Discovery / forward-motor separation
Offline relation discovery can be separated from a small exact online motor with explicit provenance.
Diagnostic work, exact algebra and local capabilities are not treated as end-to-end mining advantage.
What was tested?
Offline relation discovery can be separated from a small exact online motor with explicit provenance.
Why the test is meaningful
Offline discovery can search a large relation space without pretending that this search is free at mining time. A valid architecture freezes discovered formulas, their domains and provenance; the online motor then performs only declared checks and evaluation.
offline: data → (formula,domain,proof metadata)online: (context,x) → verify precondition → evaluate formulaW_total = W_setup/amortization + W_dispatch + W_online + W_fallbackHow it was tested
Split search artifacts from executable contracts, require the online side to consume only frozen capabilities and audit each dispatch.
What happened
The separation was implemented across seven variants with explicit discovery-only and forward-runtime roles.
Exactness and statistical controls
Seven variants enforced separate artifact roles and audited each online dispatch against direct evaluation. The experiment validates architectural separation, not a favorable value of W_total.
What the result means
Architectural separation prevents discovery cost from being hidden, but total amortization is still unmeasured.
Limitations
- Amortization across jobs is unknown.
- Artifact size and cache behavior are unmeasured.
- An exact online motor may still cost more than the baseline.
Evidence trail
Frozen capability artifacts are independently consumable by the forward verifier.
Canonical variants
SUBENGINE-V12ASUBENGINE-V12BSUBENGINE-V12CSUBENGINE-V12DSUBENGINE-V12ESUBENGINE-V12FSUBENGINE-V12GSource: internally audited canonical reports. Local filesystem structure, private headers and operational identifiers are excluded from publication.