fix(algebra): rotor_power smoke — α≈0 identity + simple B² float-dust tol #33

Merged

core-labs merged 1 commit from feat/fix-rotor-power-smoke-simple-dispatch into main

2026-07-14 04:32:51 +00:00

core-labs commented

2026-07-14 04:30:05 +00:00

Owner

Summary

Unblocks smoke after wave-field merge: rotor_power was raising non-simple bivector under simple dispatch on live chat/runtime paths.

Root cause

Denormal stream weights (α ~ 1e-40) ran the invariant split on heavy multivectors instead of returning identity.
Invariant-split factors are approximately simple; B² higher-grade residual in the 1e-6..1e-3 float-dust band tripped a too-tight fail-closed check (> 1e-6).

Fix

R^0 → I, R^1 → R early exits (_NEAR_ZERO_TOL).
Named _SIMPLE_BSQ_HIGHER_TOL = 1e-3 for simple-dispatch B² higher residual (structural non-simplicity still fail-closed).
Closed-form still uses only the scalar part of B².

Tests

test_rotor_power_near_zero_alpha_is_identity
test_rotor_power_multiplane_transition_half_stays_closed
Full smoke: 108 passed (was 11 failed)
Prior smoke reds + goldtether_alpha + dynamic_manifold: green

Test plan

pytest tests/test_rotor_power.py -q
previously failing achat/runtime/pipeline tests
python3 -m core.cli test --suite smoke -q → 108 passed
CI smoke / lane-shas on PR

Invariants

No silent identity fallback on structural non-simple bivectors (still raises above 1e-3).
Deterministic algebra-native path only.
Does not weaken versor_condition thresholds.

## Summary Unblocks **smoke** after wave-field merge: `rotor_power` was raising `non-simple bivector under simple dispatch` on live chat/runtime paths. ### Root cause 1. **Denormal stream weights** (`α ~ 1e-40`) ran the invariant split on heavy multivectors instead of returning identity. 2. **Invariant-split factors** are *approximately* simple; `B²` higher-grade residual in the `1e-6..1e-3` float-dust band tripped a too-tight fail-closed check (`> 1e-6`). ### Fix - `R^0 → I`, `R^1 → R` early exits (`_NEAR_ZERO_TOL`). - Named `_SIMPLE_BSQ_HIGHER_TOL = 1e-3` for simple-dispatch B² higher residual (structural non-simplicity still fail-closed). - Closed-form still uses only the scalar part of `B²`. ### Tests - `test_rotor_power_near_zero_alpha_is_identity` - `test_rotor_power_multiplane_transition_half_stays_closed` - Full smoke: **108 passed** (was 11 failed) - Prior smoke reds + goldtether_alpha + dynamic_manifold: green ## Test plan - [x] `pytest tests/test_rotor_power.py -q` - [x] previously failing achat/runtime/pipeline tests - [x] `python3 -m core.cli test --suite smoke -q` → 108 passed - [ ] CI smoke / lane-shas on PR ## Invariants - No silent identity fallback on *structural* non-simple bivectors (still raises above 1e-3). - Deterministic algebra-native path only. - Does not weaken `versor_condition` thresholds.

core-labs added 1 commit 2026-07-14 04:30:13 +00:00

fix(algebra): rotor_power smoke — α≈0 identity + simple B² float-dust tol

smoke / smoke (-m "not quarantine") (pull_request) Successful in 5m55s

Details

lane-shas / verify pinned lane SHAs (pull_request) Failing after 1h43m35s

Details

e124d83bfa

Smoke reds came from (1) denormal stream weights powering non-simple splits
and (2) invariant-split factors with B² higher residual in 1e-6..1e-3 dust
band raising fail-closed. Early R^0→I / R^1→R; raise simple B² higher tol
to 1e-3 with named constant. Pins multiplane transition + near-zero alpha.