Verifying a Genomarker Receipt

What this proves

• Byte-equivalence: the receipt JSON you hold matches the one Genomarker issued.
• Integrity of metadata: parameters, software environment, dataset fingerprint, and result hashes are unmodified.

What this does not prove

• Bit-replay: the receipt does not by itself let a third party re-run the analysis. Replay requires the Replay Sandbox CLI (planned in a future release).
• Identity: there is no cryptographic signature in the 1.x schema — the hash anchor + URL trust suffices today; future receipts may add signing.

Reference verifier — Python

pip install rfc8785

import hashlib
import json
import sys

import rfc8785

with open("receipt.json", "rb") as f:
    payload = json.loads(f.read())

expected = payload.get("integrity", {}).get("canonical_json_hash")
subject = {k: v for k, v in payload.items() if k != "integrity"}
recomputed = "sha256:" + hashlib.sha256(rfc8785.dumps(subject)).hexdigest()

print("PASS" if expected == recomputed else "FAIL")
print(f"expected={expected}")
print(f"recomputed={recomputed}")
sys.exit(0 if expected == recomputed else 1)

Reference verifier — Node

npm install canonicalize

import { readFileSync } from "node:fs";
import { createHash } from "node:crypto";
import canonicalize from "canonicalize";

const payload = JSON.parse(readFileSync("receipt.json", "utf8"));
const expected = payload.integrity?.canonical_json_hash;

const subject = { ...payload };
delete subject.integrity;
const recomputed = "sha256:" + createHash("sha256").update(canonicalize(subject)).digest("hex");

console.log(expected === recomputed ? "PASS" : "FAIL");
console.log("expected=" + expected);
console.log("recomputed=" + recomputed);
process.exit(expected === recomputed ? 0 : 1);

JSON Schema

The full schema for the receipt 1.x family (1.0.0, 1.1.0, 1.2.0, 1.3.0, 1.4.0, 1.5.0, 1.6.0, 1.7.0, 1.8.0, 1.9.0, 1.10.0, and 1.11.0) is published at https://genomarker.com/schemas/receipt-1.0.0.json. Minor bumps add optional fields only and keep this URL — 1.1.0 adds the optional method.handler_specific keys calibration_metrics, decision_curve_analysis, and subgroup_performance (emitted by model-training receipts). 1.2.0 adds the optional method.handler_specific.data_quality confounder pre-flight block (emitted by differential-expression receipts). 1.3.0 adds the optional method.handler_specific.analysis_classification field (one of "pre-specified", "post-hoc", or "exploratory") and the optional method.handler_specific.sap provenance block ({id, payload_sha256, zenodo_doi, locked_at, amendment_chain_depth}) — pre-registered SAP support added in Phase 3 (2B.7.7). 1.4.0 adds the optional method.handler_specific.reporting_checklists block (TRIPOD+AI / REMARK / STARD-AI auto-emission emitted by every inferential handler with applicable checklists; Phase 4 / 2B.7.8) and the optional method.handler_specific.reporting_supplements block (frozen prose snapshot at publish-time). 1.5.0 adds the optional method.handler_specific.repro_score block (Reproducibility Score emission — REPRO-10 / Phase 5 / 2B.7.9; composite 0-100 + letter A/B/C/D/F + 12 per-axis grades + evidence- link map). 1.6.0 adds the optional method.handler_specific.ingest_quality block (Robust Ingest Layer — INGEST-01..04 / Phase 1 v3.0; upload-time gene-ID harmonization + Excel repair + PHI screen + colData alignment + TPM rejection, with the tiered verdict hard-refuse / block-with-override / warn / ok). 1.7.0 adds the optional method.handler_specific.gate block (Tiered Gate + Signed Record — GATE-01 / Phase 2 v3.0; the composite severity-tiered gate verdict across confounder / colData / PHI / high-unmappable / ML-leakage / SAP-amendment surfaces, plus the signed-override attribution when a block-with-override was waived). 1.8.0 fills the previously-reserved integrity.signing slot (RO-Crate / PROV emission — REPRO-12 / Phase 4 v3.0; a detached ES256 JWS over the count-only rigor verdict, verdict_status "origin-signed-replay-pending" — PHI-free and offline-verifiable; null on legacy receipts). 1.9.0 adds the optional method.handler_specific.methods_parse block (Methods-Diff v1 — REPRO-11 / Phase 8 / 2B.7.10; the reproducibility-relevant metadata of the Claude Sonnet 4.6 LLM call that parsed external paper Methods prose into a typed MethodsCandidate — source + prose/system-prompt/parsed-candidate SHA-256s + parser provider/model/version + target handler type + user-confirm timestamp + token counts; emitted only by the methods_parse handler). 1.10.0 adds the optional method.handler_specific.sc_aggregation block (PHI-safe counts-only pseudobulk reduction, emitted by the sc_pseudobulk handler) and the optional method.handler_specific.mast block (MAST mixed-model method, emitted by the sc_mast handler) — Single-Cell DE, STAT-05 / Phase 9. 1.11.0 adds the optional method.handler_specific.evidence_sources_used block (version-only evidence provenance — bundle_version + per-source {version, license} + bundle sha256, no gene list; embedded only at export time when a gene-evidence reference is included) — Gene Evidence Card, GENE-EVID-02/03 / Phase 10. The schema URL is forever-stable; future major versions get new URLs.

Scope notes

The stdlib-only verifiers shipped inside receipt.zip use a hand-rolled canonicalization that is correct for typical payloads but does not cover every RFC 8785 edge case (exotic floats, surrogate-pair keys, U+2028/U+2029). For production verification or audit, prefer the reference snippets above.