Shay
b40422e9db
perf(rust): versor_apply f64 parity port — 29x over Python, bit-identical
...
Closes the last open Rust parity gate from ADR-0020.
Kernel: new versor_apply_closed_f64 in core-rs/src/versor.rs performs
the full sandwich V·F·rev(V) + closure in f64, mirroring Python's
algebra.versor.versor_apply + _close_applied_versor exactly:
- no null-vector early branch (Python doesn't have one)
- unitize_versor with dense-support seed fallback gate
- post-unitize versor_condition < 1e-6 recheck
- seed_to_rotor on failure, passthrough as last resort
PyO3 binding: versor_apply_with_closure_f64 accepts/returns float64
arrays through new extract_f64_slice / f64_array_to_numpy helpers.
algebra/backend.py::versor_apply routes through it under CORE_BACKEND=rust.
Parity gate re-enabled (was skipped pending this port). 8/8 bit-
identical across normalized hot-path + identity-versor cases.
Bench (5000 iters, runtime hot path):
python: 213.0 us/call
rust: 7.4 us/call → 28.8x speedup
All lanes green: algebra 132 (was 124+8skip), smoke 54, runtime 19,
cognition 57, teaching 17, packs 6. Cognition eval 100% across all metrics.
PROGRESS.md updated: versor_apply marked passing; Phase 5 Rust parity
track now 5/5 surfaces gated and enabled.
2026-05-16 20:43:01 -07:00
Shay
e36998d25d
perf(rust): zero-copy vault_recall — Rust beats Python at scale
...
ADR-0020 follow-on (task #35 ). Two-pronged fix:
1. Kernel: ported ADR-0019 Stage 1 diagonal-metric kernel to
core-rs/src/vault.rs. Per-versor scoring is now 32 multiplies
+ 32 adds via the precomputed Cl(4,1) metric, not the
1024-op full geometric_product the prior path computed.
Bit-identity preserved by serial fold order matching Python.
2. Zero-copy marshalling: replaced Vec<&PyAny> + extract-per-
versor with PyReadonlyArray2<f32> via the numpy Rust crate.
The Rust binding now reads a slice view directly into the
numpy buffer — no Python→Rust copy, no Vec<[f32;32]>
re-chunk. Python caller passes the (N, 32) ndarray as-is
(ascontiguousarray ensures C-contiguous f32).
Result:
N python rust speedup
1k 0.20ms 0.26ms 0.77x (Python wins on fixed overhead)
10k 1.62ms 1.45ms 1.12x
100k 19.22ms 12.93ms 1.49x
1M 251.50ms 131.36ms 1.91x
Parity bit-identical (raw f32 bytes) at every scale across the
parameterised test in tests/test_vault_recall_rust_parity.py.
Both ADR-0020 first-surface gates now pass: parity AND
performance at the scales where Rust is meant to win. Python
remains the default per CLAUDE.md sequencing rule 5;
CORE_BACKEND=rust is now a legitimate opt-in acceleration.
Smoke 27/27, algebra 70/70, runtime 19/19 all green.
2026-05-16 17:25:41 -07:00
Shay
eb30c75810
feat: Full Proof — surface realizer join, Rust diffusion parity, benchmark harness
...
Surface realizer join: pulse output_versor → vault recall → ground_graph fills
<pending> obj slots with recalled words → realize_semantic produces deterministic
sentences. PulseResult replaces bare word list. Every intent type surfaces.
Rust backend parity: unitize_f32 (exponential-map with boost/rotation blade
distinction) and graph_diffusion_step now in core-rs. Python dispatches through
algebra.backend, falls back transparently. 37x speedup on 200-step diffusion.
Benchmark harness (core bench): determinism (100% trace stability), latency
(~150ms median), backend speedup, versor closure audit (0 violations across all
intermediate states), convergence proof (41/45 exact, 4 bounded oscillation),
realizer coverage (8/8 intent types).
Proof property tests (31 tests): Rust/Python parity, pulse determinism across
prompts, V3 convergence for 10+ topologies, coupled V4 output validity, realizer
coverage per intent, versor closure at every intermediate step.
CLI: core pulse, core bench, core test --suite pulse, core test --suite proof.
Fix test_correction_pulls_toward_target (diffuse first, then correct).
2026-05-15 17:39:14 -07:00
Shay
523c072818
feat: vault recall index, Rust versor parity, cognitive pack expansion
...
Phase 3 — vault exact recall index:
- Replace O(N) np.array_equal scan with hash-based exact-match index
- Add optional max_entries with deterministic FIFO eviction
- Index rebuilds on reproject for consistency
Phase 4 — Rust versor_apply parity:
- Fix CGA metric signature (+,+,+,+,-) and blade ordering to match Python
- Implement versor_apply_closed with null-vector preservation, f64 unitize,
and construction seed fallback matching Python closure semantics
- Gate Rust dispatch behind CORE_BACKEND=rust; Python remains default
- Add f64 geometric product for closure-path precision
Phase 5 — cognitive quality pack expansion:
- Expand lexicon from 55 to 70 entries (evidence, inference, procedure,
verification, distinction, relation, thought, understanding, judgment,
principle, order, connectives)
- Improve semantic templates for cause, procedure, comparison, recall,
verification intents
- Expand eval cases from 20 to 45 across all categories
Validation: 491 tests pass, 45 eval cases at 100% all metrics.
2026-05-15 15:34:39 -07:00
Shay
df9ced7104
Activate and verify Rust backend
...
Add Rust backend CLI controls, fix core-rs build/test configuration, align Rust Cl(4,1)/CGA conventions with Python, and validate core_rs activation.
2026-05-13 22:23:48 -07:00
Shay
0063259584
init: Rust extension crate (core-rs) with PyO3 bindings
2026-05-12 19:19:07 -07:00
