# Intercompany validation — 25 October 2026

Private source/script/asset readiness. All data and receipts are synthetic. No source authentication, reporting opinion, accepted worksheet, durable journal, closed period, audio or encoded/published video is claimed.

## Reproduce from the repository root

```powershell
python algorithms/domains/D30-banking-ledger-and-reconciliation/D30-F04-period-close/D30-F04-A03-intercompany-elimination/tests/run_parity.py
python -m unittest discover -s algorithms/domains/D30-banking-ledger-and-reconciliation/D30-F04-period-close/D30-F04-A03-intercompany-elimination/tests -p test_intercompany.py
node algorithms/domains/D30-banking-ledger-and-reconciliation/D30-F04-period-close/D30-F04-A03-intercompany-elimination/tests/visual-qa.mjs
```

The exact model and independent expectations preceded implementation: 58 initial complete outputs include four literal math cases; 11 later strict/resource regressions bring the complete-output suite to 69. Both self-contained learner implementations match every expected branch, source copy, pair diagnostic, worksheet leg and role rollup. The complete 256-pair/512-leg case, 60-digit magnitudes, missing/late evidence, omitted internal positions, retained external rows, strict identities and signs all pass. No sibling runtime imports are used. Strict ES2022/NodeNext compiles into tests/.tmp-compiled. The lab embeds that actual final module, removing only export syntax.

Four Python methods compare independent outputs; check no entity/input mutation, detached mutable result branches and permutation invariance; verify positive balanced worksheet legs and all-or-nothing final branches; and compare all 24 actual browser-captured complete outputs against Python. Browser focus steps cannot alter the full evaluated result.

Eight scenarios at 390/768/1440 produced 189 meaningful observations: five evaluated cases ×12 intended pair positions ×3 widths plus three whole-source/request failures ×3 widths. The 24-position slice stays present; external adds one retained position. Exact residual 100 and 1 withhold, 0 permits; a matched second pair stays conditional if another internal pair is unresolved. Omitted specs preserve two unassigned internal positions. Late, duplicate and missing source/request evidence returns no partial certification.

All parameter/reset/keyboard/transport controls, real Play/Pause/cancellation, reduced-motion one-step, 44px targets, no document overflow and zero console warnings/errors pass. Both SVGs and Mermaid were rendered; phone mismatch, wide external and both static screenshots were inspected. Exact tables retain local horizontal scrolling. Actual receipts: research/visual-qa-results.json, research/browser-reference-outputs.json, research/qa and tests/parity-results.json. Public-safe copies: examples/browser-qa and examples/parity-results.json.

The trusted host must bind exact dedicated eligible positions to passed Trial-Balance raw lines and visible p at K. Current close requires p equal to the frozen head; historical p<h does not pass. Eligibility must exclude non-eliminable FX effects and align the income/expense reporting interval. These tests cannot authenticate host assertions. See the source roles in [REFERENCES.md](REFERENCES.md) and the [strict contract](data-contract.md). The coordinator separately records visitor route, article H1 and iframe checks after synchronization.