Spike outcome
outcome-VALIDATED — kill condition NOT MET on all criteria.
Key empirical findings
| Scenario | Classes | HermiT mean | stdev | n |
|---|---|---|---|---|
| T1: 9 i-ζ + BFO stubs | 19 | 0.4615s | 0.1650s | 5 |
| T2: +27 cross-module | 46 | 0.3163s | 0.0067s | 5 |
| T3+10 ext | 56 | 0.3657s | 0.0299s | 5 |
| T3+20 ext | 66 | 0.3873s | 0.0444s | 5 |
| T3+30 ext | 76 | 0.3531s | 0.0122s | 5 |
| T3+40 ext | 86 | 0.3740s | 0.0076s | 5 |
| T3+50 ext | 96 | 0.4284s | 0.0548s | 5 |
| Per-extension marginal cost | — | 0.0016s/ext | — | — |
Kill condition: >5s classify OR >0.5s/ext. Actual: 0.428s max / 0.0016s/ext. NOT MET by 12× / 312× margin.
Cross-cutting findings
- Axiom complexity dominates (confirmed from Z1.5): BFO+CCO-stub fixture 0.316-0.428s vs CCO-full-import 2.44s. Real Phase-1 will be between these.
- Flat scaling across 9-96 classes — no monotonic growth; JVM scheduling noise dominates variance.
- Zero consistency errors in 35 trials — φ.α’ 3-layer hierarchy is OWL-DL-consistent.
- ELK not benchmarked — standalone jar not installed; ELK Phase-1.5+ Java-binding spike queued.
- T1 cold-start variance (stdev 0.165s) — JVM startup ~0.3s included in wall-clock; warm path ~0.30-0.39s.
- Multiple inheritance confirmed classifiable — FinancialAsset + DigitalAsset both subclass InheritableAsset + InformationContentEntity (GDC) — HermiT handles OWL DL multiple inheritance without issue.
Phase-5 lock recommendation
φ.α’+ CORE — retain φ.α’ 3-layer; HermiT via owlready2 Phase-1; ELK Phase-1.5+.
Artefact paths
- T-file:
/home/richardd/off-github/library/projects/inherit/T-spike-nu-beta-D1-tbox-layering-r2-completeness-2026-05-04.md - Q-NU-010:
/home/richardd/testatetech/docs-strategy/docs/superpowers/specs/2026-04-29-multi-phase-audit/current-questions/Q-NU-010-tbox-layering-r2-completeness-deepening.md - Closure bundle:
/tmp/spike-nu-beta-D1/closure-bundle.md
Honesty caveats summary (top 3)
- ELK standalone not benchmarked — carried from Z1.5; jar not installed.
- T2/T3 isolation via ontologies.clear() (not full subprocess per-trial) — adequate for flat BFO-stub fixtures.
- D1 fixture uses BFO/CCO stubs not full imports — real Phase-1 with full imports expected 0.5-2.44s (still below 5s kill).