This is the report every partner receives from us on Friday, unchanged in format from pilot to year three. One page, written by the person who did the work. The two reports shown here are real but cover a two-day build rather than two weeks: one per deliverable (docrecon on 15 October, docflow on 16 October). They are shown so you know what you would get, not what we would promise. Week 2 →

Delivery report — docrecon

ON TRACK

Report: 1 · deliverable: docrecon · 2026-10-15
Scope: Open-source reconciliation engine, two implementations, parity-tested
Demo: dalteno.com/lab/docrecon/
Repo: github.com/dalteno/docrecon

Done this week — against the agreed acceptance criteria

Criterion	Status	Evidence
Visitor can enter values we never anticipated and get an explained verdict	✅	Live demo; every finding carries rule, field, expected, actual, reason
Rules identical in .NET 10 and JS	✅	52 golden reports; .NET parity test compares every field
≥ 40 shared cases incl. adversarial, expectations hand-written	✅	52 cases, 15 adversarial, `cases/generate.mjs`
CI green on both engines	✅	`.github/workflows/ci.yml` — JS, .NET, generator drift check, golden coverage check. Update 18 Oct: first real run green — GitHub Actions run #3, commit `99c5257`, 33 s total; `dotnet test` on CI: Dalteno.DocRecon.Tests Passed 145, Failed 0 (305 ms, net10.0)
Documented decisions	✅	ADR-0001 pure engine · ADR-0002 integer arithmetic · ADR-0003 cases as spec

Not done / changed

Multi-currency invoices with per-line FX conversion — out of scope, never planned (R08 flags mixed currencies instead); listed so nobody assumes otherwise.
Air waybill support — not in scope for v1.0; the model is sea-freight only. Would be a v1.1 item if a partner needs it.

Found and fixed

Mutation check showed the ±0.01 amount tolerance was masking a rounding-mode change at the integration level. Added C17b_half_cent_zero_tolerance. (See case study, "what went wrong".)
The gross-weight boundary case was initially written against the smaller value; the rule measures tolerance against the larger value (documented). Case C34 corrected before the suite was frozen.

Measured

JS: 62 tests passing · 38 µs per bundle (1,000 bundles / 38 ms, Node 22)
.NET: 136 tests defined (52 expectation, 52 parity, 32 unit) at the time of this report. Update 18 Oct: after the external review added three cases, the suite is 145 tests — 145 passed, 0 failed on GitHub Actions (run #3, 305 ms, net10.0)
Mutation checks: 3 run, 3 caught

Decisions needed from you

None blocking. Optional: should weight tolerance default stay at 2 % or follow your broker's practice? One-line change in ReconcileOptions.

Risks

Low: party-name normalisation strips one trailing legal suffix; names with a suffix in the middle ("Acme GmbH & Co. KG") compare literally. Documented in R17; a longer suffix list is a config change, not a code change.

Next week (if this were a continuing engagement)

v1.1: configurable rule set per client (enable/disable, per-rule severity), air waybill model, CSV import for line items.
Demo on Friday 16:00 CET in your repo or ours.

Format: one page, every Friday, from the person who did the work. Questions: hello@dalteno.com.