G1 — Profound testing gate (HARD STOP)

Parent: S1–S13 · Gate · Gate bar

What this is

A dedicated validation milestone between Phase 1 (machinery proven) and Phase 2 (hardening). No Phase-2 issue starts until every box below is green. The purpose is to prove the whole machinery works at enterprise-level quality, security, and reliability before investing in per-stage perfection — and to record the baselines Phase-2 must beat.

Exit criteria (all required)

• End-to-end on ≥3 real messy corpora (mixed file types, scans, office docs, and deliberately dangerous files): Source→…→Export yields KG + RAG corpus + Quality Report with every contract satisfied and no crash.
• Security review: written threat model covering malicious files, prompt injection, PII leakage, secret handling, SSRF via Source URIs, and a dependency CVE scan; every finding closed or explicitly risk-accepted with rationale.
• Reliability: crash-mid-run resumes correctly; re-running a Delta is idempotent; add/update/Retract/Purge correctness proven on a fixture corpus with assertions on what survives.
• Performance baseline: throughput and peak memory recorded at 10k and 100k docs; documented ceiling and backpressure behavior; no OOM at the stated ceiling.
• Quality baseline: gold-set precision/recall/F1 per Stage recorded — the numbers Phase-2 must improve.
• Provider-swap: proven for every Capability with ≥2 providers, including one cloud/endpoint variant; outputs consistent modulo model.
• Determinism: identical input + config ⇒ identical Corpus Version fingerprint, verified across two clean runs.
• License audit: every shipped default Provider's weights+code verified permissive; any restricted model is opt-in with flow-down surfaced (ADR-0012).
• Defects: zero known Sev-1/Sev-2; core coverage threshold met.

Output

A signed-off gate/G1-RESULTS.md capturing every baseline number and the security/threat-model findings. Phase 2 references these baselines.

Blocked by

All of S1–S13.