SAPIENSLEDGER demo

SAPIENS · public validation demo

SAPIENS LEDGER

Kepler-10 b re-derivation — a public validation demo of the SAPIENS hash-chained evidence ledger.

Verify it yourself See the evidence ledger Source + data + tests ↗

What this demonstrates

Every step of a small scientific pipeline — data hash, hypothesis, analysis, adversarial challenge, verdict — is appended to a JSONL evidence ledger in which each entry SHA-256-links to the previous one. Anyone can re-run the pipeline and the verifier and check that the published record has not been tampered with. The hypothesis / adversarial-challenge steps sit behind an adapter interface whose optional live backend is Grok (xAI); this run used the deterministic offline mock adapter, so all model text below is a recorded fixture. Grok/xAI is named only to identify the API — no affiliation or endorsement is implied.

1Reference data

Real NASA Kepler Quarter-1 long-cadence light curve of Kepler-10 (KIC 11904151), 1,432 quality-filtered cadences over 33.5 days (MAST). Answer key: NASA Exoplanet Archive ephemeris for Kepler-10 b. One Kepler-10 c transit in the segment is masked and disclosed.

2Model text — recorded fixture

The hypothesis and adversarial-challenge entries in this run are grok-mock output: a deterministic offline stand-in seeded by the dataset hash. A live Grok adapter exists and is ready to run, but was not used here.

3Deterministic verification

Pure-stdlib box-least-squares transit search, the hash-chained ledger, and a verifier that recomputes every link and flags any modification. Reproducible with zero credentials and zero network.

Result — measured vs published

QuantityThis demo (Q1 only)PublishedCheck
Orbital period0.8373346 d0.8374912 dΔ = 1.6×10-4 d  ✅
Mid-transit timeBKJD 131.579626BKJD 131.574858 + k·P (KOI)Δ = 7.0 min  ✅
Transit depth158.3 ppm191 ppmratio 0.83  ✅
Transits / SNR40 / 34.0

Verdict: MATCH — validation (re-derivation), not a discovery. Honest caveat: a single 33.5-day quarter yields several near-equal periodogram peaks; the tabulated period is the global BLS maximum and sits 0.019% off the 17-quarter published value, inside the demo's stated tolerance (ΔP ≤ 5×10-4 d).

Adversarial self-checks (challenge response)

CheckResultVerdict
Odd vs even transit depth156.2 vs 160.2 ppm (0.42σ)✅ pass
Secondary dip near phase 0.52.3 ppm (0.49σ)✅ pass
Harmonic powers (P/2, P, 2P)1378510.76 / 3127157.96 / 1608656.16 (×1012)✅ pass

Model text (recorded fixture)

recorded fixture — grok-mock offline stand-in, not live model output

Hypothesis: The light curve contains a repeating, shallow, box-like dimming: a planetary-transit candidate. Expect a short period, a depth of tens-to-hundreds of ppm, and a duration of a few hours.

Adversarial challenges issued:

  • harmonic_confusion — Is the detected period the true period or a harmonic? Compare box power at P/2, P and 2P.
  • odd_even_depth — Do odd- and even-numbered transits differ in depth? A significant difference flags an eclipsing binary whose true period is 2P.
  • secondary_eclipse — Is there a secondary dip near phase 0.5? A deep secondary flags a self-luminous companion rather than a planet.

one-time live call — 2026-07-19T13:45:17Z, not part of the deterministic run

A single bounded proof call to the live Grok adapter (xAI API): HTTP 200, model grok-4.20-0309-non-reasoning, 274 tokens total, 0.74 s. Live Grok's one-sentence adversarial question:

“Would the claimed re-detection of Kepler-10 b still be statistically significant if the light-curve validation demo had instead been performed on an independent, equally long segment of Kepler data that was known a priori to contain no transits of that planet?”

Record: grok-live-probe.json (model, status, usage, sanitized reply only — no credentials). The committed ledger and tests remain on the deterministic offline mock; this was exactly one minimal proof call.

The ledger (all 6 entries)

Six append-only entries; each entry's SHA-256 folds in the previous entry's hash, so altering any byte anywhere upstream breaks every link downstream.

  1. SEQ 1data_ingestedsapiens-demo-pipeline3059a67bc532…
  2. SEQ 2hypothesisgrok-mock (offline stand-in)1d4ff4b976fb…
  3. SEQ 3analysissapiens-demo-pipeline792c7b9e1eda…
  4. SEQ 4adversarial_challengegrok-mock (offline stand-in)16a53942b6c7…
  5. SEQ 5challenge_responsesapiens-demo-pipeline3e7160992b6b…
  6. SEQ 6verdictsapiens-demo-pipeline5ab061ebf4b0…

Entry 1 links to a genesis previous_hash of 64 zeroes; the chain head below is entry 6.

Chain head (entry 6) SHA-256:

5ab061ebf4b097fb8106804cf9ef94bc29476b329ff070e7da8fadd7d6cc4291

# entry 1 — data_ingested — actor: sapiens-demo-pipeline
{
  "actor": "sapiens-demo-pipeline",
  "entry_hash": "3059a67bc5320e89d22701e33268dafad1a4bbcbc0767f651921d361f1ba41ca",
  "kind": "data_ingested",
  "payload": {
    "bytes": 32069,
    "file": "kepler10_kic11904151_q1_lc.csv",
    "masked_events": [
      {
        "name": "Kepler-10 c single transit (known planet, P=45.3 d, out of search scope)",
        "t_end_bkjd": 138.86,
        "t_start_bkjd": 138.54
      }
    ],
    "rows": 1432,
    "rows_masked": 16,
    "sha256": "1d82ef8ced447bb44ac80c7389a8ef3dde93bf1983d619d56fe2f6cda3a56ded",
    "source": "NASA Kepler long-cadence light curve, KIC 11904151 (Kepler-10), Quarter 1 (FITS header QUARTER=1), via MAST \u2014 provenance in data/README.md",
    "span_days": 33.471,
    "t_end_bkjd": 164.983391,
    "t_start_bkjd": 131.512315
  },
  "previous_hash": "0000000000000000000000000000000000000000000000000000000000000000",
  "seq": 1,
  "ts": "2026-07-19T13:11:56.018Z"
}
# entry 2 — hypothesis — actor: grok-mock (deterministic offline stand-in)
{
  "actor": "grok-mock (deterministic offline stand-in)",
  "entry_hash": "1d4ff4b976fbd79cbd44b44efda6f60deda6f33bfb7a012a7a683a0794998eb5",
  "kind": "hypothesis",
  "payload": {
    "adapter": "mock",
    "hypothesis_id": "mock-hyp-82bf1d5a",
    "predicted_depth_ppm_range": [
      20,
      2000
    ],
    "predicted_duration_hours_range": [
      0.5,
      8.0
    ],
    "predicted_period_range_days": [
      0.5,
      10.0
    ],
    "rationale": "Deterministic mock prior derived from dataset summary statistics (rows=1416, span_days=33.47); a live Grok adapter would reason over the same context.",
    "shape": "box (flat-bottomed transit)",
    "text": "The light curve contains a repeating, shallow, box-like dimming: a planetary-transit candidate. Expect a short period, a depth of tens-to-hundreds of ppm, and a duration of a few hours."
  },
  "previous_hash": "3059a67bc5320e89d22701e33268dafad1a4bbcbc0767f651921d361f1ba41ca",
  "seq": 2,
  "ts": "2026-07-19T13:11:56.019Z"
}
# entry 3 — analysis — actor: sapiens-demo-pipeline
{
  "actor": "sapiens-demo-pipeline",
  "entry_hash": "792c7b9e1eda389d11f7dea37f4a6887f9d20e51f8d76e744fc3f3bef67386e4",
  "kind": "analysis",
  "payload": {
    "bls_power": 3.127157961690592e-06,
    "bls_power_scaled_1e12": 3127157.962,
    "depth_ppm": 158.32,
    "duration_fraction_q": 0.08969,
    "epoch_bkjd": 131.579626,
    "grid": {
      "durations": [
        0.03,
        0.05,
        0.07,
        0.09,
        0.11
      ],
      "nfreq": 6000,
      "pmax": 10.0,
      "pmin": 0.5,
      "refined": true
    },
    "method": "running-median detrend + box-least-squares grid search (simplified Kovacs et al. 2002), pure stdlib",
    "n_transits": 40,
    "period_days": 0.8373346,
    "phase_center": 0.141036,
    "snr": 33.95,
    "std_out_ppm": 52.35
  },
  "previous_hash": "1d4ff4b976fbd79cbd44b44efda6f60deda6f33bfb7a012a7a683a0794998eb5",
  "seq": 3,
  "ts": "2026-07-19T13:12:18.015Z"
}
# entry 4 — adversarial_challenge — actor: grok-mock (deterministic offline stand-in)
{
  "actor": "grok-mock (deterministic offline stand-in)",
  "entry_hash": "16a53942b6c7426646f4c237ebc60dbda97540fa3ad9a8d173906859968ef80d",
  "kind": "adversarial_challenge",
  "payload": {
    "adapter": "mock",
    "challenges": [
      {
        "id": "harmonic_confusion",
        "question": "Is the detected period the true period or a harmonic? Compare box power at P/2, P and 2P.",
        "severity": "high",
        "suggested_check": "harmonic_powers"
      },
      {
        "id": "odd_even_depth",
        "question": "Do odd- and even-numbered transits differ in depth? A significant difference flags an eclipsing binary whose true period is 2P.",
        "severity": "high",
        "suggested_check": "odd_even_fold"
      },
      {
        "id": "secondary_eclipse",
        "question": "Is there a secondary dip near phase 0.5? A deep secondary flags a self-luminous companion rather than a planet.",
        "severity": "medium",
        "suggested_check": "secondary_fold"
      }
    ],
    "overall_skepticism": "Deterministic mock challenge set (standard false-positive screen for transit candidates); not a live model output."
  },
  "previous_hash": "792c7b9e1eda389d11f7dea37f4a6887f9d20e51f8d76e744fc3f3bef67386e4",
  "seq": 4,
  "ts": "2026-07-19T13:12:18.015Z"
}
# entry 5 — challenge_response — actor: sapiens-demo-pipeline
{
  "actor": "sapiens-demo-pipeline",
  "entry_hash": "3e7160992b6bf021d938ca94be1905491fa653a250a264450475c1fd3847fad8",
  "kind": "challenge_response",
  "payload": {
    "harmonic_confusion": {
      "interpretation": "power at 2P is expected to stay high (transits coincide modulo 2P); the informative comparison is P vs P/2, where transits split into two boxes",
      "pass": true,
      "powers": {
        "double": {
          "period_days": 1.6746692,
          "power_scaled_1e12": 1608656.164
        },
        "fundamental": {
          "period_days": 0.8373346,
          "power_scaled_1e12": 3127157.962
        },
        "half": {
          "period_days": 0.4186673,
          "power_scaled_1e12": 1378510.76
        }
      }
    },
    "odd_even_depth": {
      "delta_sigma": 0.421,
      "depth_even_ppm": 156.23,
      "depth_odd_ppm": 160.16,
      "pass": true
    },
    "secondary_eclipse": {
      "depth_ppm": 2.3,
      "pass": true,
      "sigma": 0.494
    }
  },
  "previous_hash": "16a53942b6c7426646f4c237ebc60dbda97540fa3ad9a8d173906859968ef80d",
  "seq": 5,
  "ts": "2026-07-19T13:12:18.019Z"
}
# entry 6 — verdict — actor: sapiens-demo-pipeline
{
  "actor": "sapiens-demo-pipeline",
  "entry_hash": "5ab061ebf4b097fb8106804cf9ef94bc29476b329ff070e7da8fadd7d6cc4291",
  "kind": "verdict",
  "payload": {
    "checks": {
      "adversarial_clean": true,
      "depth_consistent": true,
      "epoch_match": true,
      "period_match": true
    },
    "claim": "validation \u2014 re-derivation of a published result; NOT a new discovery",
    "deltas": {
      "depth_ratio": 0.829,
      "epoch_minutes_vs_published_ephemeris": 7.01,
      "period_days": 0.0001566011501311637
    },
    "match": true,
    "measured": {
      "depth_ppm": 158.32,
      "epoch_bkjd": 131.579626,
      "n_transits": 40,
      "period_days": 0.8373346,
      "snr": 33.95
    },
    "published": {
      "depth_ppm": 191.0,
      "duration_hours": 1.8076,
      "epoch_bjd": 2454964.574756,
      "name": "Kepler-10 b",
      "period_days": 0.8374912,
      "source": "NASA Exoplanet Archive Planetary Systems (ps) table via TAP (pl_orbper/pl_tranmid/pl_trandep/pl_trandur), retrieved 2026-07-19"
    },
    "target": "Kepler-10 b"
  },
  "previous_hash": "3e7160992b6bf021d938ca94be1905491fa653a250a264450475c1fd3847fad8",
  "seq": 6,
  "ts": "2026-07-19T13:12:18.019Z"
}

Verify it yourself

git clone https://github.com/thebeastagi/SAPIENS
cd SAPIENS
python demos/ledger-grok/run_demo.py        # ~25 s, stdlib only, offline
python demos/ledger-grok/verify_ledger.py   # recomputes the chain + data hash

The verifier recomputes every entry hash, checks each previous_hash link, and re-hashes the data file against the hash recorded in entry 1 — any tampering exits non-zero.

Artifacts: ledger.jsonl · results.json · SHA256SUMS · source + data + tests

Roadmap

  1. PHASE 0Shipped — standalone SAPIENS core, this LEDGER demo, tests + CI.
  2. PHASE 1Legal/licence gate for ASTRA-family reuse; trust-tiered adapter registry; signed ledger checkpoints.
  3. PHASE 2Validation framework v1 — holdout protocols, leakage controls, calibration before confidence.
  4. PHASE 3+Structured L3 review panels, real domain adapters (only after licence review), external-review workflows.

Full plan: ROADMAP.md. No discovery claims at any phase — the honesty invariant scientific_discoveries_claimed: 0 holds until independently validated results exist.

Credit

Developed by The Beast AGI engineering fleet in collaboration with the ASTRA research program (Prof. Glenn White) — attribution only; no ASTRA code is used. Kepler data: NASA/MAST. Ephemeris: NASA Exoplanet Archive. Demo generated 2026-07-19T13:12:18.019Z UTC. License: Apache-2.0.