SCORECOMPUTE / EVIDENCE

Inspect the result.
Keep the limits.

A reproducible calculation is useful because you can question it. Models, failed research gates and verification boundaries belong beside the result.

Three things a receipt can tell you.

What was calculated

The supplied inputs, the selected engine and the returned scope describe a specific calculation. A compatible result is conditional on those inputs and assumptions.

Run a scoped check

Who issued these bytes

An Ed25519 receipt can be checked against the published key. Its signature protects the payload from alteration; it does not make observations true or provide an independent scientific audit.

Published verification keys

What was actually observed

Service availability includes sampling coverage. A kernel duration, a complete tool call and a browser round trip measure different intervals.

Recorded service observations

VERSIONED RESEARCH · INCLUDING REJECTED RESULTS

Results that did not earn promotion.

Method Forge · campaign 1

Benefit not demonstrated

A generated routing rule was evaluated on 48 held-out cases. Its predeclared mean-quality and time gates did not pass. The candidate was not promoted.

Read the campaign 1 study

Method Lab · campaign 2

Four gates passed. One did not.

The frozen rule matched the nearest-neighbor reference’s mean route length on 96 held-out cases. It did not deliver the required 1% mean improvement and was not promoted.

Read the campaign 2 study

Observation policy · v2

The complete gate still matters.

The study records a failed completion-and-equivalence gate despite other encouraging measurements. Its overall acceptance is false; no general benefit is claimed.

Read the policy outcome

These are internal, versioned experiments with declared references and separate evaluation cases. They are not external certification. The reference routing heuristic is not a guarantee of globally optimal routes.

Explore the routes and counterexamples

The atlas replays historical evidence; viewing it does not run a new scientific validation.

Boundaries you can check.

A result says	It does not establish
Arithmetic is consistent within its declared scope	That units, the model or the original factual claim are correct. Unsupported floating calculations remain inconclusive.
An observation fits the geometric model	Image authenticity, object identity or the absence of measurement bias.
An exact route search completed	Road conditions, travel time or precision beyond the supplied geometric distance model.
A client declared its name	An attested model identity. Software names can be supplied by callers.

Tool contracts · Data handling · Report a missing capability or a concrete problem

19 SCIENTIFIC ENGINES / SOURCE-BOUND EVIDENCE

A reference for each calculation.

These cards identify the reviewed sources, test references and limitations. The 0.14.0 release checks report 600 passing native tests and five explicitly ignored tests. Historical studies remain separate. No current per-engine hardware benchmark is claimed where a normalized measurement is absent.

Download all 19 evidence cards, source hashes and reproduction commands

simulate_pi · version identified by source hash

Monte Carlo estimate of π

Quarter-circle area π/4, known π, seeded sequence reproducibility and preservation under partitioning.

Limits, source and reproducible checks

A seeded finite Monte Carlo estimate, not an exact value of π.
The Wilson interval describes the sampling model, not a guarantee for each draw.
No current GPU speedup or combined-device throughput is established by this evidence inventory.

SHA-256: 99b263fac08d3432200a85dcf05b92209780ecdf69afbeaea53bb8c464b5db90

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server simulation::tests -- --test-threads=1

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server gpu_matches_cpu -- --ignored --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

optimize_tasks · version identified by source hash

Exact task selection

A bounded 0/1 knapsack example whose optimum beats greedy selection, plus a valid empty selection when no task fits.

Limits, source and reproducible checks

Independent tasks with integer durations and declared values; no precedence, shared-resource conflicts or uncertain durations in this engine.
Optimality applies to the supplied discrete selection model, not the realism of the values.

SHA-256: 3177d5dc10ce36fd14df19e188d075c0e8b9659f907c728fade2154cc3205f16

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server optimization::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

simulate_plan · version identified by source hash

Task-duration risk simulation

Zero uncertainty reduces the total duration, success rate and empirical percentile to a deterministic hand-calculated sum.

Limits, source and reproducible checks

Independent symmetric uniform duration variations are an explicit illustrative assumption.
The percentile is empirical; no real-world deadline calibration or general forecast guarantee is established by this limiting-case test.

SHA-256: 5631d8095efedcf272a21ec6a15df6425f8e5851f74f24af55436670dcbae6de

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server robustness::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

compute_orbit · 0.1.0

Ideal two-body orbit

Circular orbit radius and period, plus residual checks for Kepler’s equation at eccentricity 0.9.

Limits, source and reproducible checks

Ideal two-body ellipse around one solar mass, elapsed time from periapsis.
No perturbations, observed-body fitting or real ephemeris accuracy is claimed.

SHA-256: 452319bf9438a874c12d6f897a09ff49c84376b9925546b432ddf8aa8e9c09d0

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server astronomy::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

verify_shadows · version identified by source hash

Solar-shadow consistency

Skyfield 1.55/JPL DE421 solar directions at 37 predeclared 2026 samples, including the antimeridian; 0.05-degree test bound.

Limits, source and reproducible checks

Requires supplied image control points and a horizontal planar ground model; no automatic image interpretation.
Verdict thresholds are screening conventions, not an authenticity certificate.
Solar primitives are shared with ephemeris, tracks and golden-hour tools, so they are not independent corroborating sources.

SHA-256: abae56e3d8ef08a430ba19c80c77d0bca7d190ffda6e9b14ea366057bf6a2c13

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server soleil_ombres::tests -- --test-threads=1

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test solar_azimuth_contract -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

check_ephemeris · version identified by source hash

Claimed Sun or Moon position

Meeus sidereal-time and lunar-position examples, lunar phase and geometric topocentric parallax limiting cases.

Limits, source and reproducible checks

Approximate Sun/Moon models, standard refraction and stated observer assumptions; no object identification.
The extra angular margin is conventional, not a certified astronomical error bound.
Solar fixtures do not validate every lunar/date/location case.

SHA-256: 102dee0aac2c8bcbf21c1367719376bda7b59f41912a6a5f21d92095ded16e3a

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server ephemeride::tests -- --test-threads=1

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test solar_azimuth_contract -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

audit_statistics · version identified by source hash

GRIM/GRIMMER consistency screening

Integer mean-grid cases, sum-of-squares bounds, rounding-window enumeration, impossible zero standard deviation and sample-SD constraints.

Limits, source and reproducible checks

Necessary arithmetic filters for bounded integer observations; passing does not prove a dataset exists.
An inconsistency is not evidence of fraud or author intent.
Reported statistics must match the declared rounding and sample convention.

SHA-256: 23d2302b1cafa26b97ea604221fe2ac0f9727e32c3701842c9c2964a321f78f2

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server audit_stats::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

analyze_kinematics · version identified by source hash

Angular-speed kinematics

Radians-per-second times distance, with independently hand-calculated speed and limiting-distance anchors.

Limits, source and reproducible checks

Transverse speed depends on an assumed distance; the observation alone does not establish that distance.
Illustrative vehicle thresholds do not identify an object or its origin.

SHA-256: 7de6c8582aee6141335ad181881f852d991f2c0fb80d14f01665becc0c5e56f6

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server cinematique::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

check_shadow_track · version identified by source hash

Shadow-direction track consistency

Tracks through independent DE421 daylight directions, together with reversal, opposing residual and time-lapse regression cases.

Limits, source and reproducible checks

Consumes supplied minute/azimuth measurements, not video pixels.
Shares the solar model with the shadow and ephemeris engines.
A compatible track does not authenticate a video; thresholds and low-Sun limitations remain explicit.

SHA-256: 18068d3a1b7c32bc41626e52fee104925a35cc18ac5f50fa35fa4faef6251152

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server piste_ombre::tests -- --test-threads=1

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test solar_azimuth_contract -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

solve_equilibrium · version identified by source hash

Kuhn poker equilibrium

Kuhn poker value −1/18, probability normalization, qualitative equilibrium checks and pure best-response exploitability.

Limits, source and reproducible checks

Finite-iteration CFR+ for the specified tiny game, not a general poker solver.
Convergence checks do not imply a gambling advantage or advice.

SHA-256: d33dd25819b1219bc1dc869d4ba093a96bd3b49d1b67933ef7bea2716c269511

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server equilibre::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

analyze_chess_game · version identified by source hash

Chess move-quality profile

Legal SAN/UCI conversion and forced cleanup of an unresponsive engine; optional actual Stockfish analysis.

Limits, source and reproducible checks

Scores and move agreement depend on Stockfish version, depth, positions and analysis settings.
A descriptive move-quality profile is not a cheating verdict or a calibrated identity classifier.

SHA-256: 4060ca4799be4b61040a5131144c1b8ca3368bd5f430fbaf486a2e45a2222778

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server echecs::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

screen_wash_trading · version identified by source hash

Repeated-trade pattern screening

Designed repeating-volume and organic-flow fixtures, input sorting and overflow rejection.

Limits, source and reproducible checks

Synthetic examples do not measure real-market false-positive or false-negative rates.
Only the supplied series, time window and price band are examined; this does not establish wash trading or intent.

SHA-256: e6c4cdff6923c90a7f8f022f1f60552f393262651715d75086ba2485eeb2aae3

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server marche::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

route_optimizer · 0.1.0

Geometric route optimization

All permutations on small fixtures, independently recomputed leg costs, analytic equator/antipodal distances and bounded-interruption checks.

Limits, source and reproducible checks

Exact means a finished combinatorial search over the computed f64 distances, not exact real arithmetic.
For more than 13 points or incomplete search, returned routes are bounded heuristics.
No roads, traffic, capacity constraints or travel-time prediction.

SHA-256: 20d09cc91c01e06ef7b5f57d929a1899db9191828aa33d190a0d32131501c90d

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test route_optimizer_contract -- --test-threads=1

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server route_optimizer::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

triangulate_tracks · 0.2.0

Multi-observer track reconstruction

Independent reprojection, numerical information matrix, known synthetic motion, SciPy fit of a historical stall and translated/rotated-coordinate checks.

Limits, source and reproducible checks

Constant velocity, spherical ECEF, exact supplied stations/clocks and declared independent angular uncertainty.
Local covariance omits systematic model errors and is not calibrated field accuracy.
No universal convergence guarantee and no new held-out study was run to create this card.

SHA-256: 7fc9b1ddf7e9dd5b747e421ff8ca6b8e7ad09e28948053dc72310ce7d4a77f9f

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test triangulation_stable_contract -- --test-threads=1

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test triangulation_stable_review -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

plan_golden_hour · 0.1.0

Golden and blue hour windows

Stored Skyfield 1.55/JPL DE421 reference days, threshold tangencies, polar/boundary windows and independent azimuth samples.

Limits, source and reproducible checks

Geometric solar-center windows on a UTC day; sunrise at −0.833° is a separate convention.
DE421/reference fixtures validate their particular dates, not the whole accepted 1900–2100 interval.
No terrain, buildings, clouds, weather or guarantee of photographic light quality.

SHA-256: 444f3168e7ec20b439d7f0083d9d8fc3f9de14933ec206c2f67cea97dfe971dd

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test golden_hour_contract -- --test-threads=1

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test solar_azimuth_contract -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

propagate_uncertainty · 0.1.0

Measurement uncertainty propagation

JCGM 101 mass-calibration examples, exact moments of squared normals, lognormal quantiles and an independently seeded NumPy oracle.

Limits, source and reproducible checks

Depends on the declared measurement expression, dimensions and input probability models.
Fixed caller-specified draw count; not the full adaptive JCGM 101 procedure.
Undefined draws and nonconvergence remain possible; no current CPU/GPU speedup is asserted.

SHA-256: 5ee2d2b93b358d02768e0e7623061bea03fbc6448595f2fa99bda0c7bcedeebd

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test uncertainty_contract -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

hear_the_shape · 0.1.0

Polygonal membrane spectrum

Fixed/free square spectra, equilateral triangle, L-shaped membrane and Driscoll’s first 12 GWW drum eigenvalues.

Limits, source and reproducible checks

Ideal polygonal membrane with fixed or free boundary, not a damped real drum or elastic plate.
Richardson error estimates are not rigorous enclosures of all numerical error.
Matching finitely many modes does not prove isospectrality.

SHA-256: 5ccb62e29e1c5e5bee5df248fe2a535dfbc2ceb3a150f6b51b10fe5e53b05dd3

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test hear_shape_contract -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

kepler_stability · 0.2.0

Finite-time orbital sensitivity

Zero-perturber Kepler limit, Jacobi/convergence diagnostics, resonance examples and independent DOP853 encounter comparisons recorded in the review.

Limits, source and reproducible checks

Planar circular restricted three-body model and bounded finite-time integrations.
Sensitivity bands are heuristic diagnostics, not proof of chaos or long-term stability.
Encounter guards, uncomputed cells, numerical warnings and conservation errors must be read with the map.

SHA-256: eb4e3a0ad16db9767578512de36d5e5f38110d8881c20465a8c7ca46dc0b2841

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test kepler_stability_contract -- --test-threads=1

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --bin scorecompute-server kepler_stability::tests -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

check_arithmetic · 0.2.0

Arithmetic claim comparison

262 stored Python Fraction comparisons, four real-JSON digest vectors checked with hashlib, and explicit cancellation/overflow/parser-limit regressions.

Limits, source and reproducible checks

Checks arithmetic of decoded numeric inputs and expression literals, not units or factual/model meaning.
Numeric claims/variables are limited to the safe JSON integer-magnitude range; original JSON decimal token spelling is not retained.
Unsupported or overflowing exact operations and libm estimates are undetermined, not a verified approximate result.

SHA-256: add11968562a70ce7c62a187956026443f016f351d5f816edc6e2afc818d964a

CARGO_TARGET_DIR=/tmp/scorecompute-engine-evidence-target node scripts/cargo.mjs test --release --manifest-path server/Cargo.toml --test check_arithmetic_contract -- --test-threads=1

Commands require the matching repository checkout and stated dependencies. The evidence exporter does not run them automatically.

Inspect the result.Keep the limits.

Three things a receipt can tell you.

What was calculated

Who issued these bytes

What was actually observed

Results that did not earn promotion.

Benefit not demonstrated

Four gates passed. One did not.

The complete gate still matters.

Boundaries you can check.

A reference for each calculation.

Monte Carlo estimate of π

Exact task selection

Task-duration risk simulation

Ideal two-body orbit

Solar-shadow consistency

Claimed Sun or Moon position

GRIM/GRIMMER consistency screening

Angular-speed kinematics

Shadow-direction track consistency

Kuhn poker equilibrium

Chess move-quality profile

Repeated-trade pattern screening

Geometric route optimization

Multi-observer track reconstruction

Golden and blue hour windows

Measurement uncertainty propagation

Polygonal membrane spectrum

Finite-time orbital sensitivity

Arithmetic claim comparison

Inspect the result.
Keep the limits.