Proof Report
Proof Report — SAR Multi-Crop Acreage Estimator (Round 1)
Strictness: PROOF_STRICTNESS = IRS_AUDITOR Status: PROOF_INCOMPLETE
CERTIFIED is not used. Under IRS_AUDITOR rules, CERTIFIED is permitted only when all required artifacts exist, all checks pass, reproduction instructions exist, raw evidence is preserved, unsupported-claim count = 0, **and disclosed seams do not contradict the customer outcome**. The customer outcome is an official leaderboard submission; the official dataset is unavailable in this workspace, so that outcome is a DISCLOSED_SEAM. Status is therefore PROOF_INCOMPLETE.
What is claimed, and the evidence for each
All numbers below have lineage in proof/CLAIM_EVIDENCE.json and verification-report.json. Reproduce with node verify.mjs (see proof/REPRODUCE.md).
| Claim | Evidence | Status |
|---|---|---|
| Pipeline passes its structural + synthetic checks | verification-report.json (12/12), proof/evidence/verify.log | SUPPORTED |
| Synthetic MSE / baseline / skill / R² / per-crop RMSE | verification-report.json#/syntheticBenchmark | SUPPORTED (synthetic input) |
| Official evaluation path executes (k-fold CV + submission) | proof/evidence/verify-official-fixture.{json,log} | DISCLOSED_SEAM (synthetic-derived fixture inputs) |
| Official leaderboard MSE | none | DISCLOSED_SEAM — NOT PRODUCED |
| Raster→zonal-stats (Stage A) on real tiles | run-deploy-instructions.md (documented only) | DISCLOSED_SEAM — NOT EXECUTED |
Downgraded statements (per IRS_AUDITOR rule 8)
- "Solves multi-crop acreage estimation" → "Estimates five crop areas via ridge
linear unmixing; evaluated only on a synthetic Kharif benchmark under documented conditions (see verification-report.json)."
- "Certified / production-ready" → "Passed 12/12 documented checks on synthetic
data; production readiness, deployment, security, and operational tests are not claimed and were not run."
- "Leaderboard-ready" → "Submission-schema-valid; official accuracy is a
disclosed seam pending the real dataset."
Disclosed seams (DISCLOSED_SEAM)
- Official dataset absent — no real SAR tiles/labels in workspace; no
leaderboard MSE produced.
- Stage A not executed — calibration + zonal statistics from GeoTIFFs is
documented, not run (needs GDAL/rasterio/SNAP/GEE).
- All metrics are synthetic — generated by
src/synth.mjs; not field data. - Optional residual forest disabled by default — synthetic numbers are
ridge-only.
See proof/LIMITATIONS.md and the verification report's "Disclosed seams & limitations" section for the authoritative list.
Required proof artifacts (IRS_AUDITOR proof package)
Present under proof/: EXECUTIVE_EVIDENCE.md, CLAIM_EVIDENCE.json, EXECUTION_TRACE.json, ARTIFACT_MANIFEST.json, CHECKSUMS.json, REPRODUCE.md, VERIFY.md, LIMITATIONS.md, AUDITOR_OBJECTIONS.md, PROOF_SCORECARD.json, and raw runs under proof/evidence/. Regenerate with node tools/proof.mjs (or, from the repo root, node tools/forge-proof.mjs --outcome delivery-package/sar-crop-acreage). Verify integrity with `node tools/forge-proof-verify.mjs --outcome delivery-package/sar-crop-acreage`.
Path to closing the disclosed seam
Provide data/official/zonal.csv + data/official/train-labels.csv, run node verify.mjs (real out-of-fold CV MSE) and node tools/proof.mjs. If all checks pass on real data with evidence preserved and the seam closed, status may be reconsidered against the certification criteria in proof/PROOF_SCORECARD.json.