Plaid v1↔v2 writeback parity harness

A destructive Salesforce E2E script that generates the same Plaid quote in both Dealops v1 and v2, writes both to SFDC, reads them back, and reports every structural diff with v1 as ground truth.

1. Why this exists

Dealops v2 is being rolled out as the replacement for the v1 (Penguin) pricing flow for Plaid. Before flipping customers over, the team needs confidence that a v2 quote, once written through to Salesforce CPQ, produces the same SBQQ__QuoteLine__c rows and field values that v1 would have produced for the same input.

Unit tests and golden-file fixtures can't reach that far — the truth lives in what Salesforce stores after both sides hit the SFDC writeback path. So this PR adds a destructive end-to-end harness that exercises the real path on disposable prod opportunities, then mechanically compares the readback payloads.

2. How the harness runs

Each case in the input file describes one quote and two Salesforce opportunities. For every case the script does six things, in order:

The same logical quote (a use-case bundle or an explicit SKU list) is materialized twice — once in v1's product/price shape, once in v2's — written to two separate but topologically-equivalent disposable opportunities, then both quote line item sets are read back from Salesforce and diffed row-by-row.

3. Case file shape

The runner accepts JSONL, a JSON array, or { "cases": [...] }. Each case identifies both sides and describes the quote. Two case styles are supported:

{
  "caseId": "income-verification-10k",
  "quote": {
    "useCase": "Income Verification",
    "pricebook": "CPQ",
    "currency": "USD",
    "defaultVolume": 10000
  },
  "v1": { "organizationId": "...",
          "userId": "...",
          "opportunityId": "..." },
  "v2": { "organizationId": "...",
          "userId": "...",
          "opportunityId": "..." }
}

{
  "caseId": "custom-identity-skus-10k",
  "quote": {
    "pricebook": "CPQ", "currency": "USD",
    "defaultVolume": 10000,
    "products": [
      { "productCode": "AUT - UF", "volume": 10000 },
      { "productCode": "IDV1 - UF", "volume": 10000,
        "tiers": [...],
        "flattenTierStrategy": "blended" }
    ]
  },
  "v1": { ... },
  "v2": { ... }
}

Product resolution

4. Safety surface

This script writes to real Salesforce. The PR description leads with a "safety / false-positive audit" for that reason. Four guards stack on top of each other:

What counts as "acceptable" vs structural

5. The 21 June destructive run

The PR also checks in the final prod run that motivated shipping the harness. The input is the Plaid pricing-team sheet, converted to JSONL. S11–S15 were empty templates in the sheet and are skipped.

Top recurring diff fields (21 June run)

How to read a diff

{
  "ProductCode": "PILDL-UF",
  "SBQQ__CustomerPrice__c": 6.30,
  "SBQQ__Discount__c": 0.05,
  "SBQQ__DefaultSubscriptionTerm__c": 12,
  "SBQQ__RequiredBy__c": "<parent: Plaid Income (LDL)>"
}

{
  "ProductCode": "PILDL-UF",
  "SBQQ__CustomerPrice__c": 6.30,
  "SBQQ__Discount__c": null,            // diff
  "SBQQ__DefaultSubscriptionTerm__c": null,  // diff
  "SBQQ__RequiredBy__c": "<no parent>"        // diff
}

6. Run artifacts checked in

Two run directories ship in this PR. Each contains both the structured Salesforce readback payloads and the human-readable reports.

7. What this PR does not change

8. Risks & open questions

Things worth asking in review