HARBINGERAPI and Sandbox

PUBLIC SYNTHETIC EVIDENCE REVIEW

PUBLIC DEMO · V3Stateless GET over authored synthetic text · public demo evaluators · private product Core not connected · authority NONE

Run the comparison. Inspect what changes.

Choose financial record review or industrial supply chain. Compare the supplied records, check the evidence, and see what changes when you add supporting material.

01

Read the records, then hand off the review

Open the selected records before running the review. Tables show their contents alongside the original fields and line numbers. These synthetic records were written for the example, not extracted automatically. The published result was calculated from the same fields you see.

The review summary brings together supported findings, unresolved questions, requested material, the selected review path and its synthetic requirements. Completed comparisons and withheld calculations are listed separately; no measured time savings are claimed. The before/after table compares the current packet with the original packet of the same case. Returning to the original resets the packet, review path and synthetic identity and runs a fresh GET.

A shared link opens the review summary and repeats the request. Its view=summary and run=1 values control the page, not the API. Findings, sources, and unmet requirements come from the verified record; display headings and layout are not hashed.

02

Run either domain without a key

The endpoint parses fixed key=value source records and computes their reference, period and quantity relationships. It does not retrieve external material or interpret arbitrary documents. Input selects a published packet, not a desired finding.

Financial record reviewbash
curl --fail-with-body 'https://developer.calamophyton.com/api/demo/v3/golden-thread?release=2026-09-05.1&domain=finance-dd&packet=initial&path=review-case&grant=reviewer'
Industrial supply chainbash
curl --fail-with-body 'https://developer.calamophyton.com/api/demo/v3/golden-thread?release=2026-09-05.1&domain=supply-chain&packet=requested-material&path=review-case&grant=reviewer'
03

Add material—not a “resolved” flag

  • Finance: match appendix reference, entity, period and version; recalculate 710 / 1,000 × 100 = 71%. Wrong-period or conflicting records leave the check open.
  • Supply: compare dated stock, demand and receipts in the same item/site/unit. In requested-material, assuming the September 25 receipt, 500 opening units produce a minimum closing balance of 40. In low-stock, 200 opening units first produce a negative balance on September 12. The unreconciled transport date remains a separate question. aligned-transport reconciles that date in the same case; reviewable-shortfall reconciles it while retaining a known stock shortfall for human review.
  • The reconciled packet is a separate complete reference case, not a resolution of the original incident. It has no outstanding defined evidence check. This is not an investment conclusion or supply-chain all-clear.
04

Keep the three responsibilities separate

The public implementation does not call private HARBINGER, PÓLYA or RAIFFA packages. Matching product names or contracts do not establish Core integration. Real product integration requires exact permitted package bytes, compatible runtime and a separately verified export boundary.

  • HARBINGER surface: source-linked comparisons, changes and unresolved questions.
  • PÓLYA surface: review paths marked available, waiting or not needed from the current evidence state. No recommendation or decision.
  • RAIFFA surface: compare fixed synthetic role, scope, route, validity, revocation and used-ID records. These are real comparisons of mock records, not real credentials, approval or a durable replay ledger.
05

Verify the record independently

release=2026-09-05.1 pins source data and evaluator behavior. Hash the UTF-8 bytes of receipt.canonical with SHA-256 and compare receipt.digest. Also parse canonical and compare it with record. The complete record includes reasons, source fields and limits. Only meta.requestId and meta.servedAt are excluded.

The browser performs both checks before displaying a result. Hash equality proves neither truth nor authorization. Public links with run=1 repeat a GET; no personal history is stored. Old unversioned links open the retained v2 execution, not v3. Unsupported releases are rejected.

Independent SHA-256 check (Node.js)javascript
import { createHash } from 'node:crypto';
import assert from 'node:assert/strict';
const origin = process.env.DEMO_ORIGIN ?? 'https://developer.calamophyton.com';
const http = await fetch(origin + '/api/demo/v3/golden-thread?release=2026-09-05.1&domain=finance-dd&packet=requested-material&path=review-case&grant=reviewer');
assert.equal(http.status, 200);
const response = await http.json();
assert.equal(createHash('sha256').update(response.receipt.canonical, 'utf8').digest('hex'), response.receipt.digest);
assert.deepEqual(JSON.parse(response.receipt.canonical), response.record);
console.log({ digest: response.receipt.digest, authority: response.record.authority.issuedAuthority });
06

Understand the assumptions and measure the real work

Supply calculations use calendar days, constant daily demand, fully usable stock and receipts available at the start of the stated date. Safety stock, reservations and inspection delays are not modelled. Roles and review routes are illustrative, not practitioner-validated operating policy.

Compare identical review tasks with and without assistance: source preparation, review and correction time; critical omissions; false alarms; and rework. Counts of checks are not measured hours saved. No customer performance claim is made.