ζ-Q6 LOCKED σ.η-on-demand 2026-05-04T20:10 BST
A-150 amendment registered in arch-state §4 (v3.44 → v3.45). Twenty-fifth architectural amendment under refined-prompt-era ζ.2 numbering (A-126..A-150).
Lock decision
- σ.η-on-demand ★ — σ.γ ELK 0.6.0 OWL API Java incremental classification (3.78ms steady-state) for ALL interactive partner-firm T-box authoring + σ.α HermiT/owlready2 invoked ON-DEMAND ONLY when partner-firm explicitly triggers DL-completeness validation. NO nightly batch / NO scheduled cron.
- Rich’s framing 2026-05-04T20:00 BST: “I prefer on-demand ‘there and then’ rather than relying on nightly tasks.” Narrowed σ.η from “nightly OR on-demand” to “on-demand only”.
- Wsum 4.857 unweighted margin 0.214 over σ.δ paired second (4.643). 8/8 sensitivity perturbations favour σ.η-on-demand. NOT sensitivity-borderline at 0.20 robust threshold.
- σ.α-HermiT-only strong fallback under c8 +50% extreme-Phase-1-budget-pressure scenario.
Empirical evidence
ζ-Q6 spike (T-spike-zeta-q6-elk-owlapi-incremental-2026-05-04 at ~/off-github/library/projects/inherit/) commissioned by Rich during scorecard authoring + executed inline 2026-05-04T19:30-19:58 BST (~28 min):
| Scenario | Tool | Result | vs Z1.5 baseline |
|---|---|---|---|
| E (CCO+1-ax-incr, n=20 in-proc) | ELK OWL API Java | 3.78ms steady-state | 691× faster than HermiT 2611ms |
| E.1 (CCO+5-ax-incr, n=20) | ELK OWL API Java | 5.81ms steady-state | 449× faster |
| D (CCO 1431 initial, n=3 subprocess) | ELK OWL API Java | 308ms classify-only | 2.4× faster than ELK CLI |
| F (5K flat initial, n=3 subprocess) | ELK OWL API Java | 253ms classify-only | 2.2× faster than ELK CLI |
Z1 v1.0’s previously-unverified <10ms claim VALIDATED at 3.78ms (was conservative). Z1.5’s drop-the-claim discipline was correct given lack of direct measurement; underlying claim was accurate.
Architectural commitments
- NEW Phase-1: JVM-resident ELK sidecar service in IAS for interactive classify (£2K + 3-5d timeline)
- NEW Phase-1: “Validate DL-completeness now” UI affordance OR event-anchored on-publish/on-approval trigger (UX design decision; not architectural lock)
- CRITICAL CAVEAT: ELK is OWL 2 EL — INCOMPLETE on Q-003 ε.ε CCO/BFO DL T-box (irreflexivity / transitivity / nominals). σ.γ at 3.78ms cannot replace σ.α for DL-completeness; the σ.η-on-demand pairing makes EL-incremental + DL-validation co-exist correctly.
Reconsideration triggers preserved
- (T1) ELK 0.7+ release with improved DL handling
- (T2) Real partner-firm extension axiom-density measured at Phase-1.5 milestone → if >50ms incremental classify on real axioms, σ.δ paired sidecar may become preferred
- (T3) Faith-pillar wave-1 fires at Phase-1.5+ → revisit on-demand cadence
- (T4) IAS UX research reveals partner-firms regularly publish without validating → auto-fire σ.α on-publish (event-anchored variant within σ.η-on-demand spec)
- (T5) JVM-resident sidecar architecture proves operationally fragile → consider σ.α-only Phase-1 with σ.γ deferred Phase-1.5+ (Reversion-style cascade)
Methodological observations
- SECOND refined-prompt-era lock outside Group 1 strict-order Q-001..Q-004 batch (first was Q-005 κ.δ 2026-05-03).
- FIRST formal-asking Q where spike was commissioned by Rich during scorecard authoring + executed inline + immediately informed lock. Pattern:
feedback_test_theories_immediately_when_tabled(locked 2026-05-02) at substrate-architectural lock-point. - σ.η bold-synthesis option emerged FROM empirical evidence — original 6-option set σ.α-σ.ζ became σ.α + σ.β + σ.γ′ + σ.δ + σ.ε + σ.η post-spike (σ.ζ retired since spike-was-the-de-risking; σ.γ′ new ELK-only variant; σ.η NEW bold-synthesis). The empirical evidence didn’t just choose among existing options — it GENERATED a new dominant option.
- σ.η-on-demand narrowing reflects Rich’s emerging “minimise non-event-driven background work” preference — codify candidate:
feedback_user_event_anchored_over_scheduled_background_work.
Cascade artefacts
- arch-state v3.44 → v3.45 §4 changelog row landed
- Q-006 file at
answered-questions/Q-006-zeta-tbox-layering-incremental-reasoning-locked.md(renamed fromcurrent-questions/Q-NU-001-incremental-reasoning-posture.md) - INDEX.md v2.11 → v2.12 (asked 6→7, answered 5→6, placeholder 14→13)
- T-file at
~/off-github/library/projects/inherit/T-spike-zeta-q6-elk-owlapi-incremental-2026-05-04.mdv1.0 - Closure-bundle at
/tmp/spike-zeta-q6-elk-owlapi/closure-bundle.md
Phase 2 status post-lock
- 6 Qs answered (Q-001..Q-006)
- 1 Q current pre-formulated awaiting-natural-sequence (Q-015)
- 8 Qs missing in re-ask sequence (ζ-Q7..ζ-Q14)
- 18-30 Q budget; 6/18-30 = 20-33% complete
NEXT
ask ζ-Q7 cascade-Q file under refined-prompt v3.9. Probable topic per Phase-2-eligible queue: vocabulary management (Q-NU-002 — Z2 spike pre-validated; flat SKOS Phase-1) OR catalogue IRI registry pattern (Q-NU-006 — ν.β C1 pre-validated; affects every cross-module reference).