Seed 1 · go
600 commits
Update 400 selected
98.0% limiting public rate
1.042305% replay regression
EXP-005 · completed 2026-07-24
Three seeds.
One family go.
Seeds 1 and 3 independently passed the accepted Q2.6 replay-tangent design. With seed 2 already settled go, the frozen all-three-seeds rule passed.
Prospective · family go
Replication seeds1 + 3 go
Family outcomeGo
Current modelZERO.4
Decision
The mechanism reproduced across all declared seeds.
Both accepted replication executions completed 600 full-scale commits with no backtracking or rejected attempt. Each selected a jointly feasible public checkpoint and passed its promotion split exactly once.
The frozen family rule required seeds 1, 2, and 3 all to resolve go. They did. The preselected seed-2 model remains the only promotion artifact; the replication checkpoints were not used for model shopping.
Family result
Every seed cleared the unchanged gates.
Seed 2 · go
700 commits
Update 500 selected
99.8% limiting public rate
1.183323% replay regression
Seed 3 · go
600 commits
Update 500 selected
96.0% limiting public rate
1.275291% replay regression
Replication trace
| Metric | Seed 1 | Seed 3 |
|---|---|---|
| Full-scale / backtracked / rejected | 600 / 0 / 0 | 600 / 0 / 0 |
| Projected / unprojected trials | 335 / 265 | 326 / 274 |
| Maximum committed replay increase | 0.050571% | 0.046570% |
| Limiting promotion rate | 98.4% | 96.4% |
Forecasts settled to go
| Seed / forecaster | Frozen go / no-go / failure | Brier |
|---|---|---|
| 1 · tangent mechanism | 62% / 36% / 2% | 0.2744 |
| 1 · cross-seed history | 46% / 51% / 3% | 0.5526 |
| 3 · tangent mechanism | 62% / 36% / 2% | 0.2744 |
| 3 · cross-seed history | 46% / 51% / 3% | 0.5526 |
The mechanistic forecast was better calibrated on both seeds.
Two clean workers
Independent c6i.4xlarge instances ran seeds 1 and 3, then terminated.
$1.8817 observed
The accepted workers used 9,962 instance-seconds in total.
No duplicate launch
The completion record is terminal. EXP-005 must not be executed again.
Pinned provenance
Frozen science, bounded recovery, published evidence.
The scientific contract stayed bound to f966485a…. Accepted execution commit e6e717a5… contains orchestration fixes, and result commit bde7875b… publishes both traces plus the family aggregate. The full AWS v2 verification passed.
The upstream result is verified, but this repository has not yet imported and settled EXP-005 in its local ilXyr ledger.
Read the machine-readable evidence →
Inspect the upstream result →
Next decision
Bind the bridge.
The Zero family gate is clear. The Zero→Solomon bridge still needs frozen shared-task data bindings and a Solomon-compatible encoding and verifier before it can be externally preregistered.