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
@@ -51,43 +191,55 @@
What this demonstrates
Grok/xAI is named only to identify the API — no affiliation or endorsement is implied.
-
1 · Reference data
+
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.
-
2 · Model text — recorded fixture
+
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.
-
3 · Deterministic verification
+
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
+
+
Result — measured vs published
+
-
Quantity
This demo (Q1 only)
Published
Check
+
Quantity
This demo (Q1 only)
Published
Check
+
Orbital period
0.8373346 d
0.8374912 d
Δ = 1.6×10-4 d ✅
Mid-transit time
BKJD 131.579626
BKJD 131.574858 + k·P (KOI)
Δ = 7.0 min ✅
Transit depth
158.3 ppm
191 ppm
ratio 0.83 ✅
Transits / SNR
40 / 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)
+
Adversarial self-checks (challenge response)
+
-
Check
Result
Verdict
+
Check
Result
Verdict
+
Odd vs even transit depth
156.2 vs 160.2 ppm (0.42σ)
✅ pass
Secondary dip near phase 0.5
2.3 ppm (0.49σ)
✅ pass
Harmonic powers (P/2, P, 2P)
1378510.76 / 3127157.96 / 1608656.16 (×1012)
✅ pass
+
+
+
-
Model text (recorded fixture)
+
+
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.
@@ -107,8 +259,20 @@
Model text (recorded fixture)
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)
+
+
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.
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.
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
+
+
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.