Skip to content

CLAIMS_LEDGER.md

Latest commit

 

History

History
746 lines (615 loc) · 30.5 KB

CLAIMS_LEDGER.md

File metadata and controls

746 lines (615 loc) · 30.5 KB
Raw
status in_progress
created 2026-05-22
last_updated 2026-07-20
phase A

GeniusLocusKit Claims Ledger

The flat ledger of every empirical claim the GLK substrate makes. Each row is one claim with its spec citation, type, current status, and evidence pointer. The ledger drives Phases B through F: invariant negative tests close the invariant rows, theorem demonstrations close the theorem rows, the migration benchmark closes the application-validation rows, and the adversarial + perf gates close the recurring-regression rows.

Status vocabulary per row:

  • evidenced: a runnable test or demonstration exists and passes; the commit SHA, test path, or runnable demonstration path is recorded under "Evidence".
  • pending: the claim has no current evidence on file; the reason and what is needed are recorded under "Evidence".
  • waived: the claim is intentionally deferred or scoped out of v1; the rationale is recorded under "Evidence".
  • superseded: an earlier invariant or theorem that a later one replaces. Listed for historical reference; the active successor carries the evidence.

Population status by category

Category Total Populated Phase
Invariants (I-1 through I-30; I-24 split a/b) 31 31 Evidence-linking pass 2026-07-20: 19 evidenced, 7 pending, 1 partial, 2 waived, 2 superseded
Theorems (named results, section 14) 7 7 Evidence-linking pass 2026-07-20: 5 evidenced, 1 pending (T6), 1 waived (T7/T8)
Hot-path budgets (cookbook section 17.1) 3+ 3 A second batch done
Verb-surface protocol contracts 9 9 A second batch done
Manifest required keys 18+5 0 A pending
Framework profile contracts TBD 0 A pending
Standing-signals scheduler contracts TBD 0 A pending
Decision-record claims TBD 0 A pending

Phase A population continues.

Invariants

The constitutional core. I-1 through I-14 are inherited from the architecture spec section 3. I-15 through I-24 extend it. Two original invariants (I-5 and I-9) are explicitly superseded by I-16 and I-15 respectively; they are listed below as superseded rows for historical traceability.

I-1: Verbatim rung is immutable

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 68
  • Type: invariant
  • Status: active in v1.0
  • Evidence: pending. Corrections-produce-new-versions is covered by the LocusKit lineage cascade suite (packages/kits/LocusKit/Tests/LocusKitTests/LineageTests.swift), and DrawerStore exposes no rung-1 content-mutation API. Missing: a negative test that attempts mutation of a rung-1 column post-capture and asserts failure.

I-2: All bitmap mutations write audit rows

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 70
  • Type: invariant (formalized by I-20 as G-Set CRDT)
  • Status: active in v1.0; co-active with I-20
  • Evidence: evidenced. mutateAdjective and mutateOperational each write exactly one audit row with full fields, and write no row when the mutation fails; mutateProvenance writes its audit row atomically and rolls back on failure. All three bitmap columns are covered.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/BitmapAuditTests.swift; packages/kits/LocusKit/Tests/LocusKitTests/ProvenanceTests.swift; packages/kits/GeniusLocusKit/Tests/GeniusLocusKitTests/UnifiedAuditLogTests.swift.

I-3: The substrate accepts no secret + public combination

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 72
  • Type: invariant (subsumed by I-22's forbidden combinations enumeration)
  • Status: active in v1.0; co-active with I-22
  • Evidence: evidenced. Both required surfaces are on file. Verb layer: the validator throws on secret + exportable, and both capture and mutateAdjective reject the forbidden bitmap, leaving prior state and the audit table untouched. Maintenance scan: the standing maintenance signal emits the forbidden-combo proposal.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/ForbiddenCombinationTests.swift; packages/kits/GeniusLocusKit/Tests/GeniusLocusKitTests/StandingSignalsTests.swift (maintenanceSignalEmitsForbiddenComboPlusCandidateAndDiagnostic).

I-4: Every model-generated rung carries model_id and model_version

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 74
  • Type: invariant
  • Status: active in v1.0
  • Evidence: pending. modelID and modelVersion are non-optional parameters of the SynapseKit lane-store write surface (exercised throughout packages/kits/SynapseKit/Tests/SynapseKitTests/FloatLaneStoreTests.swift), so the absent case is structurally unrepresentable there. Missing: an explicit negative test asserting a rung 2/3/4 write without model identification fails.

I-5: Every drawer carries a lattice anchor (superseded)

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 76
  • Type: invariant
  • Status: superseded by I-16 (universal lattice anchor extends this from drawer-only to all noun types)
  • Evidence: historical; the active version's evidence is under I-16.

I-6: Audit rows can be tombstoned but not modified

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 78
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced. Expunge tombstones with content zeroing and an appended audit event; audit-grade (accepted) rows are refused and survive intact; withdraw and expunge are sticky tombstones in the unified log fold.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/ExpungeTests.swift; packages/kits/GeniusLocusKit/Tests/GeniusLocusKitTests/UnifiedAuditLogTests.swift (withdrawAndExpungeAreStickyTombstones).

I-7: The verb count is fixed at nine

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 80
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced. The lexicon suite asserts Verb.allCases.count == 9, the canonical declaration order of the nine verbs, and the 6/2/1 flow partition; the Rust leg mirrors it (verb_count_is_nine).
  • Evidence pointer: packages/libs/AriaLexiconLib/Tests/AriaLexiconLibTests/VerbTests.swift.

I-8: The adjective category count is fixed at four

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 82
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced. The lexicon suite asserts exactly the four named categories (state, trust, sensitivity, exportability); the Rust leg mirrors it (adjective_count_is_four).
  • Evidence pointer: packages/libs/AriaLexiconLib/Tests/AriaLexiconLibTests/AdjectiveTests.swift.

I-9: Field width is 4 bits in the bitmap tier (superseded)

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 84
  • Type: invariant
  • Status: superseded by I-15 (6-bit field-width floor)
  • Evidence: historical; the active version's evidence is under I-15.

I-10: Every bitmap field reserves at least 30% growth space

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 86
  • Type: invariant
  • Status: active in v1.0
  • Evidence: pending. No growth-space check is on file. Missing: a unit test over the manifest's bitmap layout asserting each field's used-values count is at most 70% of its capacity.

I-11: Bitmap layouts within a published version are stable

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 88
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced. A fresh estate populates bitmap_layout_version with the published v1 default, and a database carrying an unknown layout version refuses to open with EstateError.manifestMismatch — layouts are pinned to the published manifest version.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/ManifestTests.swift; packages/kits/LocusKit/Tests/LocusKitTests/EstateTests.swift.

I-12: The substrate provides storage; applications do not bring their own

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 90
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced by composition. GLK depends on LocusKit, SynapseKit, and CorpusKit per the kit-composition test surfaces; no external storage adapter exists.
  • Evidence pointer: packages/kits/GeniusLocusKit/Tests/GeniusLocusKitTests/CompositionConformanceTests.swift.

I-13: Federation is not a substrate concern

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 92
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced by absence. The substrate kit graph contains no network or RPC surface. Confirmed by inspection; a structural test enumerating GLK's exported symbols and asserting none reach outside the local kit graph would formalize this.

I-14: Provenance is set at write and mutated only via the audit-recorded path

  • Spec: docs/reference/GENIUSLOCUS_ARCHITECTURE_SPEC.md line 94
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced. Propose sets provenance at write time (non-default provenance round-trips; each provenance value writes to its own window); mutateProvenance flips bits and writes its audit row atomically, rolling back both tables on failure — the audit-recorded mutation path.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/ProposeProvenanceTests.swift; packages/kits/LocusKit/Tests/LocusKitTests/ProvenanceTests.swift.

I-15: 6-bit field-width floor

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 295 (§2.2)
  • Type: invariant (supersedes I-9)
  • Status: active in v1.0
  • Evidence: evidenced. The bitmap conformance suites pin the 6-bit field positions (bits 0-5, 6-11, ...) to the cookbook layout for the adjective and operational bitmaps, and the kernel BitField suite exercises 6-bit field extraction.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/AdjectiveBitmapConformanceTests.swift; packages/kits/LocusKit/Tests/LocusKitTests/OperationalBitmapConformanceTests.swift; packages/libs/SubstrateKernel/Tests/SubstrateKernelTests/BitFieldTests.swift.

I-16: Every row has a lattice anchor

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 580 (§2.7)
  • Type: invariant (supersedes I-5, extends drawer-only to universal)
  • Status: active in v1.0
  • Evidence: pending. Anchor persistence is covered for drawers (packages/kits/LocusKit/Tests/LocusKitTests/LatticeAnchorTests.swift, all four anchor fields) and associations (packages/kits/LocusKit/Tests/LocusKitTests/AssociationTests.swift, four-field anchor round-trip). Missing: schema-level presence assertions for the remaining noun row types (tunnels, KG facts, diary entries), or one structural test across all noun schemas.

I-17: Cross-noun fingerprint compatibility

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 1075 (§3.8)
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced. The four hyperplane families are independent per block with fixed canonical widths, fingerprints share the 256-bit four-block wire shape, and Hamming distance is exercised over the shared format (same surface as theorem T2).
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/DrawerFingerprintTests.swift; packages/libs/SubstrateTypes/Tests/SubstrateTypesTests/HammingTests.swift; packages/libs/SubstrateTypes/Tests/SubstrateTypesTests/HyperplaneFamilyTests.swift.

I-18: Runtime view is bit-sliced 3D tensor

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 1165 (§4.1)
  • Type: invariant (constitutional working-set layout)
  • Status: active in v1.0
  • Evidence: evidenced. The 3D bit tensor round-trips 6-bit values, addresses each bit position independently, scans field-equals matches exactly, grows capacity while preserving data, and sizes as six bit-slices.
  • Evidence pointer: packages/libs/SubstrateTypes/Tests/SubstrateTypesTests/ThreeDBitTensorTests.swift.

I-19: Kernel layer is portable across SIMD families

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 1234 (§4.4)
  • Type: invariant
  • Status: active in v1.0
  • Evidence: evidenced by the four-cell conformance gate and the cross-language byte-equality test for all 28 conformant primitives.
  • Evidence pointer: the four-way conformance matrix.

I-20: Audit log is a Grow-Only Set CRDT

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 1323 (§5.1)
  • Type: invariant (formalizes I-2 with CRDT properties)
  • Status: active in v1.0; co-active with I-2
  • Evidence: evidenced. The G-Set law surface is tested directly: add is idempotent (content-hash dedupe), merge is commutative and idempotent, and entries are HLC-ordered; the unified log's projection is independent of arrival order and converges across tiers.
  • Evidence pointer: packages/libs/SubstrateTypes/Tests/SubstrateTypesTests/GSetAuditLogTests.swift; packages/kits/GeniusLocusKit/Tests/GeniusLocusKitTests/UnifiedAuditLogTests.swift.

I-21: Sync convergence

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 1405 (§5.4)
  • Type: invariant
  • Status: active in v1.0
  • Evidence: waived. Sync is a 1.1.x feature under active development on develop/1.1.x; convergence evidence rides that work.

I-22: Forbidden state-combination invariants

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 2494 (§9.5)
  • Type: invariant (enumeration of forbidden combinations including the original I-3)
  • Status: active in v1.0
  • Evidence: partial. The secret + public case (subsuming the original I-3) is covered by packages/kits/LocusKit/Tests/LocusKitTests/ForbiddenCombinationTests.swift. The cookbook §9.5 enumeration remains to be extracted in a subsequent Phase A pass.

I-23: Pairing algebra (binary arity in v1)

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 3037 (§12.1)
  • Type: invariant
  • Status: active in v1.0
  • Evidence: pending. The arity axis is exercised — arity decodes bits 18-19 and reserved raw values fall back to .binary (packages/kits/LocusKit/Tests/LocusKitTests/AssociationTests.swift). Missing: an explicit negative test asserting the write surface rejects an n-ary association rather than coercing it.

I-24a: Randomized-response fingerprint noising with calibrated (ε, δ) parameters

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 3254 (§12.6)
  • Type: invariant (split from I-24, 2026-07-20)
  • Status: active in v1.0
  • Evidence: evidenced. The DP primitive is implemented — DPParameters construction and seeded noise in packages/kits/NeuronKit/rust/src/mind_overlap.rs — and the noise path is tested: k-anonymity threshold survival and suppression, and zero-mean, seed-deterministic Laplace noise.
  • Evidence pointer: packages/kits/NeuronKit/rust/src/mind_overlap.rs; packages/libs/SubstrateML/Tests/SubstrateMLTests/DPORReductionTests.swift.

I-24b: Tier-boundary application with budget tracking

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md line 3254 (§12.6)
  • Type: invariant (split from I-24, 2026-07-20)
  • Status: deferred to the 1.2.x federation series
  • Evidence: waived. Tier hierarchy is 1.2.x federation; same disposition as I-21.

I-25: One implementation per atomic

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md §1.4; docs/concepts/MOOTX01_AND_ARIA_CANON.md ("the substrate owns the atomics")
  • Type: invariant (2026-05-28)
  • Status: active in v1.0
  • Evidence: evidenced. The Swift/Rust concordance discipline is enforced mechanically: every public type in each package's Swift and Rust legs must appear in its interface doc's Swift/Rust Concordance table, checked by the strict concordance gate (run in CI and via make concordance); cross-language agreement over the atomics is exercised by the conformance suites.
  • Evidence pointer: packages/scripts/concordance_audit/concordance_audit.py; packages/libs/SubstrateLib/Tests/SubstrateLibConformanceTests/.

I-26: Capture writes a sealed genesis event

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md#53-projection-rules
  • Type: invariant (2026-05-28; retired the bitmap_audit / provenance_audit tables)
  • Status: active in v1.0
  • Evidence: evidenced. The named positive test is on file: auditTrail(rowID:) after capture returns the genesis capture event, HLC-ordered ahead of subsequent events; the gate itself rejects undeclared fields, illegal values, over-width values, and illegal transitions.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/AuditAPITests.swift; packages/libs/SubstrateLib/Tests/SubstrateLibTests/AuditGateTests.swift.

I-27: Single HLC maker (clock triangle)

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md#56-hlc-generator-placement-v10
  • Type: invariant (2026-05-28)
  • Status: active in v1.0
  • Evidence: pending. Generator semantics (monotonic send, receive advancement past remote timestamps, node-ID tiebreak) are covered by packages/libs/SubstrateTypes/Tests/SubstrateTypesTests/HLCTests.swift, and DrawerStore injects-or-makes a single generator keyed to estate identity. Missing: a test asserting one generator per estate and one send() per write.

I-28: Integrity triangle / custody mode

  • Spec: docs/engineering/GENIUSLOCUS_ENGINEERING_COOKBOOK.md#56-hlc-generator-placement-v10
  • Type: invariant (2026-05-28; rejected the dual-clock model)
  • Status: active in v1.0
  • Evidence: evidenced. The first-open identity regression suite is on file: a valid persisted estate_uuid derives the correct non-zero node id, an absent value opens a fresh estate, and a corrupt value fails loud on open rather than masking as node 0 or a random UUID; audit events are HLC-ordered per row.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/DrawerStoreManifestUuidTests.swift; packages/kits/LocusKit/Tests/LocusKitTests/AuditAPITests.swift.

I-29: Row identity is a UUID

  • Spec: docs/engineering/SYSTEM_ENGINEERING_REFERENCE.md#31-uuid-identity-and-lineage
  • Type: invariant (2026-05-28)
  • Status: active in v1.0
  • Evidence: pending. requireUuid is implemented on the gated write paths (packages/kits/LocusKit/Sources/LocusKit/DrawerStore.swift). Missing: a test asserting the throw on a non-UUID row id and identity stability across configurations.

I-30: Substrate ships as four packages

  • Spec: docs/engineering/HARNESS_REFERENCE.md#6-the-four-package-substrate-split
    • the 2026-05-29 addendum (SubstrateLib retained as the orchestration package)
  • Type: invariant (2026-05-28, restated 2026-05-29)
  • Status: active in v1.0
  • Evidence: evidenced. The four-package split shipped (commit 831dacbc): SubstrateTypes, SubstrateKernel, SubstrateML, and SubstrateLib exist under packages/libs/, and SubstrateLib's manifest declares dependencies on the other three (orchestration role).
  • Evidence pointer: commit 831dacbc; packages/libs/SubstrateLib/Package.swift.

Theorems

The canonical statements live in the substrate mathematics reference, which restates the named results the substrate claims and ties each to the demonstrating kit and conformance suite. The full formal statements live in the published paper section 13 and Appendix C. The theorem demonstrations covered Theorems 4, 6, 7, and 8 by their paper-numbering, and met Theorem 5; the named-result restatement uses descriptive names rather than the paper numbers. Reconciling the paper-numbered and named-result lists requires a direct read of paper section 13 and Appendix C, which Phase A defers to a focused later pass.

T1: Sync convergence (I-21)

  • Spec: substrate mathematics reference, named results; paper Theorem 1
  • Type: theorem
  • Status: active in v1.0
  • Evidence: evidenced. Replicas holding equal audit sets project equal state. Demonstrated by the ConvergenceKit convergence tests.
  • Evidence pointer: packages/libs/SubstrateTypes/Tests/SubstrateTypesTests/GSetAuditLogTests.swift (commutative, idempotent merge); packages/kits/GeniusLocusKit/Tests/GeniusLocusKitTests/UnifiedAuditLogTests.swift (projectionIndependentOfArrivalOrder, crossTierConvergence).

T2: Cross-noun Hamming well-definedness (I-17)

  • Spec: substrate mathematics reference, named results
  • Type: theorem
  • Status: active in v1.0
  • Evidence: evidenced by SubstrateLib and LocusKit fingerprint tests. Uniform four-block widths and deterministic null sub-hashes make d_H a well-defined distance across all noun types.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/DrawerFingerprintTests.swift; packages/libs/SubstrateTypes/Tests/SubstrateTypesTests/HammingTests.swift; packages/libs/SubstrateTypes/Tests/SubstrateTypesTests/HyperplaneFamilyTests.swift.

T3: Pruning equivalence

  • Spec: substrate mathematics reference, named results; cookbook section 5
  • Type: theorem
  • Status: active in v1.0
  • Evidence: evidenced by LocusKit recall tests. Container-pruned recall returns the same result as a full scan; OR-pruning skips a container only when no row could match.
  • Evidence pointer: packages/kits/LocusKit/Tests/LocusKitTests/RecallPruningTests.swift (recall prunes a non-matching container and returns the equivalent rows; containerSurvives prunes only when a required bit is absent).

T4: Automaton safety, liveness, and reachability

  • Spec: substrate mathematics reference, named results; cookbook section 13
  • Type: theorem (paper Thm 4)
  • Status: active in v1.0
  • Evidence: evidenced. Demonstrated by the LocusKit DrawerStateValidator suite and the theorem demonstrations.
  • Evidence pointer: LocusKit DrawerStateValidator; GeniusLocusKit/Tests/GeniusLocusKitTests/TheoremsTests.swift.

T5: Performance budget (paper Theorem 5)

  • Spec: substrate mathematics reference, named results; paper Theorem 5
  • Type: theorem (performance gate)
  • Status: active in v1.0; the load-bearing performance budget
  • Evidence: evidenced and met. P99 capture under 100 ms on the iPhone profile and enrichment at least 60 drawers per hour on the Mac profile.
  • Evidence pointer: GeniusLocusKit/Tests/GeniusLocusKitTests/PerformanceGateTests.swift; the perf gate reports P99 capture 0.262 ms (well under the 100 ms budget) and enrichment far above the 60 drawers per hour floor.

T6: Pairing non-transitivity (I-23)

  • Spec: substrate mathematics reference, named results; cookbook section 11
  • Type: theorem
  • Status: active in v1.0
  • Evidence: pending. Pairing enforcement exists at the sync boundary — pull rejects a signed envelope from a sender that is not a paired peer (packages/kits/ConvergenceKit/Tests/ConvergenceKitFederationTests/FederationPairingTests.swift). Missing: the negative test refusing a fingerprint comparison across estates without a shared pairing seed.

T7 and T8: Paper-numbered demonstrations

  • Spec: paper section 13 and Appendix C
  • Type: theorem (paper numbering 7 and 8; named-result mapping pending)
  • Status: active in v1.0
  • Evidence: waived. Source paper is Enterprise edition material; demonstrations exist, numbering alignment is an EE concern.

Hot-path budgets

The cookbook section 17.1 enumerates the hot-path budgets the substrate must meet. Three are pinned as load-bearing.

HP1: Capture P99 latency under 100 milliseconds on iPhone

  • Spec: paper Theorem 5; cookbook section 17.1
  • Type: hot-path budget
  • Status: active in v1.0; met
  • Evidence: evidenced. The perf gate measured 0.262 ms P99 capture, three orders of magnitude under the 100 ms budget on iPhone profile.
  • Evidence pointer: GeniusLocusKit/Tests/GeniusLocusKitTests/PerformanceGateTests.swift.

HP2: Enrichment rate at least 60 drawers per hour on Mac

  • Spec: paper Theorem 5; cookbook section 17.1
  • Type: hot-path budget
  • Status: active in v1.0; met
  • Evidence: evidenced. The perf gate measured enrichment rate far above the 60 drawers per hour floor on Mac profile.
  • Evidence pointer: GeniusLocusKit/Tests/GeniusLocusKitTests/PerformanceGateTests.swift.

HP3: Hamming top-K (K=10 over 1M rows) under 100 microseconds

  • Spec: cookbook section 17.1; cookbook section 17.5 bandwidth floor analysis
  • Type: hot-path budget
  • Status: active in v1.0; unmet, aspirational
  • Evidence: partial. The section 17.5 bandwidth floor on the Apple M-series profile is 533 microseconds. Best measured kernel today is Swift SimdKernel at 594 microseconds (11% above floor) and Rust SimdKernel at 631 microseconds (18% above floor); see the baseline in docs/validation/substrate_math_performance/README.md. The 100 microsecond budget is unmet by any kernel measured; section 17.5 analysis suggests the memory bandwidth floor is the binding constraint, not the algorithm. Future work would require either reducing bytes read (bit-slice layout per I-18, deferred) or hardware beyond the M-series memory subsystem.
  • Evidence pointer: the substrate-math performance baseline table.

Protocol contracts: the nine-verb surface

Invariant I-7 fixes the verb count at nine. The verbs are defined in the cookbook section 10 and implemented in the GeniusLocusKit verb surface. Each verb has signature, preconditions, and postconditions that conformance tests must verify.

V1: capture

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests.
  • Evidence pointer: GeniusLocusKit/Sources/GeniusLocusKit/Verbs/VerbSurface.swift; specific test path to confirm.

V2: recall

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests; recall is the canonical single-estate recall over the composed substrate.
  • Evidence pointer: same as V1.

V3: mutate

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests; mutate writes an audit row per I-2 and I-20.
  • Evidence pointer: same as V1.

V4: withdraw

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests; withdraw retracts a proposed-but-not-confirmed item.
  • Evidence pointer: same as V1.

V5: expunge

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests; expunge tombstones per I-6 while preserving the fact-of-expunge.
  • Evidence pointer: same as V1.

V6: reanchor

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests; reanchor updates a lattice anchor with audit row recording the change.
  • Evidence pointer: same as V1.

V7: learn

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests; learn drives matrix-tier update via the autonomic governor.
  • Evidence pointer: same as V1.

V8: propose

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests; propose is the emission path for the six standing signals.
  • Evidence pointer: same as V1, plus the standing-signals tests.

V9: associate

  • Spec: cookbook section 10
  • Type: protocol contract (verb)
  • Status: active in v1.0
  • Evidence: evidenced by the verb-surface tests; associate creates a binary association per I-23.
  • Evidence pointer: same as V1.

Categories still pending in Phase A

The following remain queued for the next focused Phase A session. They are all bounded sets and each is enumerable from a single source document.

Manifest required keys (18 required + 5 optional)

The manifest has 18 required keys and 5 optional keys, with bitmap_layout_version and provenance_bitmap_version making the manifest self-describing. The cookbook should enumerate the full list; Phase A captures each as one ledger row with a schema-validation evidence pointer.

Framework profile contracts

The framework profile declares lattice citation standard, zoom window, vocabulary loaded at estate creation, proposition vocabulary predicates for rung 1.5 KG, jsonpath to predicate mappings over blob tier content, canonical reference dependencies with activation triggers and learn calls, and learn mode per reference dependency (byReference or byIngestion). Profile evolution commits to retroactive remining sweeps. Each declared contract becomes one ledger row.

Standing-signals scheduler contracts

The scheduler runs all six standing signals under a single serial dispatch. The contracts: exactly-once dispatch, ordering preservation under the configured policy, register / status / subscribe API surface, and the six signals (dreaming, maintenance, vector-similarity, decay-sweep, byReference-validity, end-of-day-tournament) each emitting via propose without mutating state.

Consolidated engineering claims

The retired 1.0 design-record set contained 59 records. Its current behavioral claims, constraints, and empirical selections now live in the engineering masters linked by this ledger. Historical deliberation is not a live specification and is not used as an evidence pointer.

Pending evidence (gap list)

This section will accumulate the full list of pending rows as Phase A progresses. Once Phase A exits, the list drives Phase B (invariants) and Phase C (theorems).

Current count after the 2026-07-20 evidence-linking pass:

  • 7 pending invariant rows (I-1, I-4, I-10, I-16, I-23, I-27, I-29); 19 evidenced (I-2, I-3, I-6, I-7, I-8, I-11, I-12, I-13, I-14, I-15, I-17, I-18, I-19, I-20, I-24a, I-25, I-26, I-28, I-30); 1 partial (I-22 enumeration to complete); 2 waived (I-21, I-24b); 2 superseded historical (I-5, I-9).
  • 1 pending theorem row (T6 pairing non-transitivity); 5 evidenced (T1 through T5); 1 waived (T7-T8, EE paper numbering).
  • 0 pending hot-path budgets; 2 met (HP1 capture, HP2 enrichment); 1 unmet aspirational (HP3 hamming top-K, against the memory bandwidth floor).
  • 0 pending verb-surface rows; 9 evidenced (V1 through V9).

Total evidenced after the pass: 35 (19 invariants + 5 theorems + 2 hot-path budgets + 9 verbs). Pending: 8 (was 24 by row count: 16 rows flipped to evidenced on existing tests or shipped structure, I-21 was waived by the 2026-07-20 rulings, and the I-24 split produced one evidenced and one waived row). Met-with-budget-known: 3 (HP1 met, HP2 met, HP3 unmet but bounded).

Protocol contracts, manifest required keys, framework profile contracts, standing-signals scheduler contracts, and the 27 decision records still queued for a focused Phase A session.