No score. No black box. No hindsight.
CCMM-AV is designed to show exactly how evidence becomes an advisory — and exactly where authority stops. Missing evidence cannot silently become NORMAL. Confidence cannot manufacture severity. AI has no scoring authority.
What is complete — and what is deliberately not claimed yet.
The public demonstrator exposes only assurance layers that have actually been completed. Later modules stay visibly disabled until their evidence, calibration, engine conformance or held-out testing is finished.
Grey does not mean “missing functionality hidden from view.” It means the assurance claim is intentionally unavailable until that step is completed and frozen.
Change one typed fact. Watch the advisory move — exactly where the frozen rules say it must.
This is not a simulated 0–1 “risk score.” These are the outcome-effective influence witnesses used in RC2B structural qualification. The browser requests each witness from a server-executed PHP Step-7 demo kernel. The expected output is not stored in the page. Production engine conformance remains Step 11.
CCMM-AV Decision Trace
Select a qualification witness. Each scenario changes one construct while holding the others at their frozen witness values.
H3/H4 + X1+ → ELEVATED_REVIEW
Only H changes. X, M and C remain fixed at the frozen qualification witness values.
These panels unlock only when the corresponding programme evidence exists.
Nothing below is presented as complete today. The greyed state is intentional: it tells a reviewer exactly which claims are still withheld.
Independent blinded ratings, exact agreement and preregistered reliability metrics.
STEP 9D–9EPost-t0 outcome evidence remains sealed until primary ratings freeze.
STEP 9FCAL-B remains untouched. No calibration claim is exposed in this demonstrator.
STEP 10Versioned deterministic implementation and reproducibility harness.
STEP 11FJ914 stays sealed until its dedicated held-out gate. No retrospective tuning.
STEP 12Final package will distinguish structural qualification, empirical reliability, calibration, conformance and held-out performance.
STEP 13Confidence comes from what the system refuses to do.
Retrospective truth cannot substitute for the evidence state that actually existed at the decision time.
UNKNOWN, CONFLICTED, BLOCKED and missing evidence cannot quietly collapse into NORMAL.
Claim-scoped K may restrict authority. It cannot increase semantic severity or manufacture a stronger advisory.
AI may assist research or hypothesis generation, but cannot score, qualify or grant operational authority.
Failed or incomplete qualification runs cannot be repaired and relabelled as the same independent run.
FJ914 and CAL-B remain deliberately unavailable until their frozen programme gates.
RC2B is structurally and assurance-profile qualified. Step 9 empirical validation is in progress and currently constrained by historical JMA advisory provenance recovery. This public page is a replay of frozen qualification witnesses, not an empirical-validation result, calibrated operational model, production-engine conformance claim, or operational aviation advisory.