visual HTML Updates]
The Four Phases
Run these in order. Each phase produces output that the next phase consumes — skipping a phase produces incomplete or stale results downstream.
Phase 1: Evidence Collection (ev-collection)
Aggregates raw evidence from evidence/collectors/ and compiles the master index. This is the foundation — without it, Phases 2–4 operate on stale or missing data.
/ev-collection
Phase 2: Live Star Verification (ev-star-verification)
Queries the GitHub API for current stargazer counts, validates them against registry/named/ Markdown files, and partitions the data lake into tiers. Star counts drift constantly; this phase ensures the evidence lake reflects real signal, not cached values.
/ev-star-verification
Phase 3: Adversarial Audit (ev-adversarial-audit)
Deploys parallel adversarial reviewer agents to scan the data lake for evaluative noise, URL format errors (e.g. tree/ vs blob/), and proxy mismatches. Findings are appended to the daily source report. This phase exists because automated collectors make systematic mistakes — an adversarial pass catches patterns a single reviewer misses.
/ev-adversarial-audit
Phase 4: Link Validation (ev-link-validation)
Performs a live Firecrawl scrape verifying uptime and HTTP 200 status for all unique URLs in the data lake. Dead links inflate Trust Magnitude scores with phantom evidence; this phase surfaces them before ingestion locks them in.
/ev-link-validation
Post-Run Tasks
After all four phases complete, save the outputs so future pipeline runs and registry maintainers have an audit trail:
- Validation report: Write to
evidence/collectors/verification/firecrawl_validation_report_YYYY_MM_DD.md.
- Master source report: Write audit log, star updates, and adversarial findings to
evidence/source_report_YYYY_MM_DD.md.
- Visual dashboard: Update statistics and pipeline statuses in
evidence/verification_process.html.
Ingestion Handoff
For L4-approved intake rows, successful Phase 4 is the boundary between the
raw evidence lake and canonical registry mutation. Create a reviewed evidence
manifest from only live, correctly scoped rows, then hand it to
/gaia-ingest-batch. That wrapper uses /gaia-ingest for every CLI-only
gaia dev evidence write, appraises TM, and presents calibration proposals.
Do not import evidence by hand or treat requested intake stars as evidence.