# Validation record — 25 October 2026

Private package/source readiness; visitor synchronization and route evidence are a separate coordinator release gate. All inputs are synthetic Cedar evidence, not real matched customer records. No audio, encoded video, publication, database write or empirical match-accuracy result is claimed.

## Reproduction

From the repository root:

```powershell
python algorithms/domains/D30-banking-ledger-and-reconciliation/D30-F03-reconciliation/D30-F03-A02-fuzzy-transaction-match/tests/run_parity.py
python -m unittest discover -s algorithms/domains/D30-banking-ledger-and-reconciliation/D30-F03-reconciliation/D30-F03-A02-fuzzy-transaction-match/tests -p test_fuzzy.py
node algorithms/domains/D30-banking-ledger-and-reconciliation/D30-F03-reconciliation/D30-F03-A02-fuzzy-transaction-match/tests/visual-qa.mjs
```

TypeScript compiles strictly for ES2022/NodeNext into tests/.tmp-compiled, which is ignored and excluded from release hashing by its temporary path prefix. Both learner implementations are self-contained. research/PINNED-SELECTOR.json records the copied upstream selector-prefix hashes. tests/build_lab.py embeds the actual compiled final TypeScript kernel; no independent hand-written scoring approximation powers the lab.

## Executable evidence

- 59 fuzzy entire-output Python/TypeScript parity cases: 27 realized pre-code token/date/reference graphs, three pre-code revision-cutoff cases and 29 strict/resource regressions.
- 105 upstream independently authored complete selector outputs are preserved locally and checked in both languages. Combined total: 164 complete-output parity cases, including coverage, selected revisions, errors, comparison features and row decisions.
- Four Python test methods pass, including exact independent scores, one-row-once, source revision visibility, delivery permutation, no input mutation, detached output views and the full 16384-comparison 128×128 tied graph. TypeScript independently exercises that full bound and confirms no lexical tie proposal.
- Twenty-one browser scenario outputs (seven scenarios at three widths) are compared as complete JSON against Python, independently of browser display assertions. Browser comparisons use the actual materialized lab requests, not abbreviated fixtures.

## Rendered browser evidence

research/visual-qa-results.json records 219 meaningful row observations across 390,768,1440 widths: six valid scenarios ×12 row states ×3 widths plus an incomplete-source rejection at each width. The default source retains 24 events and 144 pair comparisons; future/known scenarios retain one additional revision event. Current focus never excludes competitors. Checks cover exact score attributes, row identities, full comparison count, all controls, parameter recomputation, deterministic reset, real Play/Pause, playback cancellation, reduced-motion one-step Play, keyboard navigation,44px control targets, no document overflow and zero console warnings/errors.

Both authored SVGs and Mermaid were rendered to PNG and visually inspected for arithmetic, labels, clipping and contrast. Phone and wide lab screenshots were inspected. The local audit table alone can scroll horizontally to preserve exact fractions and reasons. Public-safe receipts and screenshots are copied into examples/browser-qa; source QA files remain in research/qa.

## Interpretation boundary

Functional correctness applies to the frozen conservative policy and synthetic evidence. Scores remain points, not probability. One-row-once is within a run only; host accepted-resolution/source-trust/journal obligations remain outside. Same-namespace contradictory references can withhold true pairs. No calibration, measured precision/recall, production throughput or durable acceptance is established. Historical example decision: not useful without licensed paired records, receipts, namespaces and adjudicated truth. See [source roles and limitations](REFERENCES.md).

Evidence downloads: [parity results](parity-results.json), [browser results](browser-qa/results.json), [exact contract](data-contract.md). The copied evidence sits under examples so article downloads remain available on the visitor website. The coordinator adds actual route/iframe acceptance after synchronization; local screenshots alone do not certify that surface.

## Local package gates

The full topic validator passed with zero errors and zero warnings; the playground validator passed. The receipts are preserved in package-validation.json. This covers local source readiness. Visitor routes and embedding require the coordinator's separate synchronized website check.