■ PACK P01-PUBLIC · REC.BANK-LEDGER.V1 · SAMPLE
Harbor Ledger Demo (SAMPLE, fictional desk)
- Candidate:
- deterministic rules matcher, not a language model (matcher.js, match())
- Baseline:
- deterministic rules baseline (amount, date, exact reference)
Bookkeeping clerks, payments ops. Accountants — recon / close slice only.
VERDICT
works with controls · n=36 bank lines · 2 seeded judgement items · single fixture, single vintage
AXIS 01 · VERDICT
works with controls
AXIS 02 · READINESS
requires integration
AXIS 03 · COVERAGE
36 bank lines · 33 ledger lines · single fixture
■ RESIDUAL WORK · TWO DENOMINATORS
- BANK SIDE
- 15 of 36 lines stay with the desk9 proposed · 2 holds · 4 no counterpart
- LEDGER SIDE
- 1 of 33 lines unmatchedL22 · accrued postage $210.00 · no bank line yet
16 items of residual work in total, across two different denominators. The two counts are not added into one rate.
■ BY VALUE
Counted by value, the 21 postings carry 47.3 per cent of the money in the file against 58 per cent of its lines. The four lines with no counterpart proposed are 11 per cent of the file by count and 26.9 per cent by value.
■ RUN EVIDENCE
- PRE-RUN COMMITMENT
- COMMITMENT.md in that repository
- TRANSPARENCY LOG
- 2582597216 (commitment) · 2582616386 (result)
Residual human work · 15 of 36 bank lines still a desk · 16 items
Rubric · computed this run
Precision measures errors in what was posted. Exception rate measures what stayed human. Both can be true at once.