66 full-scale
The trajectory matched Q2.4 through the original cumulative boundary.
EXP-003 · settled 2026-07-19
Five more updates.
The same boundary.
Deterministic backtracking reached 1/128 and preserved the 1.5% cumulative replay authority, then exhausted every scale before the first public checkpoint.
Valid no-go
Committed71 updates
Smallest scale1 / 128
Public checkpoints0
Decision
Scalar continuation did not reopen the learning path.
Attempts 1–66 committed at full scale. Five more committed only after backtracking. Attempts 72–79 then exhausted all eight registered scales, triggering the frozen stop rule at 71 committed updates.
The maximum committed replay increase was 1.499444%, below the unchanged 1.5% authority. No rejected state was admitted. Because the run never reached 100 commits, quantity/public feasibility and promotion were not evaluated; replication seeds 1 and 3 stayed sealed.
Observed trajectory
Safe by construction. Insufficient by result.
5 backtracked
Geometric retries extended the safe path, including one commit at the minimum 1/128 scale.
8 exhausted
Every registered scale failed on eight consecutive outer attempts, ending the phase.
Recorded metrics
| Metric | Value |
|---|---|
| Outer attempts / candidate trials | 79 / 156 |
| Full-scale / backtracked commits | 66 / 5 |
| Warning / hard trial exceedances | 96 / 85 |
| Maximum trial replay increase | 1.605636% |
| Maximum committed replay increase | 1.499444% |
| Public / promotion evaluations | 0 / 0 |
Forecasts settled to no-go
| Forecaster | Frozen go / no-go / failure | Brier |
|---|---|---|
| Mechanistic backtracking model | 18% / 80% / 2% | 0.0728 |
| Q2.3/Q2.4 trace model | 8% / 89% / 3% | 0.0194 |
Both independent forecasters assigned their largest probability to the observed outcome before execution.
16 objects
All content-addressed ledger objects verified.
17 events
The complete contribution-to-settlement chain verified.
7 source hashes
Contract, results, logs, checker, and driver match the run attestation.
Evidence
Exact commits, exact artifacts, one live run.
Execution used the frozen Zero implementation at 616220e2. The observed artifacts are published without source changes at result commit 16b1fc9f. ilXyr registered the experiment prospectively at 392aaac2 and resolved run f4cf8ffe… as a valid no-go.
Read the machine-readable evidence →
Inspect the upstream result →