Measured performance. Defined limits.
Schedule Forensics is measured against stored fields in a fixed collection of P6 exports. The figures below describe that collection and a frozen engine snapshot—not a guarantee for a new schedule or proof that a professional conclusion is correct.
Latest completed minute-mode snapshot
| Comparison | Matching fields | Fields compared | Agreement |
|---|---|---|---|
| Dates and day-based float comparison | 542,897 | 584,687 | 92.85% |
| Exact timestamps and minute-based float comparison | 537,411 | 584,687 | 91.91% |
These are two scoring methods on the same run. The first can count values as matching when their dates agree but their times differ. The second retains that finer distinction. Neither percentage is a report score, a clause pass rate or an estimate of time saved.
The retained results contain 82 scored project records and one additional error record for a cyclic network. That error is not represented as a successful comparison. The comparison denominator remains fixed at 584,687 fields.
Verified on 10 September 2026 from the completed frozen full run. Download the aggregate measurement record, including source-manifest and result-file hashes. The raw corpus and project-level results are not distributed by this website.
What improved
The newer scheduler can retain working-minute precision through calendars, activity durations, relationship lags and scheduling calculations. Subsequent corrections address progress semantics, completed work and float propagation. The historical day-mode benchmark was 396,735 / 584,687 (67.85%); it remains in the technical history and in explicitly historical report disclosures.
Do not substitute the newer percentage into an older report. A valid comparison needs the engine revision, calculation mode, field definitions and population. The public sample demonstrates recorded command outputs; it is not the benchmark dataset.
What this measurement does not establish
- Universal P6 equivalence. Stored exports can contain contradictory fields or uncertain exporter provenance. This is comparison against retained values, not a new licensed P6 recalculation of every file.
- Representative performance for every project. The collection is concentrated in a small number of sources and includes many unstarted activities. It is not a random sample of agency or forensic review work.
- Correct professional conclusions. Governing terms, delay responsibility and method selection require a qualified reviewer. Reproducing a computation does not prove its assumptions.
- Production readiness for your environment. Security requirements, installation, support and representative acceptance cases belong in the enterprise deployment scope.
- A percentage of clauses that will pass. Schedule quality, review coverage and numerical agreement are different measures.
Inspect the output, then evaluate the fit
The guided sample provides generated inputs and full engine reports. Passed, failed, referred and unevaluated clauses remain distinct. Every report retains its qualifications; computed dates and floats should be reconciled with the source scheduling application before reliance.
Review the limitations, explore verification, or request an enterprise walkthrough. Detailed historical measurements remain in the technical evidence reference.
Source: web/pages/is-this-real.md. Source commit date: 2026-09-10.