E005 — Policy Search milestone
Search means zero model calls
Automated policy search explores movement and firing schedules. No Qwen or other model is called during a Search run. Candidate policies can be discovered in accelerated research; milestone confirmation runs in real time.
Verified real-time milestones: 3,590 → 4,450 → 5,050.
The plateau was useful evidence
Final E005 registry: 2,002 games, comprising 1,996 accelerated research games and 6 real-time confirmations. In the earlier 1,980-run sample, 12 overnight batches did not exceed 5,050.
1,052 of 1,980 policies contained planned switches after the final death. Those particular switches could not affect the run. This does not imply that every affected policy had an entirely identical executed prefix.
- Run ID
pleiads-20261003-201906-9c3595- Experiment / controller
- E005 / native5
- Score
- 5,050 points
- Survival
- 183.239 s (emulated game time)
- Real-time / replay
- Confirmed / exact recorded-input replay
- Screenshot matches
- 1,898 / 1,898 captured samples
- Original manifest SHA-256
618aa85fd28807f78c37f13b466fbcd555e555a44e19e328b6223f4c26d0be23- Evidence files
- Original manifest · Ordered actions CSV · Replay comparison · Screenshot hashes
All captured samples were checked, at one capture every six emulated frames. Screenshot bytes and game software are excluded. CSV preserves the original TSV action order. This is a laboratory milestone, not a world-record claim.
Validated checkpoint: 2026-10-05. Data last synchronized: .
Latest record verified from reviewed published run evidence. Registry counts retain the last full dataset checkpoint; record publication does not refresh the complete dataset. This is a snapshot, not a live counter.
Registry count checkpoint: 2026-10-05T08:28:16.420013+00:00. The latest reviewed record and its evidence update separately from full dataset totals.
What we learned
The search plateaued at the frozen checkpoint. Scheduling a switch after the last death cannot change any executed input. Effective-policy canonicalisation can collapse those duplicate behaviours rather than counting them as meaningful new exploration.
This diagnostic motivated E006 — Diversity Search. It changes the search space instead of simply increasing the number of near-identical runs.
Evidence and media
The original run manifest, ordered action sequence, metrics and replay hash comparison are now ingested and downloadable. The public record above preserves the original manifest SHA-256. Images are excluded: the comparison publishes hashes for every captured sample, rather than game pixels.
The video publication candidate remains pending editorial and rights review. The existing 3:06 edit uses captured snapshots and research captions; it is not a claim of complete native audiovisual capture. This is a laboratory milestone, not a world-record claim.
The next verified milestone
