project_zeta_3_Z1_5_clean_2026_05_03

ζ.3 Spike Z1.5 — Incremental reasoning posture CLEAN RE-RUN — CLOSED 2026-05-03

Outcome: outcome-VALIDATED-WITH-METHODOLOGICAL-SUBSTITUTION (unchanged from Z1 v1.0; Konclude/RDFox still not directly tested)

Why Z1.5 existed: Rich-review of Z1 v1.0 at 2026-05-03T21:30 BST identified three methodological flaws:

1. default_world quadstore accumulation invalidated Scenario B HermiT isolation (onto_path.clear() clears search path only, not loaded ontologies)
2. Scenario C labelled “incremental” but was full re-classify-on-write (owlready2 0.50 exposes no true incremental API)
3. Phase-1.5+ ELK <10ms claim was unverified secondary evidence (ELK design docs, not direct measurement)

Z1.5 corrected (1) and (2); flaw (3) deferred to a separate Java-binding spike.

Why: Methodology flaws in empirical spike evidence can propagate incorrect framing into ζ-Q6 formal asking and Phase-1 architecture decisions. Z1.5 ensures Q-NU-001 rests on clean measurements.

How to apply: When referencing INHERIT v2 HermiT performance for ζ-Q6 or Acquirer-DD:

• Use Z1.5 T-file as primary substrate (not Z1 v1.0)
• Scenario B HermiT (5K flat synthetic): 0.845s mean (subprocess-isolated; new clean number)
• Scenario C re-classify-on-write (CCO 1431 DL + 1 ext): 2.611s mean, n=20 (confirmed within ±2% of Z1 v1.0)
• Phase-1.5+ ELK OWL API: throughput TBD — Java-binding spike required; do NOT quote <10ms

Key empirical findings

Scenario	Tool	Z1.5 result	Z1 v1.0	Delta
A — CCO 1431 DL	ELK CLI	0.753s (carried from Z1)	0.753s	0
A — CCO 1431 DL	HermiT	2.44-2.48s (carried)	2.44-2.48s	0
B — 5K flat	ELK CLI	0.563s (carried)	0.563s	0
B — 5K flat	HermiT (subprocess-isolated)	0.845s mean	POLLUTED — no valid number	NEW clean
C — CCO+1ext	HermiT re-classify-on-write	2.611s mean n=20	2.661s mean n=10	±2% — confirmed

Critical insight: axiom density dominates reasoning cost

HermiT on 5K flat SubClassOf Thing (0.845s) is ~3× faster than 1431 DL-complex CCO classes (2.44s). Class count is not the dominant factor — axiom complexity (transitivity, irreflexivity, property chains) is. Phase-1.5+ extrapolation must use the DL-complex baseline (2.44s), not the flat baseline (0.845s).

Artefacts

• T-file: ~/off-github/library/projects/inherit/T-spike-zeta-3-Z1-5-incremental-reasoning-posture-clean-2026-05-03.md (off-github commit 425b42f)
• Z1 v1.0 T-file: marked superseded_by: Z1.5 (off-github commit 0c11aec)
• Q-NU-001 cascade-Q: v0.2 → v0.3 (docs-strategy commit 010e709)
• Closure-bundle: /tmp/spike-zeta-3-Z1.5/closure-bundle.md

Supersedes

Memory file project_zeta_3_Z1_incremental_reasoning_posture_2026_05_03.md (Z1 v1.0 — superseded by this entry)