# Prop-01 E1b §5 gate re-score

| Field | Value |
| --- | --- |
| **Protocol** | `LAX-PROP-2026-09-04-01a-E1b` |
| **Parent scorecard** | `writing/prop01-e1-section5-scorecard.md` (E1 full — retained) |
| **E1 full md5 (untouched)** | json `777b2dca461f2d3f1aac379a4cd2e794` · md `57ab776ef6e8562cdcb870c1389bfd2f` |
| **Scored** | 2026-09-06 (America/New_York) from on-disk E1b artifacts only |
| **prop01_support_claim** | **false** — not flipped |

**Sources:** `sims/sr-ladder/results/prop01_live_e1b_damped_ladder.*`, `prop01_live_e1b_dimviol.*`, `prop01_live_e1b_a3_smoke.*` + `prop01_live_e1b_a3.*` (a3.json md5 `36084a5e71347d069ccb117cf0707009`); parent E1 full scorecard for §5.1.

Frozen §5 thresholds unchanged (01a): recovery ≥0.8 at ≤0.5× N for ≥5/7; mid-noise A4 ≥ A0 − 0.05; clean-fail ≥0.9 at R\*=0.25; non-triviality via conservation-only.

---

## Gate table

| Gate | Verdict | Measured evidence |
| --- | --- | --- |
| **§5.1 Sample efficiency** | **Fail** | Prior E1 full scorecard: **0/7** clear ≤0.5× passes (need ≥5/7). E1 full artifacts left untouched; no new lit-backend full-grid efficiency re-score that clears the gate. |
| **§5.2 Noise robustness** | **Fail (KILLED)** | Damped-only budget ladder: A4 mid-noise recovery **0** at every rung including **r32** (epochs=32, max_n_evaluations=12000) vs A0 0.6–0.8; drop >0.2 on all cells. Repair lane stopped. Do **not** reopen big damped runs. Artifacts: `prop01_live_e1b_damped_ladder.*`. |
| **§5.3 Clean fail** | **Pass** | A4 `dim_violating` clean-fail **0.975** (117/120 not-accepted); aggregate mean of six A4 rates **0.975** (min 0.95). Target ≥0.9; R\*=0.25 fixed. Artifacts: `prop01_live_e1b_dimviol.*`. |
| **§5.4 Non-triviality** | **Pass (surrogate close)** | STLS-surrogate A3 pack: trivial accept **0.000**; dyn_gate_fail **1.000** (n=2800). A3 recovery by law (pooled n=200 cells/law): linear_decay 0.945, harmonic 0.560, damped 0.985, lotka 0.995, diffusion 0.000, burgers 1.000, gravity 0.500. Closes prior Incomplete: conservation-only does **not** succeed via trivial identities. **Surrogate never flips support.** Artifacts: `prop01_live_e1b_a3_smoke.*` + `prop01_live_e1b_a3.*`. |

---

## Overall

| Question | Answer |
| --- | --- |
| All four gates Pass? | **No** (§5.1 Fail; §5.2 Fail/Killed) |
| Flip `prop01_support_claim`? | **No — stays false** |

**Verdict:** E1b closed two gates (§5.3 Pass; §5.4 Pass via STLS-surrogate) and **irreducibly failed** §5.2 on `damped_oscillator` under matched wall-cap (A4 recovery identically 0 through r32). §5.1 remains Fail on the untouched E1 full lit-backend grid. Completing E1b patch lanes does **not** authorize support. Prop-02 GNN remains held.

---

## Recommended next science step

**Do not** open Prop-02 GNN primary. Prefer one of:

1. **Prop-01 honesty close-out (preferred first):** CoS escalate to Michael — demote or amend A4/PhySO on dissipative / mid-noise recoverable laws after irreducible §5.2 kill; keep §5 thresholds frozen; optional Lit Scout cite-only inventory for alternate lit backends (no threshold edits). Publish measured honesty only.
2. **Parallel compute:** start Prop-03 B0 Cobaya smoke (`proposals/prop-03-b0-cobaya-smoke-checklist.md`) — claim language separate; no tension-solved language.
3. **Fingerprint-only secondary** (Prop-02 reporting track) only if CoS wants light parallel work — still cannot declare `prop02_success_claimed`.

Stop further damped PhySO budget ladders.