wick total recall · checkpoint · 2026-07-02 · sessions 07-01 → 07-02
The Engine Works. The Record Is Clean.
Two-day arc: the complete recall engine solved, scaled, hardened, and honestly scoped — 15 experiments, zero false entries, 5 harness self-catches, everything patent-sealed in 4 locations.
The Complete Engine (all measured)
anchor-set recall
1068×
graph-local BEAM field @ N=1M (1.27ms vs 1361ms), top-1 recovery 100%. 412× @ 19k dense at 99.8% clarity. Scaling law: O(beam) vs O(N·edges) — grows with N.
item↔item wormholes
30–164×
hypervector + E8 lattice tunnels, measured wall-clock @ 0.95+ retention. E8 cells best ≥50× (0.754 @150× vs hash 0.295). Beam DIFFERENTIATED from ACL-push (CI-backed).
durability (3 lanes)
40%
coordinate erasure survived UNCODED (intrinsic redundancy, tunnel-retention 0.96 flat) · shard loss → RS any-k · tamper → two-plane seal.
accuracy lever
0.924
PPR-field/deg^0.5 (from 0.858 plateau). α holds out (0.9147 test). Even-compute test: the FIELD OPERATOR is the win (+0.69), not compute.
Laws Discovered This Arc
| law | statement | status |
|---|
| Task-Split | Value-reconstruction of the field ≈ random at rank-256; tunnels serve item↔item only; set-recall needs the (beam-local) exact field | BANKED |
| Color = Regime Axis | Rare targets live in BLUE/fine spectrum (13×), strong ties in RED/coarse; every construction splits coarse-vs-fine regime | BANKED |
| Conditional Independence | LSH independence is conditional-on-similarity; claim suppression at the operating regime via measured curves (1541×) | REFRAMED + BANKED |
| Role-Specificity | BIBD wins the codebook role, loses the tunnel role — combinatorial wins don't transplant across roles | BANKED |
| Intrinsic Damage-Tolerance | Spectral signatures survive 40% coordinate loss uncoded — spread-spectrum redundancy by construction | BANKED (led by tunnel-retention evidence) |
Honest Ledger (the discipline record)
| item | outcome |
|---|
| Beam vs ACL forward-push | WIN differentiated: 0.925 @114 nodes vs 0.914 @286 (paired CI excl 0) |
| Magnetic-Laplacian phase | PARKED null + mechanism: phase needs CYCLES; near-tree DAGs have none |
| Hyperbolic route-by-class | NULL/DATA-ABSENT wrong tree (file-paths); CXDB parent-tree unpopulated; needs a populated semantic hierarchy |
| Torus cyclic rescue | AWAITING CORPUS harness invalid on access-trace (no rhythm variance); retest on the 2,859-record daily-rhythm corpus |
| Erasure-coding-for-recall | QUARANTINED not needed — intrinsic redundancy covers coordinate damage (tested twice) |
| 5 harness self-catches | tie-inflation ×2 · mask-poisoned normalization ×2 · insensitive control ×2 — all caught before contaminating the record |
Queue (next up)
| # | item | state |
|---|
| 1 | Hyperbolic-capacity nesting (last §8 zero-to-one; sub-dimension nesting for unlimited capacity) | NEXT — buildable now |
| 2 | Torus on the daily-rhythm corpus | needs corpus path (revive) |
| 3 | Real-1M beam re-check on live B3 data | B3 accruing |
| 4 | C/SIMD implementation measurement plan (production) | spec committed (revive) |
Preservation state: SESSION_2026_07_01 + 07-02 patent-evidence docs (with all addenda) · PATENT_MASTER §7.5–7.6 · ATTORNEY_PACKAGE + tested-and-declined appendix (revive) · git-clean-proof backups (17 scripts + 17 results) · CXDB snapshots through cxdb-20260702-1615 · memory pointers current · revive's ledger synchronized in near-real-time. Nothing depends on any single session surviving.