The reconciliation step where we use no model at all

Description and use

A deterministic reconciliation engine without models matched 3,000 subscription pairs between HubSpot and Stripe, identified 240 discrepancies, and reported a €900 total delta, validating the comparison rules.

Problem

Reconciling 3,000 subscriptions a week between HubSpot and Stripe produced 240 discrepancies; a read‑only discrepancy report is needed for an operations lead.

Outcome

The deterministic engine classified all 3,000 pairs as expected, found all 240 planted discrepancies with no false positives or missed cases, and reported a supported amount delta of €900.00. The comparison ran in milliseconds and the build regenerates and re‑verifies each time.

Workflow

• Define comparison contract: stable ID, same period, EUR cents, agreed status vocabulary.
• Index both sides by subscription ID.
• For each ID, determine outcome: if reasons exist → review; else compute amountDeltaCents, check safe integer, push reason codes for mismatches, set outcome to discrepancy or match.
• Collect source rows, reason codes, and amount delta for each pair.
• Generate expected‑result manifest with known counts for matching, amount differences, status differences, and unsupported currency cases.
• Verify engine output against manifest: classification, reason codes, per‑pair delta, and supported total.
• Publish summary, examples, review queue, and evidence file for external verification.

Evidence

The engine indexes both sides by subscription ID and gives every ID one of three outcomes: match, discrepancy, review.

On this dataset, the run classifies all 3,000 pairs as expected. It finds all 240 planted discrepancies with no false positives and no missed cases, and every reason code matches.

The comparison itself runs in milliseconds; the build regenerates and re‑verifies it each time.