The runs first differ at match, where price_tolerance went from 2.00% to 5.00%.
Every stage before it is the same. The stage after it differs.
read_invoiceSame
matchFirst difference
approve_paymentChanged after it
Checked from the records. The same answer every time.
What may have happened
AI-generated, may be wrong · model anthropic:claude-sonnet-4-6
Hide the detailsShow the details
Lead (possible culprit)
The `match` node is identified as the first difference point, and its output `match.json` differs in both size and digest between baseline and candidate.
The `approve_payment` node is downstream of `match` and its differing input `match.json` and output `payment.json` are a consequence of the upstream change in `match`, not an independent source of divergence.
Both runs' records agree. read_invoice was the same in both. match's settings differ in price_tolerance: 2.00% → 5.00%. Its output differs, and the stage after it changed too.
11 {22 "amount": "512.00",3- "matched": false,3+ "matched": true,44 "order": "PO-501",5- "problems": [6- "unit price 12.80 is 2.40% over the order's 12.50, past the tolerance of 2.00%"7- ]5+ "problems": []86 }
approve_payment · Changed after it
settings: none changed
input match.json (operator-authored) changed: 5b0f7e319dbf -> e61819271007; content not stored
11 {22 "amount": "512.00",3- "decision": "hold",4- "problems": [5- "unit price 12.80 is 2.40% over the order's 12.50, past the tolerance of 2.00%"6- ]3+ "decision": "pay",4+ "problems": []75 }