# Sundog Active TODO Last assembled: 2026-05-29. This is the operator queue for outstanding roadmap work, experiment gates, and blockers that were getting scattered across docs, anniversary notes, and public site passes. Source specs and pre-registrations remain authoritative. If this file conflicts with a spec, the spec wins and this file should be corrected. Status tags: - `compute-blocked`: the next result needs a long run or a runner plan. - `operator-blocked`: the next result needs a human artifact, review, or manual measurement. - `design-blocked`: the next result needs a sharper spec before execution. - `public-surface`: the next result affects site copy, Ask Sundog, or launch posture. - `review-gated`: the next result is blocked on an external review packet landing — a sanity-check pass from an outside reader, a referee response, a collaborator review, or a publication-style review cycle. Distinguishes external-cycle waits from internal blockers. Riemann bounded-null synthesis is the canonical example. - `internal-resource`: the lane is on hold for internal capacity reasons (compute budget, bandwidth, operator focus elsewhere) rather than for a missing artifact, spec, or review. Use when the work is ready to run but the program has chosen to spend the cycles elsewhere. The Isotrophy K_facet v0.3-v0.9 program post-pause sits here. - `deferred`: worth keeping visible, but not a launch gate. - `dead-end`: a closed-negative result that has been audited and is retained for the record but is no longer actionable. Distinguishes "we tried this, it did not yield a structural edge, here's why" from "still queued, just not scheduled." Use when reopening would require new substrate, new theory, or new instrument — not just more cycles. Default presumption is that dead-ends do not need re-triage on each pass. ## Current Holds At A Glance (2026-05-29 inventory) A snapshot of where each tracked lane sits, sorted by tag class. This is a read-through summary, not authoritative — each item's own section below is the source of truth. ### `review-gated` — waiting for an outside reader - **Riemann bounded-null synthesis** — three-lane null with named substrate causes; external-review packet at [`riemann/EXTERNAL_REVIEW_PACKET.md`](riemann/EXTERNAL_REVIEW_PACKET.md). Public surface blocked on review return. ### `internal-resource` — paused for capacity reasons, not artifact gaps - **Isotrophy K_facet program (v0.3-v0.9)** — paused 2026-05-24 after seven pre-registered chapters and the v0.3h structural-zero verdict shipped to `/isotrophy`. Pause document: [`isotrophy/kfacet/kfacet_isotrophy_program_pause.md`](isotrophy/kfacet/kfacet_isotrophy_program_pause.md). Eight reopening avenues catalogued for any lab initiate with bandwidth. Bandwidth currently directed to three-body Phase 15+ and the coarse-graining proof trunk. - **ARC Branch E3 learned ranker** — built, compute-paused by operator decision 2026-05-29. Not blocked on a missing artifact; the program chose to spend cycles elsewhere. ### `review-gated` + draft handoff — review surface not yet built - **Yang-Mills Phase 0 handoff** — draft finite-lattice gauge-invariant certificate lane. No lit-pass, preregistration, runner, receipt, mass-gap claim, or Clay-problem claim is live. The next move is the handoff itself; until then this is `design-blocked, draft-handoff`. ### `design-blocked` — sharper spec needed before execution - **Navier-Stokes C1 regime-generality wall** — `G = 300` regime-generality attempt hit objective vacuity rather than fiber-locality. Next move is a portable-objective design decision. - **ARC Phase 3E relative-locality hold** — rank-neighbor certificate proposal after absolute fibers stayed sparse on the expanded 108-task register. - **P-vs-NP Phase 2 mesa bridge** - Phase 2 v0 named-quarantined on raw-log provenance; v1 provenance repair executed as a bounded-positive mesa-bridge receipt. Capacity threshold remains not estimated. ### `operator-blocked` — needs a human artifact, review, or measurement - **Structural-Failure Cut 3** — admission check held open; reopening requires H0 angular calibration, corpus records, baselines, edit operators, and renewed admission before any execution claim. ### `compute-blocked` — the next result needs a long run - **Coarse-graining proof trunk** — Phase 4 BF-4b second diagnostic scheduled, not run. - **Mesa Phase 7 v3** — basin-attractor caveat extension. - **Three-Body Phase 15C** — multi-step counterfactual sweep. ### `public-surface` — site / chat / launch posture - **Ask Sundog claim-map freshness** — chat library rewrite is the next major surface move; the 2026-05-22 regen against the refreshed claim map is in place, library rewrite still to come. - **Application evidence assets** — receipt-grade visuals across the workbenches. - **Anniversary rollout aftercare** — post-anniversary surface follow-ups. - **Canonical routes / deploy verification** — pre-deploy hygiene. ### `deferred` — visible but not a launch gate - Vortex / Wishing-Well toy - Repo Map / Agent Navigation Surface (delivered for /repo-map this pass; the agent-navigation generator remains deferred) - V0.3h K_facet Tooling Polish — ~10-hour budget, no compute lock ### `dead-end` — closed, retained for audit only - *(none currently flagged)* — V0.4b through V0.9 of the K_facet program are chapter-closed verdicts that landed inside the `internal-resource` Isotrophy pause envelope rather than dead-ends. Reserve this tag for lanes that would require new substrate, theory, or instrument to reopen. ## P0/P1 Load-Bearing Queue ### 1. Coarse-Graining Proof Trunk Sources: [`COARSE_GRAINING_PROOF_ROADMAP.md`](COARSE_GRAINING_PROOF_ROADMAP.md), [`proof/PHASE4_THREEBODY.md`](proof/PHASE4_THREEBODY.md), [`proof/PHASE4_BAYESIAN_FLOOR_BUILDOUT.md`](proof/PHASE4_BAYESIAN_FLOOR_BUILDOUT.md), [`proof/PHASE6_LAMBDA_CONTROL.md`](proof/PHASE6_LAMBDA_CONTROL.md). Status: `compute-blocked`, active. Current state: - Phases 0-3 are positive on the proof track: definitions lock, LQG proof, finite-MDP proof, and boundary theorem. - Phase 4 has a three-body measured-substrate spec and a smoke-passed Bayesian-floor controller path, but the gate is not closed. - BF-4b's first off-set calibration was a useful negative: the off-set cell gave regret CI `[0, 0]`, while the satisfiability probe showed oracle-signature headroom. That means the next diagnostic is real work, not bookkeeping. - Phase 5 cross-substrate sameness is blocked on a positive Phase 4. - Phase 6 lambda-control is staged, with empirical result still open. Blocker: The compute-unconstrained Information-Accessibility Diagnostic and BF-5 sharded/full-lock path need a long-run plan. Treat this as multiple-day compute until measured otherwise. Do not abandon the trunk for a cheaper target just to get a faster public result. Next actions: 1. Add the exact runnable PowerShell commands to [`proof/PHASE4_THREEBODY.md`](proof/PHASE4_THREEBODY.md), including module or npm form, output directory, resume behavior, and read-back manifest path. 2. Run only a capped rate probe inline, if it is under the repo's ~10 minute rule. Record per-cell or per-trial rate in the spec. 3. Choose the runner for the real lock: local operator over multiple days, GitHub Actions `workflow_dispatch`, or self-hosted runner. Do not use an interactive coding-agent session for a multi-hour/multi-day lock. 4. Pre-register the branch table before the run: measurable-set regret goes to zero -> Phase 5 entry; measurable-set regret remains bounded away -> halt/falsify on real substrate; off-set regret goes to zero -> reopen the boundary theorem. 5. After Phase 4 closes, decide whether Phase 5 compares mesa, geometry, and a third toy, or stays strictly mesa plus geometry. ### 2. Structural Failure Boundary Map / Cut 3 Sources: [`prereg/structural-failure-coincidence/BOUNDARY_MAP.md`](prereg/structural-failure-coincidence/BOUNDARY_MAP.md), [`prereg/structural-failure-coincidence/PUBLICATION_PLAN.md`](prereg/structural-failure-coincidence/PUBLICATION_PLAN.md), [`prereg/structural-failure-coincidence/P2_CUT3_ADMISSION.md`](prereg/structural-failure-coincidence/P2_CUT3_ADMISSION.md), [`prereg/structural-failure-coincidence/P2_CUT3_H0_CALIBRATION.md`](prereg/structural-failure-coincidence/P2_CUT3_H0_CALIBRATION.md), [`prereg/structural-failure-coincidence/P2_CUT3_H0_2_SCHEMA.md`](prereg/structural-failure-coincidence/P2_CUT3_H0_2_SCHEMA.md). Status: `operator-blocked`, active. Current state: - P0 boundary map is frozen and passed. - P1 admission passed. - P2 first cut was corrected as vacuous. - Cut 2 produced closed-form regime-separability, not a controller result. - Cut 3 rendered escalation is opened but execution is held. Blocker: H0 angular calibration needs real-frame measurement. The operator has to mark or draw the relevant anchors on actual pictures/renders, not on modeled stubs: sun center, scale ticks or angular map, anchor loci, scored feature, valid span, and any filename/configuration fields that might leak `h`. Next actions: 1. Prepare an image packet for operator annotation, with one sidecar template per frame and a short instruction sheet. 2. Identify the known-PASS full-span fixture from real HaloSim output. 3. Fill measured sidecars for the Phase-15 known-FAIL frames, the known-PASS fixture, and each proposed Cut-3 corpus frame. 4. Generate the anchor-residual table from the sidecars. 5. Record the operator decision on whether HaloSim compound codes such as `e13` encode sun elevation and should trip the H-leak rule. 6. Re-run Cut-3 admission only after the render corpus manifest, H0 residual table, concrete agent path, baselines, edit operators, and write-path guard all exist. Stop condition: If honest angular coverage fails, Cut 3 stays blocked. Do not stretch the span, substitute anchors, or score off-ruler features to force a result. ### 3. Ask Sundog Claim-Map Freshness Sources: [`SUNDOG_V_CHAT.md`](SUNDOG_V_CHAT.md), [`WEBSITE_DEVELOPMENT.md`](site/WEBSITE_DEVELOPMENT.md), [`../chat/contents.json`](../chat/contents.json), [`../chat/claim_map.json`](../chat/claim_map.json). Status: `public-surface`, launch-sensitive. Current state: The widget has deterministic trace and boundary machinery, but the site changed quickly: atlas/legend disambiguation, alignment page, anniversary posture, commercial translation language, and homepage proof panels all moved. The next Ask Sundog pass should be isolated so it does not casually rewrite or contaminate the eval. Blocker: `chat/claim_map.json`, prompt gold slates, and static answers may be behind the current public copy. Next actions: 1. Inventory new claim phrases and routes from `index.html`, `about.html`, `alignment.html`, `sundog.html`, `legend.html`, `applications-gallery.html`, `mesa.html`, and `structural-failure.html`. 2. Update `chat/claim_map.json` and the relevant prompt slates only where the public copy now invites a real visitor question. 3. Rebuild generated data with `npm run chat:index`. 4. Run, at minimum: ```powershell npm run chat:eval:static npm run chat:eval:phase3 npm run chat:eval:phase3:adversarial npm run chat:eval:phase3:differential npm run chat:eval:phase4 ``` 5. Publish only if the gate preserves evidence tiers, source support, and active boundaries under adversarial pressure. ### 4. Mesa Phase 7 v3 / Basin-Attractor Caveat Sources: [`mesa/PHASE7_V2_RESULTS.md`](mesa/PHASE7_V2_RESULTS.md), [`SUNDOG_V_MESA.md`](SUNDOG_V_MESA.md), [`../internal/roadmaps/fix_roadmap.md`](../internal/roadmaps/fix_roadmap.md). Status: `compute-blocked`, active. Current state: Large Phase 7 v2 shows a U-shape recovery at `lambda=0.99`, but the eval summary does not compute `old_basin_pref`. High terminal alignment is consistent with both "reaches the basin" and "collapses onto a fixed attractor that co-points with the basin direction." Blocker: We need a Phase 4-style intervention battery on Large checkpoints before saying the recovery avoids the mesa-trap attractor. Next actions: 1. Draft Phase 7 v3 with `old_basin_pref` or an equivalent intervention-confirmed basin-avoidance metric. 2. Reuse the coherent-signal framing: signature-pure and reward-pure may both be coherent classes; mixed-signal controllers destabilize when the mixture creates inference noise. 3. Keep public language at "basin-reaching by current eval metric, not yet verified basin-attractor-avoiding." 4. Measure compute cost before running a Large-tier battery. ### 5. Three-Body Phase 15C Multi-Step Counterfactual Sources: [`threebody/PHASE15_SPEC.md`](threebody/PHASE15_SPEC.md), [`threebody/PHASE15_RESULTS.md`](threebody/PHASE15_RESULTS.md), [`threebody/PHASE15B_RESULTS.md`](threebody/PHASE15B_RESULTS.md), [`threebody/PHASE15C_SPEC.md`](threebody/PHASE15C_SPEC.md), [`threebody/PHASE15C_RESULTS.md`](threebody/PHASE15C_RESULTS.md), [`../copilot_test_readme.md`](../copilot_test_readme.md). Status: `compute-blocked`, active. Current state: Phase 15 full lock is complete and landed as **Fail-Magnitude**. Phase 15B's 1,728-trial normalizer lock is complete and landed as **Mixed / Partial Diagnostic**: denominator-floor collapse is real and near-universal, but it does not explain the Phase 15 magnitude miss. Phase 15C's multi-step counterfactual horizon audit is now implemented and smoke-passed. Operator sign-off landed 2026-05-29, and the 12-shard lock is running in waves; as of [`threebody/PHASE15C_RESULTS.md`](threebody/PHASE15C_RESULTS.md), 6 / 12 shards are logged. Blocker: The binding read is still incomplete. The interim trial-pooled note is not a branch read because the candidate-split readback has to wait for all 12 shards. The audit remains additive and must not retune the controller or revise the Phase 15 verdict. Next actions: 1. Finish the remaining six shards and append the shard run log. 2. Aggregate the candidate-split readback across all 12 shard directories. 3. Read the candidate-envelope horizon table before assigning any branch. 4. Preserve the non-candidate pooled observations only as diagnostics. 5. Keep Phase 15C interpretation separate from coarse-graining Phase 4 unless Phase 4 explicitly chooses this evidence path. ## High-Stakes Math Loose Ends These are the active "hit a wall / waiting on review / do not accidentally promote" lanes opened during the late-May high-stakes generality push. They are tracked here so the math-ledger documents can stay precise without becoming the operator inbox. ### Riemann Bounded-Null External Review Sources: [`SUNDOG_V_RIEMANN.md`](SUNDOG_V_RIEMANN.md), [`riemann/README.md`](riemann/README.md), [`riemann/RIEMANN_BOUNDED_NULL_SYNTHESIS.md`](riemann/RIEMANN_BOUNDED_NULL_SYNTHESIS.md), [`riemann/EXTERNAL_REVIEW_PACKET.md`](riemann/EXTERNAL_REVIEW_PACKET.md), [`riemann/EXTERNAL_REVIEW_EMAIL_DRAFT.md`](riemann/EXTERNAL_REVIEW_EMAIL_DRAFT.md). Status: `operator-blocked`, review-gated. Current state: Three Riemann-adjacent lanes now have clean bounded-null receipts with distinct causes: Path-i Z2 parity, C1 explicit-formula/cell-set reading, and nonlinear S2 gap-pair reversibility. The synthesis explicitly says no structural-zero edge was found and no RH progress is claimed. Blocker: Public or promotional language is blocked on an external sanity check. The next step is not another probe unless the reviewer identifies a viable new lane. Next actions: 1. Send the short review email using [`riemann/EXTERNAL_REVIEW_EMAIL_DRAFT.md`](riemann/EXTERNAL_REVIEW_EMAIL_DRAFT.md). 2. Attach or link only the bounded-null synthesis and external-review packet. 3. Record the send date, reviewer category, and response disposition in the Riemann folder or a dated receipt note. 4. If the reviewer says the null is ordinary / unsurprising, close the public surface as "bounded null, externally sanity-checked." 5. If the reviewer points to a real mathematical opening, require a new pre-registration before any new run or claim language. ### Navier-Stokes C1 Regime-Generality Wall Sources: [`SUNDOG_V_NAVIERSTOKES.md`](SUNDOG_V_NAVIERSTOKES.md), [`proof/PDE_C1_LOCK_EXECUTION_SYNTHESIS.md`](proof/PDE_C1_LOCK_EXECUTION_SYNTHESIS.md), [`proof/PDE_C1_KNN_CONVERGENCE_CHECK.md`](proof/PDE_C1_KNN_CONVERGENCE_CHECK.md), [`proof/PDE_C1_TWIN_STATE_CERTIFICATE.md`](proof/PDE_C1_TWIN_STATE_CERTIFICATE.md), [`proof/PDE_C1_REGIME_GENERALITY_v0.md`](proof/PDE_C1_REGIME_GENERALITY_v0.md), [`proof/PDE_C1_REGIME_GENERALITY_v1.md`](proof/PDE_C1_REGIME_GENERALITY_v1.md). Status: `design-blocked`, high-stakes math. Current state: The v5 Kolmogorov cell has a tightly scoped Reading-2 regime-2 witness: control-sufficiency on fibers plus sampled-support twin-state non-injectivity. The first Grashof-axis generality attempt at `G = 300` deferred as objective vacuity, not as a fiber-locality failure. Blocker: [`proof/PDE_C1_REGIME_GENERALITY_v1.md`](proof/PDE_C1_REGIME_GENERALITY_v1.md) is a design proposal only. It needs sign-off on the portable objective, split sizes, and run order before any harness build or long run. Next actions: 1. Decide the v1 open questions: 50k/50k split, `q = 0.70`, and build trigger. 2. If signed off, implement `--objective portable-quantile` and the `lock_v7_g200` / `lock_v7_g300` presets behind the new objective mode. 3. Run only a smoke inline; stage the expected 25-40 minute verdict-bearing runs as operator commands under the repo's long-run rule. 4. Do not promote C1 beyond "one-cell, finite-Galerkin, sampled-support witness" without external PDE review. ### Yang-Mills Phase 0 Handoff Sources: [`SUNDOG_V_YANG_MILLS.md`](SUNDOG_V_YANG_MILLS.md). Status: `design-blocked`, draft-handoff. Current state: The Yang-Mills lane is now a roadmap only. It proposes a finite-lattice, gauge-invariant certificate program influenced by the ARC relative-locality lesson: certify bounded invariant structure before attempting reconstruction or any continuum-facing claim. Blocker: No lit-pass, domain lock, pre-registration, source library, runner, or receipt exists. The lane must not be promoted as a Yang-Mills, mass-gap, confinement, or Clay-problem result. Next actions: 1. Draft `docs/YANG_MILLS_LITPASS_MEMO.md`. 2. Create `docs/prereg/yang-mills/P0_DOMAIN_AND_RECEIPT_LOCK.md` with the finite-lattice domain, observable labels, controls, and exact claim boundary. 3. Choose the first source library or imported ensemble path before writing runner code. 4. Register gauge-invariance, metadata-shortcut, raw-link, random-neighbor, coupling-stratified, and label-permutation controls. 5. Keep the gallery/card language at "draft finite-lattice certificate lane" until Phase 0 exists. ### P-vs-NP Phase 1 v6 Op-Count Receipt Sources: [`SUNDOG_V_P_V_NP.md`](SUNDOG_V_P_V_NP.md), [`pvnp/README.md`](pvnp/README.md), [`pvnp/PHASE1_V6_SLATE.md`](pvnp/PHASE1_V6_SLATE.md), [`pvnp/PHASE2_MESA_BRIDGE_V0_SLATE.md`](pvnp/PHASE2_MESA_BRIDGE_V0_SLATE.md), [`pvnp/receipts/README.md`](pvnp/receipts/README.md), [`pvnp/PHASE3_CAPACITY_ONE_WAYNESS_V0_SLATE.md`](pvnp/PHASE3_CAPACITY_ONE_WAYNESS_V0_SLATE.md). Status: `phase3-v0-slate-opened-for-review`, v6 and Phase 2 v1 bounded positive. Current state: Phase 1 v0-v6 produced a useful chain of verifier receipts. v5 is a corrected named quarantine: safety-complete, wall-time cost unadjudicated. v6 is bounded positive under the registered op-count protocol: `C_total_signature_ops / C_rollout_ops = 0.948587 <= 1.0`, with wall-time diagnostic-only. Phase 2 v0 is filed as named quarantine, and Phase 2 v1 closes the raw-log provenance gap as a bounded positive under the frozen mesa-bridge contract. Phase 3 v0 is now opened for review to register attacker capacity, inversion, and spoof tests. Blocker: The next move is not another Phase 1 cost repair or Phase 2 provenance repair. Those claim boundaries are fixed. The remaining open question is capacity-relative one-wayness / spoof resistance under a frozen Phase 3 attacker battery. Next actions: 1. Review and freeze the Phase 3 capacity-relative one-wayness v0 slate. 2. Decide whether v0 runs seed-extension batteries or named-quarantines spoof search as unfaithful. 3. Freeze the attacker/capacity battery before any execution. 4. Carry forward the v6 boundary verbatim: op-count positive, wall-time diagnostic-only, no complexity-theoretic claim. 5. Do not use the v3/v5 favorable wall-time samples as evidence. 6. Treat `capacity_threshold = not_estimated` as still open for any larger attacker or domain expansion. ### ARC Phase 3E Relative-Locality Hold Sources: [`SUNDOG_V_ARC.md`](SUNDOG_V_ARC.md), [`prereg/arc/README.md`](prereg/arc/README.md), [`prereg/arc/PHASE3E_SIGNATURE_FIBER_CERTIFICATE_SPEC.md`](prereg/arc/PHASE3E_SIGNATURE_FIBER_CERTIFICATE_SPEC.md), [`prereg/arc/PHASE3E_PROGRAM_SKETCH_ORACLE_V2_SPEC.md`](prereg/arc/PHASE3E_PROGRAM_SKETCH_ORACLE_V2_SPEC.md), [`prereg/arc/PHASE0_CONTEXT_EXPANSION_FOR_FIBERS_SPEC.md`](prereg/arc/PHASE0_CONTEXT_EXPANSION_FOR_FIBERS_SPEC.md), [`prereg/arc/PHASE3E_RELATIVE_LOCALITY_CERTIFICATE_SPEC.md`](prereg/arc/PHASE3E_RELATIVE_LOCALITY_CERTIFICATE_SPEC.md). Status: `design-blocked`, execution hold. Current state: The deterministic low-capacity ARC lanes have hit repeated full-grid exact-match floors across learner families and framings. The expanded Phase 3E oracle lane also found absolute-epsilon fibers still sparse and surfaced an oracle leakage regression on the 108-task register. The relative-locality spec asks a narrower rank-neighbor question without erasing those findings. Blocker: The relative-locality spec explicitly forbids execution until runner tooling, npm wiring, result ignore path, leak-check coverage, smoke fingerprint, and a freeze-marker amendment are filed together. Candidate tooling is present in the working tree (`docs/prereg/arc/phase3e_relative_locality_certificate.py`, `scripts/arc-phase3e-relative-locality-certificate.mjs`, plus package/result path wiring), but it is not a receipt and should be admitted and tracked with the freeze marker and smoke fingerprint before any run is treated as official. Next actions: 1. Admit the runner in one freeze-marker change with npm wiring, result path, leak checks, and smoke fingerprint. 2. Run only the smoke inline if it stays under the ~10-minute rule. 3. If the full relative-locality receipt is longer, stage the exact command and readback path for the operator. 4. Keep any positive language caveated as rank-local sketch coherence under a leakage-caveated oracle, not a solver or sufficiency result. ### ARC Branch E3 Learned Ranker (built, compute-paused) Sources: [`prereg/arc/PHASE3_BRANCH_E3_LEARNED_RANKER_SPEC.md`](prereg/arc/PHASE3_BRANCH_E3_LEARNED_RANKER_SPEC.md), [`prereg/arc/PHASE3_BRANCH_E_PROGRAM_SEARCH_SPEC.md`](prereg/arc/PHASE3_BRANCH_E_PROGRAM_SEARCH_SPEC.md), [`prereg/arc/PHASE3_BRANCH_E_V2_PROGRAM_SEARCH_SPEC.md`](prereg/arc/PHASE3_BRANCH_E_V2_PROGRAM_SEARCH_SPEC.md), [`prereg/arc/README.md`](prereg/arc/README.md). Status: `compute-blocked`, paused by operator decision 2026-05-29. Current state: Branch E v1 cleared a modest capability floor (2 distinct held-out tasks/lane); v2's deterministic expansion replicated but did not lift it (top-2 crowding). E3 tests whether a learned MLP ranker over the FROZEN v2 candidate set beats that crowding. The runner is built and byte-verified: `enumerate_admitted` reproduces v2's candidate set exactly (identical `n_admitted` on all `be94b721`/`f25fbde4` instances; the `v2_deterministic_selector` control reproduces the v2 gated solves), with a 591-dim target-free feature schema, the RankerMLP, the five controls, a three-split no-target barrier, and shard/merge plumbing. py_compile + leak-check green; dry-run + candidate-gen sanity pass. Blocker: A single-shard timing probe (12 aux tasks -> 49 held-out instances, 38m45s) measured **~47.5 s/instance** -- the intrinsic frozen-v2 enumeration cost over uncurated auxiliary grids. The full ~892-task auxiliary pool extrapolates to **~48 h** of candidate generation (memory is fine: ~220k training rows, ~0.5 GB). The ~48 h sharded binding run was deliberately NOT launched. Next actions: 1. To resume, append the E3 tooling freeze-marker amendment, then either (a) stage + run the full sharded slate (32-48 shards, concurrent as cores allow; resumable shard+merge with a runner-content SHA audit / byte-equivalence to a serial run), or (b) first amend the spec to bound the aux pool (a deterministic sha256 sample of N tasks and/or a lower aux candidate budget) to cut the wall ~3-4x, then run. 2. Do not run the multi-hour/multi-day lock from an interactive coding-agent session; use a local operator slate or a dispatched runner. 3. On completion, adjudicate `branch_e3_ranker_*` on `learned_ranker` over `U_primary` vs `S_v1v2 = {be94b721, f25fbde4}`; keep all language as held-out public-training capability, not sufficiency / solve / eval / Kaggle. ## P2 Active Follow-Ups ### 6. Bayesian Floors Across Workbenches Sources: [`BAYESIAN_FLOOR_PROFILE_TEMPLATE.md`](BAYESIAN_FLOOR_PROFILE_TEMPLATE.md), [`../alignment.html`](../alignment.html), [`../internal/feedback/quarantine/attack_vectors.md`](../internal/feedback/quarantine/attack_vectors.md). Status: `design-blocked`, cross-cutting. Blocker: The Bayes reference should be a same-observation baseline inside each workbench, not a separate `bayes_v_sundog` storyline. Some workbenches still need their Bayesian-floor profile and readout shape. Next actions: 1. Add or update a Bayesian-floor row in each active workbench roadmap. 2. State the observation lane, privileged oracle lane, random/naive lane, and Bayes lane separately. 3. Keep public comparisons in the form "Sundog did X; Bayes did Y under the same information lane; oracle did Z with privileged access." ### 7. Geometry Calibration / Re-Anchoring Debt Sources: [`SUNDOG_V_GRAVITY.md`](SUNDOG_V_GRAVITY.md), [`calibration/PHASE10_OPTICAL_AUDIT_SYNTHETIC_MEMO.md`](calibration/PHASE10_OPTICAL_AUDIT_SYNTHETIC_MEMO.md), [`MESA_CROSSOVER_NOTE.md`](MESA_CROSSOVER_NOTE.md). Status: `operator-blocked`, not a launch blocker unless reopened by Cut 3. Blocker: The tangent/supralateral/pyramidal route still has specialist-photo and rendered-vs-anchored debt. The known discipline is to mark it as settled only at the strength earned by anchored receipts. Next actions: 1. If reopened, package p2/p7/p13/p27 anchors and the post-audit wording for specialist review. 2. Do not promote named-only or weak halo families into evidence without new anchored photos or HaloSim receipts. 3. Keep `sundog.html` and `legend.html` split: atlas for visual geometry, legend for vocabulary. ### 8. Application Evidence Assets Sources: [`../applications-gallery.html`](../applications-gallery.html), [`UI_UX_THEME_FOUNDATION.md`](site/UI_UX_THEME_FOUNDATION.md), [`WEBSITE_DEVELOPMENT.md`](site/WEBSITE_DEVELOPMENT.md). Status: `public-surface`, active. Outstanding assets: - Photometric Alignment: experiment video, graph package, short reviewer-facing summary. - EyesOnly / Gone Rogue: matched-seed multi-policy study, compressed-perception ablation, volatility sweep, behavior clips. - Dungeon Gleaner: orbit telemetry capture, side-by-side planner comparison, tuning-sensitivity sweep. - Money Bags: frozen terrain fixtures, disturbance scripts, metric glossary. - Pressure Mines / Balance / Pushable Occluder: keep rail posters and clips tied to their owning evidence docs; do not activate missing-evidence cards just to fill space. Next actions: 1. Prioritize one chart or clip per application that clarifies the evidence tier. 2. Keep gallery language as "surfaces" or "workbenches" unless a result doc earns stronger wording. 3. Replace decorative placeholders with claim-boundary visuals only when the visual adds interpretation. ### 9. Anniversary Rollout Aftercare Sources: [`../internal/outreach/anni_spam_roadmap.md`](../internal/outreach/anni_spam_roadmap.md), [`../internal/outreach/first_public_statement.md`](../internal/outreach/first_public_statement.md), [`../internal/feedback/quarantine/attack_vectors.md`](../internal/feedback/quarantine/attack_vectors.md), [`ANNIVERSARY_ROADMAP_TRIAGE.md`](ANNIVERSARY_ROADMAP_TRIAGE.md). Status: `public-surface`, active after posting. Blocker: The promotional circuit needs critique capture, not just more copy. Human feedback is already pointing at comprehension and commercial-value questions. Next actions: 1. Collect serious objections into `internal/feedback/quarantine/attack_vectors.md`. 2. Record recurring plain-language failures separately from technical objections. 3. Keep every broad public post attached to the failure-boundary sentence. 4. Add a young-reader/plain-language path only after the main copy stops accumulating contradiction and redundancy. ### 10. Canonical Routes / Deploy Verification Sources: [`WEBSITE_DEVELOPMENT.md`](site/WEBSITE_DEVELOPMENT.md), [`../public/_redirects`](../public/_redirects), [`../scripts/check-public-routes.mjs`](../scripts/check-public-routes.mjs). Status: `public-surface`, pre-deploy check. Blocker: Local route smoke cannot prove Cloudflare `_redirects` actually emits 301s for legacy `.html` URLs. A static dev server can pass by serving the old files directly. Next actions: 1. On the next Cloudflare preview deploy, run: ```powershell npm run site:routes -- --base .pages.dev ``` 2. Consider tightening the route checker so legacy `.html` paths must return a redirect to the canonical extensionless route, not just any `200`. 3. Keep sitemap, canonical tags, `og:url`, and `_redirects` in agreement before anniversary traffic. ## Deferred, But Do Not Lose ### Vortex / Wishing-Well Toy Source: [`../internal/theory/postulations.md`](../internal/theory/postulations.md). Status: `deferred`. This is a tempting cheap fourth test for coarse-graining language, but it should not replace the proof trunk or the structural-failure boundary work. Keep it as a candidate workbench only if it gets a Bayesian floor and a pre-registered failure boundary. ### Repo Map / Agent Navigation Surface Source: [`../repo-map.html`](../repo-map.html). Status: `deferred`. A living repo map could become a useful human and agent navigation surface, but it should not sit on top of the hero until it is accurate enough to route a visitor to evidence tiers, not just folders. ## v0.3 Cross-m_3 Sentinel (II) ### V0.3 Gamma cross-m_3 sentinel sweep (~3-4 hours staged compute) Sources: [`../docs/isotrophy/kfacet/kfacet_v03_gamma_crossm3_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v03_gamma_crossm3_preregistration.md), [`../docs/isotrophy/kfacet/kfacet_v03_freeze_b_comparison.md`](../docs/isotrophy/kfacet/kfacet_v03_freeze_b_comparison.md), [`../docs/isotrophy/kfacet/kfacet_v03h_writeup.md`](../docs/isotrophy/kfacet/kfacet_v03h_writeup.md). Status: **VERDICT LANDED 2026-05-22**. Sentinel sweep executed against the pre-registration; joint verdict `(Q1.D, Q2.D) = gate pathology on both axes`. Resolved by the targeted `sigma_3-scan` symmetry probe: all seven supp-B sentinels fail `sigma_3`; the initial coarse-gauge read placed six of seven in `Z_2 = (12)`-swap and flagged `O_434(0.4)` as smaller-symmetry, but v0.4a0 later showed `O_434` was a gauge artifact and v0.4a classified all 273 supp-B rows as `Z2_clean`. v0.3 `Gamma_i` is `D_3`-equivariant by construction and structurally inapplicable to this catalog. **Domain-of-applicability finding**, not within-domain falsification. v0.3 epilogue closed; v0.4 chapter opens on `Z_2`-equivariant mechanisms (paper-side design first, no runner code until representation theory closes). Current state: - `alpha` closed the strict-G.2 m_3 = 1 cell of a **2 x 2 design** as structural null: zero daughters predicted from 20 resolved strict-G.2 rows against 273 observed piano-trios. II tests the remaining two testable cells. - The 2 x 2 has rows `(m_3 = 1, m_3 != 1)` and columns `(strict G.2 family, piano-trio family)`. Strict G.2 is only defined at m_3 = 1; the off-equal-mass strict-G.2 cell is empty by construction. - The two slices answer two structurally separate questions: - **Q1 (m_3 axis):** does `Gamma` wake up where supp-B is densest? m_3 = 0.4 carries 55 of 273 rows. - **Q2 (family axis):** do piano-trio orbits carry the standard-E signal even though strict-G.2 seeds did not? The 38 supp-B rows at m_3 = 1 are entirely disjoint from the strict G.2 21 indices. - Pre-registration freezes the sentinel subset (4 rows at m_3 = 0.4, 3 rows at supp-B m_3 = 1.0), the per-axis outcome alphabet, and the joint verdict table over `(Q1, Q2)` before any run. The four main cases are `(A,A) = everywhere null`, `(A,B) = family-dependent`, `(B,A) = m_3-dependent`, `(B,B) = G.2 corner is unique`. Blocker: Each sentinel row takes 20-45 minutes wall-clock at rtol=1e-12. Seven sentinels total = ~3-4 hours of staged compute. Above the inline ~10-minute rule. Next actions: 1. Operator verifies parse: `npm run isotrophy:parse:b` returns 273 rows; sentinel indices match the pre-registration tables. 2. Operator runs the 7 sentinel commands listed in the pre-registration (4 at m_3=0.4 + 3 at supp-B m_3=1.0). Each is staged separately. 3. Operator runs the reprocessor and bridge-audit commands over the sentinel receipts. 4. Read the per-axis outcomes `(Q1, Q2)` against the pre-registered alphabet `{A null, B signal, C quarantine, D gate-pathology}`. Resolve any C / D outcomes via per-row WHY-dive / runner diagnosis so the joint verdict reads over `(Q1, Q2) in {A, B}`. Then branch on the joint verdict: - `(A, A)` everywhere null -> stage full 55-row m_3 = 0.4 sweep as a follow-on operator command; if also null, write the v0.3 epilogue. - `(A, B)` family-dependent -> halt; open family-axis review (piano-trios vs strict G.2). - `(B, A)` m_3-dependent -> halt; open m_3-axis review (F_beta cocycle off equal mass). - `(B, B)` G.2 corner is unique -> halt; open joint review (alpha verdict may be a rigidity artifact of the strict G.2 m_3 = 1 cell). Stop condition: If sentinels surface a positive standard-E sector at any m_3, do not extend to full slices before the redesign review lands. The alpha verdict on m_3=1 strict-G.2 stands either way. ## v0.4a Domain Map (VERDICT LANDED) ### V0.4a Z_2 domain map for supplementary-B (~85-120 min staged) Status: **VERDICT LANDED 2026-05-22**. Sweep executed 22:21 -> 02:46 (~4 h 25 min). Final verdict: **outcome_A_all_Z2_clean**. 273/273 rows in `Z2_clean` after Pass 2 (24 rows rescued from provisional `smaller_symmetry`, validating the two-pass design). `marginal_Z2 / smaller_symmetry / undefined` bands all empty. v0.4 body now locked as **"supplementary-B piano-trio orbit as primary Z_2 object; domain: all 273 rows"**. See the **Verdict (Landed)** section of the pre-registration for the per-row provenance and Pass 2 rescue table. Methodology rule recorded: long-period rows (T >= ~170) at m_3 in {1.4, 1.5, 1.6, 1.7} stress the coarse SO(3) gauge minimizer; treat default tolerances as a screening pass, not a final classifier. Next: **v0.4b mechanism predictor** (closed structural-negative; see the section below). ### V0.4a pre-registration reference Sources: [`../docs/isotrophy/kfacet/kfacet_v04a_domain_map_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v04a_domain_map_preregistration.md) (includes the Verdict (Landed, 2026-05-22) section with per-row provenance and the Pass 2 rescue table), [`CROSS_SUBSTRATE_NOTES.md`](CROSS_SUBSTRATE_NOTES.md) §7.2 (projection-language framing for the v0.3 -> v0.4 transition). Receipts: ```text results/isotrophy/k-facet-v04a-domain-map/manifest.json results/isotrophy/k-facet-v04a-domain-map/pass1/{aggregator_manifest, flagged_for_pass2.csv, m3eq*/} results/isotrophy/k-facet-v04a-domain-map/pass2/m3eq{0.4, 1.4, 1.5, 1.6, 1.7}/O{idx}/ ``` ## v0.4b Mechanism Predictor (chapter closed) ### V0.4b gamma_3 / gamma_3' non-circular structural predictors (closed structural-negative) Sources: [`../docs/isotrophy/kfacet/kfacet_v04b_mechanism_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v04b_mechanism_preregistration.md), [`../docs/isotrophy/kfacet/kfacet_v04a_domain_map_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v04a_domain_map_preregistration.md) (domain locked: 273/273 Z2_clean), [`../docs/isotrophy/kfacet/kfacet_v04b_gamma3prime_form.md`](../docs/isotrophy/kfacet/kfacet_v04b_gamma3prime_form.md), [`../docs/isotrophy/kfacet/kfacet_v04_writeup.md`](../docs/isotrophy/kfacet/kfacet_v04_writeup.md), [`CROSS_SUBSTRATE_NOTES.md`](CROSS_SUBSTRATE_NOTES.md) §7.2 (projection-language framing). Status: **CHAPTER CLOSED 2026-05-23, structural-negative**. Both registered Z_2-shadow predictors failed their pre-registered falsifiers without committing the compute-blocked 2-hour sweep. `gamma_3` (tangent-isotypic) retired pre-sweep when the 7-row sanity probe found `F_beta` does not preserve `K_fib` on supp-B (verdict `form_precondition_failed`). `gamma_3'_orbit_pass2` (orbit gauge-rigidity) registered and immediately tested against the v0.4a manifest (no new compute); falsified by `chi^2 = 1202.32` vs critical `26.22` (46x threshold). Rule accuracy 63.74% lies slightly below the always-U baseline 64.47%. Stability is statistically independent of the Z_2 shadow at both tested granularities. Full chapter close at [`../docs/isotrophy/kfacet/kfacet_v04_writeup.md`](../docs/isotrophy/kfacet/kfacet_v04_writeup.md); gamma_3'_orbit_pass2 verdict receipt at `results/isotrophy/k-facet-v04b-gamma3prime-orbit-pass2/manifest.json`. Current state: - v0.4a closed with verdict `outcome_A_all_Z2_clean`. The 273 supp-B piano-trio rows uniformly carry `Z_2 = (12)`-swap symmetry after the pre-registered two-pass gauge classifier. - v0.4b found a new structural gap: v0.4a proves `F_beta` is an orbit-level symmetry of supplementary-B piano-trios, but the at-anchor `F_beta` operator used by the locked form is **not** a tangent-level symmetry of `K_fib = ker(M_i - I) / N_C` on the tested rows. - Seven distinct sanity rows all show `F_beta` leakage out of `K_fib` at order `1e-1` (median approximately `0.45`) and projector overcount of `+1` to `+3` directions. The naive `(I +/- F_beta_K)/2` projectors are therefore not valid isotypic projectors on supp-B `K_fib`. - The locked threshold rule `predict S iff F_beta_even_dim >= F_beta_odd_dim` is retired **before** the 273-row sweep. It is not a failed chi-squared result; the precondition failed first. - `gamma_3'` replacement selected the orbit-feature lane: `gamma_3'_orbit_pass2`, predicting S iff the row required Pass 2 rescue in the v0.4a two-pass gauge classifier. It was tested inline from the v0.4a manifest and falsified (`chi^2 = 1202.32` vs `critical = 26.22`; accuracy 63.74%, below always-U 64.47%). Closed blockers: 1. ~~**Paper-side** gamma_3 form lock~~ -- DONE 2026-05-22, then RETIRED 2026-05-23. Threshold-rule baseline locked at [`../docs/isotrophy/kfacet/kfacet_v04b_gamma3_form.md`](../docs/isotrophy/kfacet/kfacet_v04b_gamma3_form.md): `predict S iff F_beta_even_dim >= F_beta_odd_dim`. Zero free parameters. Chi-squared df = 12, critical value `26.22` at `p = 0.01`. Retired because `F_beta` does not preserve `K_fib` on the sanity rows. 2. ~~**Paper-side**: choose and pre-register a replacement `gamma_3'` projection target.~~ DONE 2026-05-23. The `orbit_features` lane was selected after the cocycle-rescue probe returned identity cocycles. 3. ~~**Implementation/compute**: deferred.~~ The original 273-row `kfacet-row-z2-sweep` plan is historical only and should not be run for the retired baseline; the selected `gamma_3'_orbit_pass2` verdict used the existing v0.4a manifest with no new compute. Next actions: 1. ~~Lock gamma_3 form paper-side.~~ DONE 2026-05-22, retired 2026-05-23 by `form_precondition_failed`. 2. ~~Choose replacement lane.~~ DONE 2026-05-23. The cocycle-rescue probe surfaced `(R_i, phi_i) = (I, 0)` uniformly, ruling out `anticomm`-style lanes. `orbit_features` selected; baseline locked at `gamma_3'_orbit_pass2` (predict S iff Pass 2 rescue required). 3. ~~Run the verdict.~~ DONE 2026-05-23, inline from the v0.4a manifest (no new compute). Verdict: **falsifies_baseline_orbit_pass2_rule** at `chi^2 = 1202.32` vs `critical = 26.22`. 4. ~~Land the chapter close.~~ DONE 2026-05-23. v0.4 epilogue writeup at [`../docs/isotrophy/kfacet/kfacet_v04_writeup.md`](../docs/isotrophy/kfacet/kfacet_v04_writeup.md). 5. ~~**Paper-side v0.5 direction (open).**~~ DONE 2026-05-23. v0.5 opens as a **branch-shadow audit**: catalog-only branch hash on `(m_3, z_0)` indicator bits, chi-squared independence vs S/U at `df = occupied_branch_count - 1`. Frame: stop asking what symmetry the row has, start asking which branch the row belongs to. See [`../docs/isotrophy/kfacet/kfacet_v05a_branch_map_form.md`](../docs/isotrophy/kfacet/kfacet_v05a_branch_map_form.md). See the v0.5a section below for the runner state. Stop condition: If a proposed `gamma_3'` still requires an isotypic decomposition on a subspace not preserved by the relevant `F_beta` action, retire it at the precondition stage rather than launching a catalog sweep. ## v0.5a Branch-Shadow Audit (registered) ### V0.5a branch-shadow audit (chi-squared independence on the active branch hash) Status: **VERDICT LANDED 2026-05-23**. The 4x2 branch-shadow audit passes: `chi^2 = 34.986` vs critical `11.34` (`df = 3`, p ~= `1.23e-7`). Receipt: `results/isotrophy/k-facet-v05a-branch-map/manifest.json`. Frame: stop asking what symmetry the row has, start asking which branch the row belongs to. v0.5a is an AUDIT (chi-squared independence of a branch hash against S/U), NOT a predictor. A pass licenses v0.5b registration as a true predictor with held-out test; a fail joins v0.4 as another projection-limit negative and the chain moves to the next mechanism family. Locked form (`docs/isotrophy/kfacet/kfacet_v05a_branch_map_form.md`): ```text 4-bit candidate signature, with deterministic constant-bit retirement: b1 = (m_3 < 1) ACTIVE (169 / 273 true on supp-B) b2 = (z_0 < 0.3) ACTIVE (130 / 273 true on supp-B) b3 = (abs(v_z) < 1e-6) RETIRED (constant FALSE on supp-B) b4 = (m_3 * z_0^2 < 2) RETIRED (constant TRUE on supp-B) active signature on supp-B: (b1, b2) occupied buckets: 4 df = occupied_branch_count - 1 = 3 critical: chi-squared(3) at p = 0.01 = 11.34 verdicts: chi^2 > 11.34: branch_hash_passes_audit chi^2 <= 11.34: branch_hash_fails_audit ``` Catalog-only input; no new dynamical compute is required. The audit runs against the v0.4a manifest at `results/isotrophy/k-facet-v04a-domain-map/manifest.json`. 4x2 result: ```text branch_label N S U S_fraction m_3 >= 1, z_0 >= 0.3 87 18 69 0.2069 m_3 >= 1, z_0 < 0.3 17 5 12 0.2941 m_3 < 1, z_0 >= 0.3 56 11 45 0.1964 m_3 < 1, z_0 < 0.3 113 63 50 0.5575 ``` Next actions: 1. ~~Implement the audit aggregator.~~ DONE: [`../scripts/v05a_branch_map_audit.py`](../scripts/v05a_branch_map_audit.py). 2. ~~Run the audit blind.~~ DONE. Verdict: `branch_hash_passes_audit`. 3. ~~Land the verdict.~~ DONE in [`../docs/isotrophy/kfacet/kfacet_v05a_branch_map_form.md`](../docs/isotrophy/kfacet/kfacet_v05a_branch_map_form.md). 4. Register v0.5b paper-side with held-out m_3 bins or leave-one-branch-out CV. No predictor claim is licensed until v0.5b is separately registered and tested. Stop condition: Any post-audit narrowing of the active signature, threshold change, or df-formula change is a **re-registration**, not a refinement. The audit form is locked. ## v0.5b Branch Predictor (held-out verdict) ### V0.5b branch-shadow predictor (leave-one-m_3-bin-out) Status: **VERDICT LANDED 2026-05-23**. Paper-side form: [`../docs/isotrophy/kfacet/kfacet_v05b_branch_predictor_form.md`](../docs/isotrophy/kfacet/kfacet_v05b_branch_predictor_form.md). Receipt: `results/isotrophy/k-facet-v05b-branch-predictor/manifest.json`. Verdict: **branch_predictor_fails_heldout**. The v0.5a branch hash stratifies S/U in-sample, but the fold-trained branch-majority predictor does not beat always-`U` under leave-one-`m_3`-bin-out. Locked form: ```text primary partition: leave-one-m_3-bin-out over the 12 bins with N >= 5 predictor form: fold-trained branch-majority on active (m_3 < 1, z_0 < 0.3) tie rule: predict U baseline: always-U primary gate: one-sided exact McNemar/binomial test on discordant rows pass threshold: p <= 0.01 AND positive accuracy delta continuous fields: diagnostic-only; reserved for v0.5c ``` Primary result: ```text gate rows: 263 (95 S / 168 U) model accuracy: 0.619772 always-U acc: 0.638783 accuracy delta: -0.019011 McNemar: win = 28, loss = 33, n_disc = 61, p = 1.0 verdict: branch_predictor_fails_heldout ``` Reading: `m_3 = 0.4` was the load-bearing fold. Holding it out flips the low-mass/low-`z_0` branch training direction to `U` (`28 S / 33 U`), missing the large stable `0.4` block. In the remaining low-mass folds, the same branch usually predicts `S`, but its held-out wins and losses nearly cancel. The random catalog-half split and the single rule `predict S iff (m_3 < 1 and z_0 < 0.3)` are registered only as sidecars. The primary gate respects the mass-bin structure rather than mixing same-`m_3` rows across train and test. Next actions: 1. ~~Sign off or revise the held-out predictor form before any runner is executed.~~ DONE 2026-05-23. 2. ~~Implement `scripts/v05b_branch_predictor.py` against `results/isotrophy/k-facet-v05a-branch-map/per_row_table.csv`.~~ DONE 2026-05-23. 3. ~~Run the seconds-scale held-out predictor and land the receipt at `results/isotrophy/k-facet-v05b-branch-predictor/`.~~ DONE 2026-05-23. 4. ~~Land the chapter close.~~ DONE 2026-05-23. v0.5 epilogue writeup at [`../docs/isotrophy/kfacet/kfacet_v05_writeup.md`](../docs/isotrophy/kfacet/kfacet_v05_writeup.md). Chapter type: **projection-limit (audit-passes-predictor-fails)**. The v0.5a in-sample positive (chi^2 = 34.99) and v0.5b held-out negative (accuracy_delta = -0.019) together constrain the branch shadow as descriptive, not predictive. 5. ~~**Paper-side v0.6 direction (open).**~~ DONE 2026-05-23. v0.6 opens with the **conserved-quantity (E, |L|) stratification** family. Parent registration at [`../docs/isotrophy/kfacet/kfacet_v06_mechanism_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v06_mechanism_preregistration.md). See the v0.6 section below for the next-action queue. Stop condition: If a proposed v0.6 mechanism requires re-using the same active feature set that failed v0.5b held-out, retire it at the registration stage rather than launching a sweep. v0.5b's pre-registered asymmetric falsifier (McNemar p AND positive delta) is the inheritance discipline. ## v0.6 Conserved-Quantity (E, |L|) Stratification (parent + v0.6a + v0.6b verdicts landed) ### V0.6 conserved-quantity (E, |L|) mechanism family Status: **PARENT REGISTERED 2026-05-23; v0.6a VERDICT LANDED 2026-05-23; v0.6b VERDICT LANDED 2026-05-24**. Parent registration: [`../docs/isotrophy/kfacet/kfacet_v06_mechanism_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v06_mechanism_preregistration.md). **v0.6a verdict**: `energy_quartile_passes_audit_alignment_warning` (chi^2 = 33.70 vs 11.34, alignment = 0.956). Form/verdict: [`../docs/isotrophy/kfacet/kfacet_v06a_energy_quartile_audit_form.md`](../docs/isotrophy/kfacet/kfacet_v06a_energy_quartile_audit_form.md). **v0.6b verdict**: `within_branch_energy_fails_audit` (chi^2 = 6.90, permutation p = 0.029; sparse-cell fallback fired with min_expected = 2.66; |L| sidecar at permutation p = 0.074 also below loud-signal threshold). Form/verdict: [`../docs/isotrophy/kfacet/kfacet_v06b_within_branch_energy_audit_form.md`](../docs/isotrophy/kfacet/kfacet_v06b_within_branch_energy_audit_form.md). v0.6c (held-out predictor) is NOT licensed; v0.6 chapter is positioned for chapter-close writeup as a clean conditional-independence result. Frame: v0.4 ruled out the Z_2 shadow as a stability projector. v0.5 ruled out the 2-bit catalog branch shadow as a held-out predictor. v0.6 asks whether a higher-dimensional catalog-coordinate projection — orbit-level conserved quantities `(E, |L|)` — carries held-out stability. Locked parent shape: ```text Body: supp-B piano-trio orbit as primary conserved-quantity object (273 rows, Z2_clean per v0.4a). Projection: row -> (E(row), |L|(row)) computed from initial conditions and the three-body Hamiltonian. Observable: per-row (E, |L|) joined with S/U stability label, binned into a pre-registered (n_E x n_L) contingency. Compute: catalog-only, seconds (no orbit integration required). Discipline: inherits v0.5 (audit-then-predictor, asymmetric McNemar+delta falsifier, leave-one-m_3-bin-out partition default, conservative tie rule, constant-feature retirement). ``` v0.6a locked shape: ```text primary feature: E(row) = total energy at IC, computed from (m_1, m_2, m_3, r_i, v_i) and Hamiltonian. binning: quartiles over supp-B's 273-row E distribution. quantile method: numpy.quantile(E, [0.25, 0.50, 0.75], method='linear'). contingency: 4 x 2 over (Q_E, S/U), df = 3. critical: 11.34 at p = 0.01. verdicts: energy_quartile_passes_audit (chi^2 > 11.34) | energy_quartile_passes_audit_alignment_warning (chi^2 > 11.34 AND max-bin-alignment > 0.8) | energy_quartile_fails_audit (chi^2 <= 11.34). sidecar: |L| quartile chi-squared under same shape, REPORT-ONLY (no verdict claim). alignment guard: receipt MUST report max over Q_E of (fraction in any single v0.5a branch_label bucket). ``` v0.6a result: ```text receipt: results/isotrophy/k-facet-v06a-energy-quartile-audit/manifest.json sanity: PASS (max |Delta E| = 0, max |Delta |L|| = 0) bound check: PASS (E < 0 for all 273 rows) E audit: chi^2 = 33.703158, p = 2.29e-7, critical = 11.34 alignment_E: 0.955882 (> 0.8 warning threshold) verdict: energy_quartile_passes_audit_alignment_warning |L| sidecar: chi^2 = 28.954252, p = 2.29e-6, report-only alignment_|L|: 0.956522 ``` Interpretation: E quartiles carry strong in-sample S/U signal, but the registered alignment scalar says the signal is tightly entangled with the v0.5a `(m_3, z_0)` branch shadow. v0.6b must therefore be registered with an alignment-breaking partition (for example leave-one-E-quartile-out, or a constant-m_3 subcatalog design), not the default leave-one-m_3-bin-out partition. Completed actions: 1. Per-row E and |L| computation implemented in `scripts/v06a_energy_quartile_audit.py`. 2. Sanity check passed. The receipt records that local cross_m_3 receipts do not carry scalar E/|L| fields, so the parity reference is the pre-existing v0.3 `scripts.isotrophy_workbench` invariant implementation. 3. Bound-orbit check passed. 4. Per-row `(label, m_3, z_0, E, |L|, Q_E, Q_|L|, branch_label, stability)` table emitted. 5. v0.6a verdict landed. v0.6b result: ```text receipt: results/isotrophy/k-facet-v06b-within-branch-energy/manifest.json stratum: 113 rows (63 S / 50 U) sanity: PASS (N, S, U match locked counts) within-branch Q_E contingency: Q1: N=0 (Q1 entirely outside stratum -- tightest catalog orbits all sit in other branches) Q2: N=6 S=2, U=4 S_frac=0.333 Q3: N=42 S=18, U=24 S_frac=0.429 Q4: N=65 S=43, U=22 S_frac=0.662 within-branch Q_E x m_3: Q4 = m_3 in {0.4, 0.5} (52 of 65 are m_3 = 0.4) Q3 = m_3 in {0.5, 0.6, 0.7, 0.8} Q2 = m_3 in {0.8, 0.9} -- Q_E is essentially a label for m_3 sub-bin inside the stratum. E primary: chi^2 = 6.904228 fallback fired (min_expected = 2.655 < 5); permutation test seed = 20260523, n_permutations = 10000; permutation p = 0.0292 verdict: within_branch_energy_fails_audit (p > 0.01) |L| sidecar: chi^2 = 4.464524 permutation p = 0.0741 report-only; does NOT meet the loud-signal threshold (p <= 0.01) for a fresh v0.6c |L| form lock. ``` Interpretation: the within-branch direction is monotone in the same direction as the catalog-wide v0.6a finding (Q2: 33% S, Q3: 43% S, Q4: 66% S), but the permutation p = 0.029 does not clear the pre-registered p <= 0.01 floor. The v0.6a in-sample chi-squared was dominated by branch-shadow content; energy quartile is essentially a 1-to-1 label for m_3 sub-bin within the stratum, and the residual within-branch stratification is below the gating floor. The energy-shadow-as-mechanism sub-question closes; the |L| sidecar at p = 0.074 does NOT trigger a fresh form lock either. Completed actions: 1. ~~Implement `scripts/v06b_within_branch_energy_audit.py`.~~ DONE 2026-05-24. 2. ~~Run the v0.6b audit blind; land verdict.~~ DONE 2026-05-24. Next actions: 1. ~~Author the v0.6 chapter close writeup.~~ DONE 2026-05-24. v0.6 epilogue at [`../docs/isotrophy/kfacet/kfacet_v06_writeup.md`](../docs/isotrophy/kfacet/kfacet_v06_writeup.md). Chapter type: **conditional-independence close** (distinct from v0.4 structural-negative and v0.5 projection-limit). 2. ~~**Paper-side v0.7 direction (open).**~~ DONE 2026-05-24. v0.7 opens with the **gamma_1 direction-of-instability** family. Parent registration at [`../docs/isotrophy/kfacet/kfacet_v07_mechanism_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v07_mechanism_preregistration.md). See the v0.7 section below for the runner state. Stop condition: If the within-branch chi-squared has a permutation p > 0.01 but the residual structure suggests a coarser binning would pass, the pre-registered re-registration path is a fresh form lock with explicitly justified coarser bins (e.g., median-split with df=1) -- NOT a post-hoc loosening of v0.6b. v0.6b's sparse-cell fallback discipline is permanent. ## v0.7 gamma_1 Direction-of-Instability (chapter closed, qualified-positive) ### V0.7 gamma_1 direction-of-instability mechanism family Status: **CHAPTER CLOSED 2026-05-24, qualified-positive on restricted domain.** Parent registration: [`../docs/isotrophy/kfacet/kfacet_v07_mechanism_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v07_mechanism_preregistration.md). Chapter close: [`../docs/isotrophy/kfacet/kfacet_v07_writeup.md`](../docs/isotrophy/kfacet/kfacet_v07_writeup.md). **v0.7a catalog-wide verdict**: `velocity_fraction_blocked_integration_attrition` at `integration_blocked_count = 23` (8.42% of catalog > 5% threshold). Two amendments applied: R1 (symplecticity gate 1e-6 -> 1e-4) and R2.A (per-row integration-failure fallback). Form lock + verdict: [`../docs/isotrophy/kfacet/kfacet_v07a_velocity_fraction_audit_form.md`](../docs/isotrophy/kfacet/kfacet_v07a_velocity_fraction_audit_form.md). **v0.7a' restricted-scope confirmation verdict**: `velocity_fraction_restricted_passes_audit` [PASS] at `chi^2 = 16.43` (critical 11.34, df=3, p = 9.3e-4) on the 250 analyzable rows, with alignment-tightness 0.698 against the v0.5a branch hash (below the 0.8 warning floor). Non-monotone U-shape signature: Q1 (gamma_1 mostly positional) 49% S, Q2 (mixed) 18% S, Q3 29% S, Q4 (gamma_1 mostly velocity) 43% S. Form lock + verdict: [`../docs/isotrophy/kfacet/kfacet_v07a_prime_restricted_scope_form.md`](../docs/isotrophy/kfacet/kfacet_v07a_prime_restricted_scope_form.md). The Floquet velocity-fraction direction shadow is the **first** low-dimensional projection in the v0.4 / v0.5 / v0.6 / v0.7 envelope to produce a non-branch-aligned positive signal at p < 0.01 on the analyzable sub-catalog. Catalog-wide generalizability is limited by the integration-attrition. ## v0.8 Floquet Direction-Purity (parent + v0.8a verdict landed) ### V0.8 Floquet direction-purity mechanism family Status: **PARENT REGISTERED 2026-05-24; v0.8a VERDICT LANDED 2026-05-24 (FAIL).** Parent registration: [`../docs/isotrophy/kfacet/kfacet_v08_mechanism_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v08_mechanism_preregistration.md). **v0.8a verdict**: `purity_quartile_fails_audit` at `chi^2 = 4.94`, `p = 0.176` (df=3, critical 11.34). Form lock + verdict: [`../docs/isotrophy/kfacet/kfacet_v08a_purity_quartile_audit_form.md`](../docs/isotrophy/kfacet/kfacet_v08a_purity_quartile_audit_form.md). The unsigned purity transform `abs(vf - 0.5)` does NOT capture the v0.7a' positive at the registered floor; the diagnostic purity_signed contingency reproduces v0.7a's `chi^2 = 16.43` exactly, confirming the signal lives in the **signed** direction. The failure is explained by the vf distribution's density asymmetry (184/250 rows below vf = 0.5; max vf = 0.87 -- the "vf near 1" pure end is structurally under-populated). v0.8b (held-out predictor) is NOT licensed under the fails verdict. Frame: v0.7a' produced a non-monotone U-shaped signal (direction- purity correlates with stability; mixed directions correlate with instability). v0.8 promotes purity = `abs(vf - 0.5)` to the primary operational quantity and asks whether it predicts S/U **held-out** on the 250-row analyzable subset. Locked parent shape: ```text Body: the 250-row analyzable supp-B subset from v0.7a (rows where compute_monodromy_vectorized completed). Attrition carried as a permanent domain restriction. Projection: row -> purity(row) = abs(vf(row) - 0.5) inheriting v0.7a's vf operational definition (gamma_1 selection rule, tie-break cascade, CoM reduction, mass-weighted norm). Observable: per-row (purity, S/U) on 250 rows joined with v0.5a branch_label for alignment-tightness diagnostics. Discipline: inherits v0.5/v0.6/v0.7 (audit-then-predictor, asymmetric McNemar+delta predictor falsifier, alignment-tightness guard, sparse-cell fallback, conservative tie rule, constant-feature retirement). Non-circ: re-asserts v0.7a's locked sentence on Floquet eigenvectors used as geometric directions only. ``` Completed actions: 1. ~~Pick v0.8a form.~~ DONE 2026-05-24. Locked A (purity-quartile audit). 2. ~~Implement `scripts/v08a_purity_audit.py`.~~ DONE 2026-05-24. 3. ~~Run blind; land verdict.~~ DONE 2026-05-24. Verdict: `purity_quartile_fails_audit`. Result interpretation: ```text v0.8a purity quartile contingency: Q1 N=63 S=21 U=42 33.3% S (vf ~ 0.44-0.56, mid) Q2 N=62 S=18 U=44 29.0% S Q3 N=62 S=19 U=43 30.6% S Q4 N=63 S=29 U=34 46.0% S (vf far from 0.5, high purity) ----- chi^2 = 4.94 (p = 0.176) v0.8a diagnostic purity_signed contingency (reproduces v0.7a' exactly since linear transform of vf): chi^2 = 16.43 asymmetry: |S(Q1_signed) - S(Q4_signed)| = 0.064 (< 0.2 flag) asymmetric_u_shape_flag: False vf distribution: vf < 0.5: 184 / 250 (73.6%) vf > 0.5: 66 / 250 (26.4%) vf max: 0.8653 (no rows near vf = 1) ``` The U-shape signal IS real (purity_signed reproduces v0.7a's chi^2 = 16.43) but is **NOT captured by unsigned direction-purity**. The signal lives in the signed vf direction. Density asymmetry of the vf distribution (74% below 0.5) violates the purity transform's symmetry assumption. Next action: ~~paper-side direction call.~~ DONE 2026-05-24. v0.8 chapter closed as structural-negative on direction-purity (chapter close at [`../docs/isotrophy/kfacet/kfacet_v08_writeup.md`](../docs/isotrophy/kfacet/kfacet_v08_writeup.md)); v0.9 opened on signed Floquet direction-composition with explicit anti-circular framing. ## v0.9 Signed Floquet Direction-Composition (parent + v0.9a verdict landed) ### V0.9 signed Floquet direction-composition mechanism family Status: **PARENT REGISTERED 2026-05-24; v0.9a VERDICT LANDED 2026-05-24 (FAIL on both chi^2 and U-shape pattern; substantively informative).** Parent registration: [`../docs/isotrophy/kfacet/kfacet_v09_mechanism_preregistration.md`](../docs/isotrophy/kfacet/kfacet_v09_mechanism_preregistration.md). **v0.9a verdict**: `signed_vf_three_zone_fails_audit_chi2` at chi^2 = 7.42 vs critical 9.21 (df=2, p = 0.0245). Form lock + verdict: [`../docs/isotrophy/kfacet/kfacet_v09a_signed_vf_three_zone_form.md`](../docs/isotrophy/kfacet/kfacet_v09a_signed_vf_three_zone_form.md). The U-shape pattern check ALSO fails (mixed > positional, not < positional). Substantive read: **the v0.7a' U-shape was a quartile-boundary artifact at vf ~ 0.297**; under physical cutpoint at vf = 0.25 the pattern is **monotone-increasing** (positional 11% S, mixed 34% S, velocity-heavy 44% S) but does not reach the chi-squared floor. The v0.7a' chi^2 = 16.43 catalog-coordinate positive remains valid; the U-shape MECHANISM hypothesis is invalidated. v0.9b is NOT licensed. Frame: v0.7a' produced chi^2 = 16.43 (p = 9.3e-4) on vf quartiles vs S/U. v0.8a tested unsigned purity = `abs(vf - 0.5)` and failed; the v0.8a diagnostic confirmed the signal is signed. v0.9 promotes signed direction-composition (raw vf without folding) and asks **what specific pattern in vf vs S/U is testable beyond v0.7a's chi^2 of independence**. **Anti-circular framing discipline (load-bearing, locked by parent):** Any v0.9a form that: 1. Uses vf-ordering as feature, AND 2. Uses chi^2 of independence as test statistic, AND 3. Tests on the v0.7a 250-row analyzable subset **reproduces v0.7a's chi^2 = 16.43 by linear-transform invariance** and is REJECTED PRE-COMPUTE. To be honest, v0.9a must break at least one of (feature derivation, test statistic, evaluation domain). Candidate v0.9a forms: ```text A. Ordered quartile audit. REJECTED (re-test of v0.7a'). B. Binary vf < 0.5 threshold. REJECTED (binary collapse). C. Three-zone audit with PHYSICAL cutpoints (not v0.7a' quartiles). LIVE (cutpoint motivation required). D. Ordinal trend + nonmonotonicity test (e.g., Jonckheere-Terpstra). LIVE (cleanest, tests hypothesis-specific pattern not chi^2 of independence). ``` Completed actions: 1. ~~Pick v0.9a form.~~ DONE 2026-05-24. Locked C with physical cutpoints {0.25, 0.50} (positional-dominant / mixed / velocity-heavy). Anti-circular discipline: two of three conditions broken (feature derivation via physical cutpoints, test derivation via pattern-specific verdict tree). 2. ~~Implement `scripts/v09a_signed_vf_three_zone_audit.py`.~~ DONE. 3. ~~Run blind; land verdict.~~ DONE. Result: ```text verdict: signed_vf_three_zone_fails_audit_chi2 chi^2: 7.42 vs critical 9.21 (df=2, p=0.01) p_value: 0.0245 alignment_tightness: 0.5697 three-zone contingency: positional-dominant (vf < 0.25): N=19 S_frac=0.1053 mixed (0.25 <= vf < 0.50): N=165 S_frac=0.3394 velocity-heavy (vf >= 0.50): N=66 S_frac=0.4394 U-shape pattern check: mixed < positional? FALSE (0.34 > 0.11) mixed < velocity? TRUE (0.34 < 0.44) u_shape_confirmed? FALSE ``` Substantive interpretation: the v0.7a' U-shape was a quartile- boundary artifact at vf ~ 0.297. Under physical cutpoint vf = 0.25, the positional-dominant zone shrinks from 63 to 19 rows and its S-fraction collapses from 49% to 11%. The actual physical pattern is **monotone-increasing** in vf: positional < mixed < velocity- heavy. The v0.7a' chi^2 = 16.43 catalog-coordinate result remains valid; the U-shape MECHANISM hypothesis is invalidated. Next action: ~~paper-side direction call.~~ DONE 2026-05-24. v0.9 chapter closed; isotrophy program paused at end-of-v0.9. Chapter close at [`../docs/isotrophy/kfacet/kfacet_v09_writeup.md`](../docs/isotrophy/kfacet/kfacet_v09_writeup.md); program pause at [`../docs/isotrophy/kfacet/kfacet_isotrophy_program_pause.md`](../docs/isotrophy/kfacet/kfacet_isotrophy_program_pause.md). ## Isotrophy Program PAUSED (2026-05-24) The K_facet isotrophy program paused at end-of-v0.9 after seven sequential pre-registered chapters with seven distinct chapter-close types. Load-bearing substantive finding: v0.7a' chi^2 = 16.43 positive on the 250-row analyzable subset (branch-independent at alignment 0.70). Load-bearing meta-finding: under physical zones, the actual pattern is monotone-increasing in vf (positional 11% S, mixed 34% S, velocity-heavy 44% S); but chi-squared on three zones (7.42) does not pass p = 0.01. Program pause document: [`../docs/isotrophy/kfacet/kfacet_isotrophy_program_pause.md`](../docs/isotrophy/kfacet/kfacet_isotrophy_program_pause.md). Bandwidth redirects to three-body Phase 15+ (survival pocket) and the coarse-graining proof trunk per the original direction call. ### Lab Initiate Reopening Avenues (waste compute on) Eight concrete reopening directions for any lab initiate with bandwidth to spare. Full descriptions, hypotheses, implementation sketches, and non-circularity arguments at [`../docs/isotrophy/kfacet/kfacet_isotrophy_program_pause.md`](../docs/isotrophy/kfacet/kfacet_isotrophy_program_pause.md). ```text 1. Jonckheere-Terpstra trend test on v0.9a three-zone data (LOW EFFORT, HIGH SUBSTANCE). Tests the monotone-increasing pattern directly; J-T has more power than chi^2 for ordered alternatives. Non-circular by breaking test-statistic condition. 2. v0.7a' held-out predictor on the 250 analyzable rows (LOW-MEDIUM EFFORT). Re-register v0.7b on the restricted scope; leave-one-m_3-bin-out + asymmetric McNemar + positive- delta gate. 3. v0.7a relaxed-precision re-run at rtol = atol = 1e-10 (MEDIUM-HIGH EFFORT). Attempts to clear the 23 attrited rows and recover catalog-wide v0.7a claims. 4. m_3 = 0.4 sub-catalog targeted mechanism investigation (MEDIUM EFFORT, HIGH SUBSTANCE). The m_3 = 0.4 cluster has been the load-bearing positive throughout; a within-bin audit may reveal substrate-specific structure. 5. Joint (vf, Q_E) audit on the 250 analyzable rows (MEDIUM EFFORT). Cochran-Mantel-Haenszel-style stratified test of Q_vf perp S/U given Q_E. 6. Cross-substrate transfer test (HIGH EFFORT). Apply the v0.7a velocity-fraction mechanism to a different three-body sub- catalog (mesa analog, free-fall, choreography) to test substrate-specificity. 7. Action-angle / KAM-style decomposition (HIGH EFFORT, THEORETICAL). Per-row action-angle variables stratify by integrability proximity; KAM-stable orbits are predicted stable. 8. Full-catalog audit at rtol = atol = 1e-10 with v0.7a R2.A discipline (alternative to avenue 3 if precision relaxation is sufficient to clear attrition). ``` All eight avenues respect the locked methodology surface AND the anti-circular framing discipline. Reopening the program at any of these is invited; mechanism-laundering (silently re-testing v0.7a' under different labels) is not. Stop condition: If a lab initiate's reopening avenue runs and produces a clean verdict (pass, fail, or sparse-inconclusive) under the locked discipline, the program may UNPAUSE at that point and the chain extends. Until then, the isotrophy section is closed for active investigation. Frame: v0.4 ruled out the Z_2 shadow. v0.5 ruled out the 2-bit catalog branch shadow. v0.6 ruled out the (E, |L|) catalog-coordinate shadow beyond branch labeling. **v0.7 leaves catalog-coordinate space**: it asks whether the row's per-orbit monodromy operator has a geometric structure (specifically the DIRECTION of its largest-real-part Floquet eigenvector, projected onto a pre-registered geometric reference frame) that stratifies stability. Locked parent shape: ```text Body: supp-B piano-trio orbit as primary orbit-dynamics object (273 rows, M_i derivable from v0.4a receipt). Projection: row -> gamma_1(row), the eigenvector DIRECTION in phase space of the row's largest-real-part Floquet eigenvalue, projected onto a pre-registered geometric reference frame. Observable: per-row gamma_1 direction feature joined with S/U. Compute: monodromy + eigenstructure already in v0.4a; feature extraction is seconds per row. Discipline: inherits v0.5 (audit-then-predictor, asymmetric McNemar + delta falsifier) and v0.6 (alignment-tightness guard, sparse-cell fallback tree). ``` Candidate operational definitions (one to be locked by v0.7a): ```text D1. Largest-real-part eigenvalue's eigenvector. D2. Largest-magnitude eigenvalue's eigenvector. D3. Restricted to "well-defined" rows; companion sidecar for ill-defined. D4. Z_2-isotypic phase-space split of gamma_1 (even/odd fraction). D5. Phase-space angle decomposition (velocity vs spatial component ratio). ``` Candidate audit forms (one to be locked by v0.7a): ```text A. Direction-projection quartile audit. B. Velocity-fraction quartile audit. C. Z_2 isotypic-fraction quartile audit. D. Direction-angle binary audit (df = 1, critical 6.63). ``` The v0.7a form lock must address three explicit circularity risks (eigenvalue-choice, well-definedness, feature-extraction) and lock a deterministic tie-break for S-row eigenvalue degeneracy, a reference frame for the geometric projection, an alignment-tightness guard against the v0.5a branch label, and the sparse-cell fallback tree. v0.7a result: ```text receipt: results/isotrophy/k-facet-v07a-velocity-fraction-audit/manifest.json verdict: velocity_fraction_blocked_integration_attrition total rows: 273 (S=97 / U=176) analyzable: 250 integration_blocked: 23 (8.42%) attrition threshold: 14 (5% of 273) attrition fired: True sanity-gate failures (analyzable but symp>1e-4 or recip>1e-4): 11 total data-integrity issues: 34 / 273 = 12.5% vf_sd: 0.144 (would have passed constant-feature retirement at 0.01, but no chi-squared verdict is licensed under the attrition gate) runtime: 266.9 min (~4.5 h) ``` Blocked rows cluster at long-period high-m_3 (the same regime v0.4a's two-pass classifier was built to handle). At v0.7a's variational precision rtol = atol = 1e-12, the DOP853 step adapter cannot find feasible steps for the matrix variational equation along the most-challenging orbits. The audit is integration-attrited, NOT feature-falsified. v0.7a locked shape (pre-verdict): ```text operational definition: D5 velocity-fraction direction property. gamma_1 selection: largest-real-part Floquet eigenvalue's eigenvector (D1's selection rule used only to disambiguate; feature is independent of which eigenvalue is picked). tie-break cascade: within degenerate eigenspace, maximal projection onto velocity subspace; then smallest |Im(lambda)|; then smallest positive argument; then lex sign. reference frame: center-of-mass reduced frame with mass-weighted norm (body-3 mass variation does not fake a direction effect). feature: vf(row) = ||delta_v||^2 / (||delta_q||^2 + ||delta_v||^2) after CoM reduction and mass-weighted norm. audit form: B = quartile audit of vf vs S/U. binning: supp-B quartiles via numpy.quantile, linear interpolation; right-closed lower bins. contingency: 4 x 2 (Q_vf, S/U), df = 3. critical: 11.34 at p = 0.01 (chi-squared(3)). alignment guard: max over Q_vf of fraction-in-any-single- v0.5a-branch_label, threshold 0.8 (warning), 0.95 (severe). sparse-cell fallback: v0.6b inheritance verbatim (seed = 20260523, n_permutations = 10000). constant-feature retirement threshold: sd(vf) < 0.01. verdicts: velocity_fraction_passes_audit | velocity_fraction_passes_audit_alignment_warning | velocity_fraction_passes_audit_severe_alignment | velocity_fraction_fails_audit | velocity_fraction_inconclusive_sparse | velocity_fraction_retired_near_constant | velocity_fraction_blocked_sanity | velocity_fraction_blocked_circularity_audit. named sidecar (D1 + A): gamma_1 z-projection magnitude under CoM frame, quartile-binned. Report-only; a loud sidecar would require a fresh circularity audit, NOT a refinement of v0.7a. locked non-circularity sentence (paper-side): "v0.7a uses Floquet eigenvectors only as geometric directions; it does not use eigenvalue magnitude, spectral radius, unit-circle status, unstable-pair count, or any threshold that defines the published S/U label. The tested scalar is a phase-space composition ratio of the selected direction, not the growth rate of that direction." ``` Completed actions (during the v0.7a run): 1. ~~Implement `scripts/v07a_velocity_fraction_smoke.py` and run 7-row sentinel smoke.~~ DONE 2026-05-24. Initial smoke showed integrator was column-by-column slow; user signed off on vectorized `compute_monodromy` helper. Subsequent vectorized smoke + R1 amendment showed 7/7 sentinels pass at 1e-4 symplecticity gate. 2. ~~First full-catalog run crashed at row 76 (O_194 at m_3=0.5) with "Required step size is less than spacing between numbers" inside `compute_monodromy_vectorized`.~~ User signed off on R2.A per-row integration-failure fallback + R2.C append-per-row resume. 3. ~~Implement `scripts/v07a_velocity_fraction_audit.py` with R2.A try/catch + R2.C append-per-row + resume mode.~~ DONE 2026-05-24. 4. ~~Re-run full audit blind under amendments.~~ DONE 2026-05-24, ~4.5 hours runtime, 23 integration-blocked rows, attrition fired. Next action: Pre-mortem flagged action items if the run hadn't completed cleanly: 1. Implement `scripts/v07a_velocity_fraction_audit.py`: - Per-row variational integration of the 18-dimensional tangent system over one period (new compute; was not in v0.4a's emitted scalars). - Symplecticity sanity gate (M_i^T J M_i - J in tolerance 1e-6). - Reciprocal-pair sanity gate (Floquet eigenvalue pairing within 1e-4). - gamma_1 extraction under the locked selection rule and tie-break cascade. - Velocity-fraction computation under CoM reduction + mass-weighted norm. - Constant-feature retirement check (sd(vf) < 0.01). - Quartile binning + chi-squared + alignment-tightness guard + sparse-cell fallback. - D1 + A sidecar emitted under the same selection rule, different feature. - Receipt at `results/isotrophy/k-facet-v07a-velocity-fraction-audit/manifest.json`. 2. Run v0.7a blind; land verdict. 3. Conditional on `velocity_fraction_passes_audit` (clean): register v0.7b with default leave-one-m_3-bin-out partition. Conditional on `..._alignment_warning` or `..._severe_alignment`: register v0.7b with alignment-breaking partition (within-branch mirroring v0.6b). Conditional on `..._fails_audit`: close the velocity-fraction sub-question; the D1+A sidecar's chi-squared is informational only; v0.7c may register a fresh form lock with a different D + A combination after a circularity re-audit. Conditional on `..._inconclusive_sparse` or `..._retired_near_constant` or `..._blocked_*`: inspect and re-register as appropriate. Stop condition: If the v0.7a runner cannot maintain the locked non-circularity discipline at compute time (e.g., the tie-break cascade requires inspecting eigenvalue magnitude in some unforeseen way), abort with `velocity_fraction_blocked_circularity_audit`. No compute verdict is licensed. ## Onboarding / Polish (Run-Friendly) ### V0.3h K_facet Tooling Polish (~10-hour initiation) Sources (paper trail / table of contents): 1. [`../docs/isotrophy/kfacet/kfacet_v03h_writeup.md`](../docs/isotrophy/kfacet/kfacet_v03h_writeup.md) — Methodology + result hand-off. **Start here.** Sets the audit chain (sentinel → adaptive-floor reprocessor → bridge audit) and the load-bearing result: `20 of 21 m_3=1 strict G.2 rows are structural zeros; O_617 is the single quarantined row`. 2. [`../docs/isotrophy/kfacet/kfacet_v03h_o617_deep_dive.md`](../docs/isotrophy/kfacet/kfacet_v03h_o617_deep_dive.md) — The quarantined-row companion. Reads the six-probe deep dive plus the WHY-dive addendum that landed `bridge_approx_sign_isotypic` as the corrected disposition. 3. [`../docs/isotrophy/kfacet/kfacet-runner-spec.md`](../docs/isotrophy/kfacet/kfacet-runner-spec.md) — Full runner spec. Includes the adaptive-floor algorithm, the bridge audit outcome categories, and the disposition for O_617. Long; use the table-of-contents at the top to navigate. 4. [`sundog_v_isotrophy.md`](sundog_v_isotrophy.md) log lines 1305-1310 — Dated v0.3h log entries. Each line is one round of the audit chain ratcheting. 5. [`SUNDOG_V_THREEBODY.md`](SUNDOG_V_THREEBODY.md) bridge-audit section around line 933 — Parent-spine mirror of the same closure. 6. [`../scripts/isotrophy_workbench.py`](../scripts/isotrophy_workbench.py) — Workbench with three landed subcommands (`kfacet-sentinel`, `kfacet-reprocess-floor`, `kfacet-bridge-audit`) plus the pre-registered constants (`ADAPTIVE_FLOOR_*`, `BRIDGE_*`). 7. [`../scripts/o617_deep_dive.py`](../scripts/o617_deep_dive.py), [`../scripts/o617_why_dive.py`](../scripts/o617_why_dive.py), [`../scripts/catalog_near_t_separator.py`](../scripts/catalog_near_t_separator.py) — Three one-shot diagnostic scripts. These are the candidates for promotion in this polish item. Status: `deferred`, polish-ready. ~10-hour budget. No compute lock; all diagnostic scripts run in seconds against existing receipts under `results/isotrophy/k-facet-v03-*/`. Current state: - The v0.3h audit chain is **closed** for `m_3=1`. Three workbench subcommands + three one-shot Python scripts produce the load-bearing result with deterministic, pre-registered classifications. - Diagnostic scripts (`o617_deep_dive.py`, `o617_why_dive.py`, `catalog_near_t_separator.py`) are functional one-shots but hardcoded to `O_617` (deep-dive and why-dive) or to the full 21-row catalog (separator). They are not discoverable via `npm run`. - The catalog separator has a known projector double-counting artifact (T + S + E exceeds kernel dim by 1 for `O_617`) that should be fixed by switching from character projectors to simultaneous (sigma_3, F_beta) joint eigendecomposition. - The interpretive lesson from the (II) round (always compute signed ``, never just parity magnitudes) is currently a comment in the WHY-dive script; it should be hoisted into a workbench helper used by all three diagnostics. Verify-the-picture step (1 hour budget): 1. Run `npm run isotrophy:kfacet:sentinel:gamma` on `O_62` (control) and inspect `gate_receipt.json`. Expect all gates PASS, `c_i = d_i = 0`, `T(2)+S(5)+E(0)`. 2. Run `npm run isotrophy:kfacet:reprocess-floor`. Manifest summary should report `21/21 resolved or recorded` with no suspicious rows. 3. Run `npm run isotrophy:kfacet:bridge-audit`. Manifest should report `20 no_bridge_present + 1 defective_E_block_confirmed (O_617)`. 4. Run `python scripts/o617_deep_dive.py`. Six probes; expect `near_T_edge=[617]` and `e_rotation=0` per the existing receipt. 5. Run `python scripts/o617_why_dive.py`. Expect outcome `bridge_approx_sign_isotypic` (the corrected label after the isotypic-projector check). 6. Run `python scripts/catalog_near_t_separator.py`. Expect `edge_E_other=[617]` only; all 20 other rows pure `clean_T(2) + clean_S(5 or 6)`. If any of (1)-(6) drift from the recorded receipts, stop and consult [`../docs/isotrophy/kfacet/kfacet_v03h_writeup.md`](../docs/isotrophy/kfacet/kfacet_v03h_writeup.md) before continuing. Polish items: 1. **Generalize `o617_deep_dive.py` to a workbench subcommand** (~2 hours). - Add `kfacet-row-anatomy --row INDEX [--source A --m3 1.0]`. - Make all six probes accept an arbitrary row, not just `O_617`. - Receipt path: `results/isotrophy/k-facet-v03-row-anatomy/O{idx}/`. - npm: `isotrophy:kfacet:row-anatomy -- --row 617`. - Smoke: run on `O_62` (control) and `O_617` (quarantined); both should produce structured receipts without errors. 2. **Generalize `o617_why_dive.py` to a workbench subcommand** (~2 hours). - Add `kfacet-why-dive --row INDEX`. - Keep the corrected outcome hierarchy: `bridge_in_wrong_frame`, `bridge_frame_partial`, `bridge_approx_sign_isotypic`, `bridge_approx_trivial_isotypic`, `bridge_is_quasi_kernel`, `bridge_eigenvector_off_unit_circle`, `bridge_genuinely_non_D3`. - Include direct signed `` and `` in the C1/C3 receipts (the interpretive-error fix from the (II) round). - npm: `isotrophy:kfacet:why-dive -- --row 617`. 3. **Generalize the catalog separator and fix the projector double-counting** (~3 hours). - Add `kfacet-near-T-separator [--rows IDX,IDX,...]`. - Replace the character-projector decomposition with a simultaneous-eigenvector decomposition of `(sigma_3, F_beta)` on the kernel (joint diagonalization). T = (lambda_s3=+1, lambda_fb=+1) eigenspace, S = (+1, -1), E = (omega, omega^2) pair. This fixes the `T + S + E = 9` vs `kernel_dim = 8` artifact for `O_617`. - Keep the existing classification thresholds (alignment 0.99, closure 1e-3) intact. - npm: `isotrophy:kfacet:near-T-separator`. 4. **Add a shared workbench helper `compute_d3_alignments`** (~30 minutes). Encapsulate the "always signed inner product" lesson: given a direction `v`, return `` and `` (signed). Use this helper from the three new subcommands; this is the codified version of the interpretive lesson the (II) round surfaced. 5. **Smoke tests** (~1 hour). One small `scripts/test_kfacet_diagnostics.py` or pytest module that runs each new subcommand on `O_62` and `O_617`, asserts on a few headline fields (outcome name, isotypic dims, bridge alignment sign), and emits a pass/fail summary. Wire it under `npm run isotrophy:kfacet:diagnostics:test`. 6. **Doc updates** (~1 hour). - In [`../docs/isotrophy/kfacet/kfacet_v03h_o617_deep_dive.md`](../docs/isotrophy/kfacet/kfacet_v03h_o617_deep_dive.md) and [`../docs/isotrophy/kfacet/kfacet-runner-spec.md`](../docs/isotrophy/kfacet/kfacet-runner-spec.md), replace the `python scripts/o617_*.py` references with the new `npm run isotrophy:kfacet:*` commands and `--row` parameterization. - Add a one-line note in [`../docs/isotrophy/kfacet/kfacet_v03h_writeup.md`](../docs/isotrophy/kfacet/kfacet_v03h_writeup.md) pointing at the new row-anatomy / why-dive subcommands. 7. **Buffer / unblocking room** (~30 minutes). Reserved for friction on argparse, receipt-schema drift, or the joint eigendecomposition subtleties. Acceptance criteria: - All three new npm scripts run cleanly on `O_62` and `O_617` and emit versioned receipts under `results/isotrophy/k-facet-v03-*/`. - The `kfacet-near-T-separator` no longer over-counts: `T + S + E == kernel_dim` exactly for every row, including `O_617`. - The `kfacet-why-dive` receipt records signed `` and `` explicitly. The outcome label is `bridge_approx_sign_isotypic` on `O_617`. - `npm run isotrophy:kfacet:diagnostics:test` passes locally; receipts archived for the audit trail. - Documentation references the new npm-script entry points; the one-shot scripts can be marked deprecated but remain runnable. Known issues / interpretive guardrails: - **Signed-vs-magnitude trap**: the original WHY-dive inferred `F_beta v approx +v` from parity-magnitude inspection of `F_beta v_bridge`. The actual signed alignment is `-0.9999970`. Always use signed inner products for representation-theoretic classification; magnitudes alone discard the irrep label. - **Catalog-admission orientation**: `O_617` is admitted to the strict-21 catalog via the *opposite* sigma_3 orientation (residual `1.01e-8`, catalog-normal). It is not a weak-admission row. The defective E(1) bridge is a structural property of the orbit at this kernel boundary, not a tolerance artifact. - **The 20/21 firewall**: the polish work does **not** touch the load-bearing v0.3h evidence claim. The new subcommands are reproductions of existing one-shot scripts with `--row` parameters and one methodology fix (joint eigendecomposition instead of character projectors). The 20-row structural-zero result and the O_617 quarantine should be byte-identical across the rerun. Out of scope for this polish round: - Cross-`m_3` or supplementary-B verification. - Catalog-admission tightening (already settled as unnecessary; `O_617` is not a weak-admission row). - Paper-side write of the audit chain (see [`../docs/isotrophy/kfacet/kfacet_v03h_writeup.md`](../docs/isotrophy/kfacet/kfacet_v03h_writeup.md) for the existing methodology hand-off; a paper draft would be a separate deferred item). - Advancing to v0.4 / induced-rep `d_i` derivation. Exit deliverable: a single PR (or its equivalent) that adds three new workbench subcommands, three new npm scripts, one shared `compute_d3_alignments` helper, one smoke-test module, and the doc updates listed above. Reviewer should be able to verify the exit deliverable by running `npm run isotrophy:kfacet:diagnostics:test` and inspecting the new manifest under `results/isotrophy/k-facet-v03-row-anatomy/`. ## Quick Re-Inventory Commands Use these when this TODO starts to feel stale: ```powershell rg -n "BLOCK|HELD|HOLD|blocked|open|Next actions|Next Gate|operator|compute-unconstrained|old_basin_pref|Bayesian floor" . -g "*.md" -g "*.html" -g "*.json" git status --short --untracked-files=all npm run chat:index ```