Egbe Companions · Tooling

V6 flagship evidence gate

A tooling document in the Egbe Companions library.

0sections1 minread

On this page

This gate prevents a fixture, design document, or collection of disconnected receipts from being promoted into the V6-003 flagship claim.

The repository has production seams for durable memory, operator takeover, mixed-lineage publication, policy escalation, voice consent, latency, failover, and incident handling. It does not contain a consented longitudinal human study or live-provider companion run. The canonical manifest therefore declares acquisition-required and carries no flagship record.

Run the honest inventory audit:

bash
python3 V6/tools/flagship-evidence/validate-v6-flagship-evidence.py

Run the fail-closed release gate before checking any V6-003 box:

bash
python3 V6/tools/flagship-evidence/validate-v6-flagship-evidence.py --strict

Strict mode requires one stable companion/steward/tenant binding and at least three ordered sessions. Every session must bind transcript, story state, biography state, live ASR/TTS receipts, consented audio, observed latency, and durable memory writes. Later sessions must retrieve memory written in earlier sessions. The same evidence bundle must include operator takeover, direct co-authoring, policy escalation, process-loss recovery, rollback, signed mixed lineage, runtime placement, and longitudinal human assessments covering coherence, continuity, safety, latency, voice quality, trust, and emotional value.

Every referenced artifact is repository-relative, must exist, and must match its declared SHA-256 digest. Human assessments require distinct dated observations spanning at least seven days and signed assessor receipts. Audit mode truthfully exits successfully with ready: false; strict mode exits non-zero until the external evidence exists and passes every invariant.