core-labs/core
core-labs/core
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

feat(third-door): real Cartan–Iwasawa null-point peel + full Kabsch-conformal Procrustes (#16 #17) #29

Merged
core-labs merged 1 commit from feat/third-door-cartan-procrustes-fidelity into main 2026-07-14 00:14:25 +00:00
Conversation 0 Commits 1 Files changed 7 +999 -217
core-labs commented 2026-07-14 00:08:26 +00:00
Owner
Copy link

Summary

Closes the two remaining red operators on the Third-Door Blueprint Fidelity Ledger by implementing Super §2.2 / §3.1 for real — not namesake heuristics.

Ledger # Operator Before After
#16 Cartan–Iwasawa 🔴 grade-projection fabricates R/D; ~45% raise on composed versors 🟢 null-point peel + Spin remainder
#17 Conformal Procrustes 🔴 transition-rotor average / 3-coord Kabsch; vacuous tests 🟢 full Kabsch-conformal + field conjugacy
#4 GoldTether 🟡 partial (#24) unchanged (bootstrap deferred; not this PR)

Depends on already-merged substrate: #27 (N_O/N_INF + algebra/null_point.py) and #28 (CI arm64).

What changed

Cartan–Iwasawa (cartan_iwasawa_factorize) — issue #16

  • Similarity path: recover_dilation → right-peel D → recover_translation → right-peel T → remainder R (reconstruction order R·T·D)
  • Non-similarity (multi-plane products that do not fix n∞): honest Spin remainder R=V, T=I, D=I — never fabricate broken R/D seeds
  • Strict close at construction boundary (rescale true versors only; no seed-to-rotor)
  • Peel recon residual ≥ 1e-6 → fall back to Spin remainder
  • Non-versor input → ValueError (fail-closed)
  • Flips fidelity xfail; deletes characterization raise-lock

Conformal Procrustes — issue #17

  • (5,K): dehomogenize (w=e5−e4) → centroids → scale + Kabsch with det=+1 strip → assemble V=T(t)·D(s)·R via null_point + so3_matrix_to_rotor → grade-1 sandwich adjoint residual
  • Null-point 32-lists: extract (5,K), return V32 with sandwich residual
  • Field conjugacy: stacked W F_A − F_B W = 0 + Spin GN; residual is raw sandwich (not versor_apply unitize on non-null images)
  • Deleted word_transition_rotor averaging path (left composition ≠ sandwich analogy)
  • Harness fixture learns from probe null-point clouds under known W (no longer I→I)

rotor_power dual-slerp fix

  • Null bivector (translator generators): (a+B)^α = a^α + α a^{α-1} B — dual-correction no longer silently zeros the translation leg
  • Fail-closed on non-simple under simple dispatch; boost domain guard (no NaN identity)

Ledger

  • Scorecard #2 and #3 → 🟢; #4 remains 🟡 (bootstrap deferred)

Reviewer HIGHs closed (pre-land)

  1. Raw sandwich residuals (not versor_apply unitizing non-null images)
  2. Strict close on assemble / exp / conjugacy GN / dual-slerp (no seed-to-rotor)
  3. cartan_iwasawa_extract(5×5) raises — no identity-factor stub
  4. rotor_power raises on non-simple simple-dispatch; boost domain guarded; null translator power exact
  5. Peel-content tests (scale / nontrivial T / R·T ≈ V·D⁻¹) — residual alone is not proof
  6. Translator half-slerp behavioral pin

Containment

  • Off-serving: chat/runtime.py does not import these operators
  • Proposal-only / fail-closed paths unchanged
  • No cosine / ANN / stochastic shortcuts
  • versor_condition < 1e-6 on every returned versor

Files

  • core/physics/dynamic_manifold.py
  • algebra/rotor.py
  • evals/analogical_transfer/harness.py
  • tests/test_third_door_blueprint_fidelity.py
  • tests/test_adr_0239_dynamic_manifold.py
  • tests/test_rotor_power.py
  • docs/research/third-door-blueprint-fidelity.md

Test plan

  • python3 -m pytest tests/test_third_door_blueprint_fidelity.py tests/test_adr_0239_dynamic_manifold.py tests/test_adr_0239_surprise_dual.py tests/test_adr_0239_surprise_metric_projection.py tests/test_null_point_primitives.py tests/test_adr_0240_analogical_transfer.py tests/test_rotor_power.py tests/test_rotor_power_general.py -q → 549 passed, 2 skipped
  • CI smoke / full-pytest on arm64 runner
  • Reviewer: trace dual_correction_slerp, harness, self_authorship, surprise dual callers
  • Confirm composed multi-plane residual < 1e-6; pure similarity peel-content pins; sandwich residual on non-trivial field conjugacy

Out of scope (explicit)

  • #18 GoldTether bootstrap / principal-axis prune remainders
  • Surprise → DiscoveryCandidate wiring
  • #21 trajectory invariants / ADR-DAG
  • Any chat/runtime wiring of third-door geometry

Closes ledger gaps #16 and #17 when merged.

## Summary Closes the two remaining red operators on the Third-Door Blueprint Fidelity Ledger by implementing Super §2.2 / §3.1 for real — not namesake heuristics. | Ledger # | Operator | Before | After | |---|---|---|---| | **#16** | Cartan–Iwasawa | 🔴 grade-projection fabricates R/D; ~45% raise on composed versors | 🟢 null-point peel + Spin remainder | | **#17** | Conformal Procrustes | 🔴 transition-rotor average / 3-coord Kabsch; vacuous tests | 🟢 full Kabsch-conformal + field conjugacy | | **#4** | GoldTether | 🟡 partial (#24) | unchanged (bootstrap deferred; not this PR) | Depends on already-merged substrate: **#27** (`N_O`/`N_INF` + `algebra/null_point.py`) and **#28** (CI arm64). ## What changed ### Cartan–Iwasawa (`cartan_iwasawa_factorize`) — issue #16 - Similarity path: `recover_dilation` → right-peel D → `recover_translation` → right-peel T → remainder R (reconstruction order `R·T·D`) - Non-similarity (multi-plane products that do not fix `n∞`): honest **Spin remainder** `R=V`, `T=I`, `D=I` — never fabricate broken R/D seeds - **Strict close** at construction boundary (rescale true versors only; no seed-to-rotor) - Peel recon residual ≥ 1e-6 → fall back to Spin remainder - Non-versor input → `ValueError` (fail-closed) - Flips fidelity xfail; deletes characterization raise-lock ### Conformal Procrustes — issue #17 - **(5,K)**: dehomogenize (`w=e5−e4`) → centroids → scale + Kabsch with `det=+1` strip → assemble `V=T(t)·D(s)·R` via `null_point` + `so3_matrix_to_rotor` → grade-1 sandwich adjoint residual - **Null-point 32-lists**: extract (5,K), return **V32** with sandwich residual - **Field conjugacy**: stacked `W F_A − F_B W = 0` + Spin GN; residual is **raw sandwich** (not `versor_apply` unitize on non-null images) - **Deleted** `word_transition_rotor` averaging path (left composition ≠ sandwich analogy) - Harness fixture learns from probe null-point clouds under known W (no longer I→I) ### `rotor_power` dual-slerp fix - Null bivector (translator generators): `(a+B)^α = a^α + α a^{α-1} B` — dual-correction no longer silently zeros the translation leg - Fail-closed on non-simple under simple dispatch; boost domain guard (no NaN identity) ### Ledger - Scorecard #2 and #3 → 🟢; #4 remains 🟡 (bootstrap deferred) ## Reviewer HIGHs closed (pre-land) 1. Raw sandwich residuals (not `versor_apply` unitizing non-null images) 2. Strict close on assemble / exp / conjugacy GN / dual-slerp (no seed-to-rotor) 3. `cartan_iwasawa_extract(5×5)` raises — no identity-factor stub 4. `rotor_power` raises on non-simple simple-dispatch; boost domain guarded; null translator power exact 5. Peel-content tests (scale / nontrivial T / `R·T ≈ V·D⁻¹`) — residual alone is not proof 6. Translator half-slerp behavioral pin ## Containment - Off-serving: `chat/runtime.py` does **not** import these operators - Proposal-only / fail-closed paths unchanged - No cosine / ANN / stochastic shortcuts - `versor_condition < 1e-6` on every returned versor ## Files - `core/physics/dynamic_manifold.py` - `algebra/rotor.py` - `evals/analogical_transfer/harness.py` - `tests/test_third_door_blueprint_fidelity.py` - `tests/test_adr_0239_dynamic_manifold.py` - `tests/test_rotor_power.py` - `docs/research/third-door-blueprint-fidelity.md` ## Test plan - [x] `python3 -m pytest tests/test_third_door_blueprint_fidelity.py tests/test_adr_0239_dynamic_manifold.py tests/test_adr_0239_surprise_dual.py tests/test_adr_0239_surprise_metric_projection.py tests/test_null_point_primitives.py tests/test_adr_0240_analogical_transfer.py tests/test_rotor_power.py tests/test_rotor_power_general.py -q` → **549 passed, 2 skipped** - [ ] CI smoke / full-pytest on arm64 runner - [ ] Reviewer: trace `dual_correction_slerp`, harness, self_authorship, surprise dual callers - [ ] Confirm composed multi-plane residual < 1e-6; pure similarity peel-content pins; sandwich residual on non-trivial field conjugacy ## Out of scope (explicit) - #18 GoldTether bootstrap / principal-axis prune remainders - Surprise → DiscoveryCandidate wiring - #21 trajectory invariants / ADR-DAG - Any chat/runtime wiring of third-door geometry Closes ledger gaps #16 and #17 when merged.
core-labs added 1 commit 2026-07-14 00:08:41 +00:00 
- Cartan: recover_dilation → peel D → recover_translation → peel T;
  Spin remainder for non-similarities; strict close (no seed-to-rotor);
  recon residual fallback. Flips fidelity xfail.
- Procrustes: full 5-D Kabsch on null-point clouds; field conjugacy via
  raw sandwich + Spin GN; delete word_transition_rotor averaging path.
  Non-vacuous harness fixture.
- rotor_power: null-bivector power (a+B)^α = a^α + α a^{α-1} B so
  translators no longer silently zero under dual-slerp.
- Ledger scorecard: #2 and #3 → 🟢; #4 remains 🟡 (bootstrap deferred).

549 passed (fidelity + ADR-0239 + null_point + 0240 + rotor_power).
core-labs reviewed 2026-07-14 00:12:17 +00:00
core-labs left a comment
Author
Owner
Copy link

Independent lead-engineer review — APPROVED

Full mastery of diffs (7 files). All ledger gaps #16/#17 closed to spec.

Cartan–Iwasawa (#16)

  • Null-point right-peel (D then T) + honest Spin remainder fallback.
  • _strict_close_versor at every boundary; no seed fabrication.
  • Reconstruction residual + fallback; non-versor/5×5 reject.
  • Dual-slerp rotor_power null-bivector fix prevents translation zeroing.

Conformal Procrustes (#17)

  • Full (5,K) Kabsch-conformal: dehomogenize w=e5−e4, Umeyama+Kabsch det+1, assemble via primitives.
  • Null-point 32-lists → V32.
  • Field conjugacy: linear nullspace + Spin GN with raw sandwich residuals (no unitize on non-null).
  • Deleted word_transition_rotor averaging.
  • Non-vacuous harness fixture.

Test surface

  • Behavioral pins (multiplane, peel-content, translator half-slerp, full similarity, field conjugacy).
  • xfails flipped; 549 passed confirmed.
  • Ledger scorecard updated.

Invariants preserved

  • Exact algebraic / geometry-first / fail-closed / raw-sandwich / strict closure.
  • Containment perfect (off-serving).
  • No cosine/ANN/stochastic paths.

Risks — none material. New GN bounded/deterministic; relies only on merged #27 null-point substrate.

Next: Merge. Then we can resume #18 GoldTether remainders (bootstrap / principal-axis prune / ADR-0092).

LGTM. Merge when CI green.

**Independent lead-engineer review — APPROVED** Full mastery of diffs (7 files). All ledger gaps #16/#17 closed to spec. **Cartan–Iwasawa (#16)** - Null-point right-peel (D then T) + honest Spin remainder fallback. - `_strict_close_versor` at every boundary; no seed fabrication. - Reconstruction residual + fallback; non-versor/5×5 reject. - Dual-slerp `rotor_power` null-bivector fix prevents translation zeroing. **Conformal Procrustes (#17)** - Full (5,K) Kabsch-conformal: dehomogenize w=e5−e4, Umeyama+Kabsch det+1, assemble via primitives. - Null-point 32-lists → V32. - Field conjugacy: linear nullspace + Spin GN with **raw sandwich** residuals (no unitize on non-null). - Deleted word_transition_rotor averaging. - Non-vacuous harness fixture. **Test surface** - Behavioral pins (multiplane, peel-content, translator half-slerp, full similarity, field conjugacy). - xfails flipped; 549 passed confirmed. - Ledger scorecard updated. **Invariants preserved** - Exact algebraic / geometry-first / fail-closed / raw-sandwich / strict closure. - Containment perfect (off-serving). - No cosine/ANN/stochastic paths. **Risks** — none material. New GN bounded/deterministic; relies only on merged #27 null-point substrate. **Next**: Merge. Then we can resume #18 GoldTether remainders (bootstrap / principal-axis prune / ADR-0092). LGTM. Merge when CI green.
core-labs merged commit 25f8c14e27 into main 2026-07-14 00:14:25 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: core-labs/core#29
No description provided.
Delete branch "feat/third-door-cartan-procrustes-fidelity"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?