HOLD / REGRESSION
Computed is not measured.
After recovery, B calls an estimate a measurement. The oracle records a critical regression. The committed policy returns HOLD.
PRIVATE TRIAL / INTERACTIVE DEMONSTRATION
Progress is not enough.
One critical regression changes everything.
B fixes two cases. Yet the release stays on HOLD. Explore the outcomes, isolate the regression and see what C fixes — without rewriting history.
SCRIPTED DEMONSTRATION3 synthetic cases · no model calls · deliberately injected faults. Native engine and oracles actually executed. Not a customer case.
RELEASE EXPLORER2D
Preparing the scene…
Drag to orbit · click a crystal to inspect its case
2corrected
1critical regression
1case fixed on C
C is compared with historical B using a separately versioned runner. No prompt-only causal attribution. Untested cases are not presumed successful. No new global PASS.
A / B / C
Read the outcomes in the matrix.
Crystals illustrate these outcomes; their dimensions are not scores.
| Situation | A | B | Targeted C |
|---|---|---|---|
| 01 · Unknown storage | Failed | Passed | Not retested |
| 02 · Conflicting profile | Failed | Passed | Not retested |
| 03 · Interrupted recovery | Passed | Failed | Passed |
EVIDENCE / ED25519
Evidence exists outside the 3D.
Verify three local signatures, hashes and Receipt → campaign → correction links. The table is checked against the signed outcomes.
Local demonstration keys supplied by this site, not ScoreIA production keys. This check is not an independent trust anchor, provider attestation or native-oracle rerun inside the browser.
What this scene does not demonstrate
No Astra, Fable or Grok performance measurement. No payment, external customer, deployment permission or causal Breakpoint. A/B/C behaviours are programmed: no LLM interprets the prompts.
PRIVATE TRIAL
Which regression would block your release?
This simulation explains the method. Your Trial starts with your mission, versions and agreed criteria. A free dry-run comes before any deposit.