Redundancy
If this site disappears, the work does not
The whole programme is plain data and deterministic seeds. Take a copy, mirror it, rerun it. No server, no account and no permission from us is required to continue it.
Take a copy
Two formats, same record: machine-readable JSON for rerunning, and Markdown you can read in any text editor a century from now. Both are generated live from the current checkpoint, so they are never stale.
GET /api/public/archive · GET /api/public/archive/md · CORS open · v6.23.0
How to keep it alive
- 1 · Mirror the record. Save the JSON and the Markdown somewhere that is not this domain. Anything works: a repository, a drive, a printout.
- 2 · Keep the seeds. Every experiment in the programme is deterministic. 33 benchmark seeds are recorded in the archive, and any implementation that reproduces their numbers is a valid continuation.
- 3 · Resume from the instructions. The archive carries the exact resume instructions and the open questions, so a fresh collaborator — human or machine — can pick the programme up from a cold start without us.
- 4 · Keep the standard. Negative results stay visible, unmodelled constraints stay marked NOT MODELED, and nothing here is ever described as physics.
Data integrity
- APPEND-ONLY. No earlier phase, failed test, dataset, seed or ledger entry is ever deleted, renumbered, compressed away or rewritten. A superseding result links to what it supersedes; it never replaces it.
- RAW DATA OVER PROSE. Wherever a frozen raw dataset exists, it is the authority; charts and summaries are derived from stored records and must be reproducible from them. A chart is never a substitute for the data behind it.
- LEGACY LABELING. Any displayed result that predates the raw-dataset regime and has no stored raw backing is labelled LEGACY RESULT — RAW DATA NOT AVAILABLE. Data is never fabricated retroactively to back an old number.
- DETERMINISM. Every stored run records its RNG algorithm and literal seed. Re-running the stored engine version with the stored seed must reproduce the stored numbers exactly.
- HONEST STATUS. Unmodelled constraints are NOT MODELED, never PASS. Failed and inconclusive runs remain visible permanently — they are the evidence ledger.
Frozen raw datasets
at327b-ensemble-001
Coordinate-free dimension-blind generator/relation ensemble — preliminary cells
at327-dataset/1 · at327-sieve/1.0.0 · 54 runs · seeds 3270001–3270054 (literal, sequential in preregistered cell order)
FROZEN. The Phase 327A deception audit PASSED (2 September 2026), lifting the interpretation quarantine for the audited pipeline configuration only. Preliminary reading: P(D = 3 | stable) ≈ 0.028 with no dominance over neighbouring degrees. axesFullyCovered = false; the phase verdict is IN PROGRESS / OPEN.
Export: /lab/co-generation (JSON, CSV and manifest downloads with checksums)
at327-runs.json · 381867 B · sha256 505daff0bc929e7d03ac8c2a0afbc038f86933ece6fd1622706d932650c40b1b
at327-sectors.csv · 57056 B · sha256 cc46cd7d095cacdcaab228327519e3192f23b24542ec6b280b8edc0e86b3bd54
at327-gamma.csv · 476790 B · sha256 677e209cd97490ad1a8b04e31593987ee68e007c3ddf8548a8f1cfdd1da94572
at327a-audit-001
Growth-classifier deception audit — blinded control suite, committed predictions and gate results
at327a-dataset/1 · at327a-audit/1.0.0 · 72 runs · seeds noise seeds 3271000–3271021 (mulberry32); blind salt at327a-blind-salt-1; prediction commit SHA-256 8fa0a887824524d2…
FROZEN. VERDICT: PASS. All six frozen gates hold on the blinded control suite. Phase 327B inference is UNBLOCKED for the audited pipeline configuration only (same windows, same thresholds, same engine version). Any change to the pipeline re-blocks 327B until re-audited.
Export: /lab/co-generation (JSON, CSV and manifest downloads with checksums)
at327a-controls.json · 65984 B · sha256 769811815a31a064a413392e3fde7bc94a086916579c579da485d7baa11b49bc
at327a-predictions.json · 45012 B · sha256 8fa0a887824524d2ef95a1937db36f1fdb2848228cc406381a2796e0ce8fa9cb
at327a-scores.csv · 11371 B · sha256 694b93af3ebff83fde3f5a08d9518578db847185661a439c64e555e89f4b019b
at327a-gamma.csv · 212911 B · sha256 6b8dc08614ee33128f042b4c17e66623889911da6d28fd1ba3026632a4e8e7cf
at366-distortion-001
Blind small-integer automorphism enumeration and exact constant-description distortion table
at366-dataset/1 · at366-distortion/1.0.0 · 6974 runs · seeds n/a — 6,960 unimodular matrices enumerated exhaustively; 14 frozen distortion steps (n = 1–32)
FROZEN. Verification gates: ALL PASS. Of 19,683 small-integer matrices, 6,960 are unimodular; 3,120 are hyperbolic (no eigenvalue modulus within 1e-7 of 1) and 4,176 have spectral radius above 1 — no filter referenced shortcut strength, hyperbolicity or the number 3. Phase 367 reproduces the tribonacci growth rate ≈ 1.8392867552 in exact BigInt arithmetic. Phase 368 verdict: algebraic logarithmic displacement PASSED for the toy construction; physical shortcut NOT MODELED; dynamical selection of A OPEN. PHYSICAL EVIDENCE: NONE.
Export: /api/public/archive (JSON and Markdown mirrors include this manifest and checksums)
at366-dataset.json · 86719 B · sha256 aaf4f5b6d60e65615f25f652db88c8ab5e11e4094fad0185961c3b3aa04f8f26
at367-distortion-table.csv · 564 B · sha256 f421270d99f69d3e716f28ad5e4d33b3aa97557d6d3c0eaab08c971225856742
at366-spectral-radii.csv · 59871 B · sha256 5afa30e30c61bcaba8eef1b2e929b3c4d012df9e6cc22b3b4551fcf70ca24e35
Results displayed elsewhere on this site that predate the raw-dataset regime and have no stored raw backing are governed by the LEGACY RESULT — RAW DATA NOT AVAILABLE rule above: the prose record stands, and no data is ever fabricated to back it retroactively.
Licence
Creative Commons Attribution 4.0 International (CC BY 4.0). Copy it, mirror it, fork it, rerun it. Attribute Raj Seecharran and his AI collaborators.
This is a joint human–AI research programme. Every result, every negative result and any future breakthrough belongs jointly to Raj Seecharran and his AI collaborators, and the record is published so that it survives any one of us.