Data and code
The reproducible material for these papers is in github.com/openIE-dev/openie-research-experiments: preregistrations with their SHA-256, scripts, results, receipts, a checksum file for every file, and pinned references for large corpora (CORPORA.md). Code is Apache-2.0 and data is CC BY 4.0. Energy labels: est_j is an estimate, reported_j an operating-system or on-chip counter, measured_j a wall or shunt meter, empty in every result.
Thermodynamic Bounds and an Optimal Tempo for Goal-Holding Agents
| Result | Repository path | Contents |
|---|---|---|
| §8.11 preregistered acceptor run | agency-run/ | Preregistrations v1, v2 and second machine; agents, meters, tasks, analysis; episodes and receipts |
| §8.11 results, run 1 | agency-run/results/ | Episode rows and receipts |
| §8.11.1 and §8.11.2 results | agency-run/results2/ | Second run; the second machine is in results_jetson/ and results_jetson_cyc/ |
The Reality Code Engine
| Result | Repository path | Contents |
|---|---|---|
| §2.3 and §6.1 to §6.3 ladder: L0 and L1 held-out results, RAPL subsample | rce/bench/ladder/ | LADDER.md, PREREG.json with splits, RESULTS.md, report.py, per-task rows, results/jetson-rapl-2026-10-11 |
| §4.5 and §5.1 Form compiler, five-target differential test | rce/bench/forms/ | Harness and recorded traces; test in rce/tests/form_targets.rs |
| flowG cross-backend conformance (RCE §1, flowG) | flowg-conformance/ | Suite, fixtures by tier, recorded 18 of 18 report, CI job |
| §8.1 Defects4J | rce/bench/defects4j/ | Harness, TBar port comparison, REVIEW.md, results |
| §3.8 dimension runs dim-r1 and dim-r2 | rce/bench/dimension/ | PREREG.json with amendments and hash sidecar, pilot, RESULTS.md, per-program rows in results/ |
| §3.5, §3.7 and §12.4 surveys of 22, 23 and 24 October 2026 | rce/surveys/ | The three design reports, the database survey, the versioning survey (continuous improvement without versions), the new-entrant languages survey (111 languages of 2024 to 2026) and their reading notes |
RCE-JEPA
| Result | Repository path | Contents |
|---|---|---|
| W1 Form corpus v1 and v2, H1 (Form space against source space), the intent mutation score and the cost of a check | rce/bench/form-space/ | PREREG.json with amendments and hash sidecar, CARD.md (dataset card), RESULTS.md, results/forms.jsonl, results/negatives.jsonl, summary.json; generator scripts/corpus.py in openIE-dev/rce |
| H7 adversarial Forms, fuzzed arm: 560 accepted mutants against five invariants | rce/bench/adversarial/ | PREREG.json with hash sidecar, run.py, RESULTS.md, results/summary.json and violations.jsonl (empty) |
| H8 determinism: 10,500 streams on six targets | rce/bench/determinism/ | PREREG.json with hash sidecar, RESULTS.md with the workflow run, per-target jobs and limits; generator scripts/xarch-streams.py and the parity workflow in openIE-dev/rce |
Metered Commit Gates
| Result | Repository path | Contents |
|---|---|---|
| Part I and II | n/a | The gate and cascade use the WCA reference implementation described in the paper. Its board logs are cited in the paper's Appendix A. |
Economics companions: AI-Coding Market, Task Satiation, Spell Check
| Result | Repository path | Contents |
|---|---|---|
| Measurements cited | rce/ | The essay's measured rows come from RCE §6; the projections cite public sources in each paper's references. |