# Three-Body Phase 13 - Longer-Horizon Result Note This document records Phase 13 results for [`PHASE13_SPEC.md`](PHASE13_SPEC.md). Phase 13 asks whether the Phase 11 positive pocket survives a longer rollout, or whether guarded accelerometer TRACK mostly delays failures inside the original 8-second horizon. Status: smoke run and full lock complete. Cross-doc verification (2026-05-15): result trail confirmed consistent across `SUNDOG_V_THREEBODY.md` (claim ladder, rollup, receipt), `threebody-writeup.md`, and `CROSS_SUBSTRATE_NOTES.md`. Metrics (3,456 / 88·324 / 81·108 / 100·108 / 0.229 vs 0.218) agree on every surface; no Phase 12 bleed or transposition. Reconciled the Phase 13 claim-ladder article in `SUNDOG_V_THREEBODY.md` and added the pre-registered-branch verdict to `threebody-writeup.md`; low-velocity boundary negatives carried in `CROSS_SUBSTRATE_NOTES.md` as designed (no numeric trail expected there). ## 1. Smoke Run Command: ```bash npm run threebody:phase13:smoke ``` Output: - `results/threebody/phase13-long-horizon-smoke/` - 32 trials at `duration=16` - 2 candidate envelope rows out of 12 - Terminal outcomes: 13 bounded, 15 escape, 4 close approach The smoke covered mass ratio `1`, timestep `0.01`, radii `1.025` and `1.075`, velocities `0.9` and `1.1`, four modes, and two seeds. ## 2. Smoke Read The larger-radius, high-velocity cell (`radiusScale=1.075`, `velocityScale=1.1`) remains promising at the longer horizon. Guarded accelerometer TRACK bounded both seeds while passive bounded none, with mean time delta `+9.85` seconds against passive and mean delta-v `1.948`. The smaller-radius, high-velocity cell (`radiusScale=1.025`, `velocityScale=1.1`) is neutral: passive already bounded both seeds for the full 16-second horizon, so guarded TRACK does not earn a survival delta there. Both low-velocity boundary cells (`velocityScale=0.9`) are risky. In those cells, passive often survives longer than the controllers, and the dominant failure mechanism is `controller_destabilized_or_shortened_passive`. ## 3. Smoke Claim Impact The smoke alone did not justify upgrading the public claim. It supported running the full Phase 13 lock because the high-velocity larger-radius pocket still existed at 16 seconds, but it also confirmed that horizon extension makes the failure boundary harder, not softer. Current earned wording remains: > In the tested planar restricted setup, a guarded accelerometer-proxy TRACK > controller improves survival over passive and naive local baselines in a > robust high-velocity near-escape operating pocket. The full lock below supplies the ratchet test: the pocket must remain positive across mass ratio, timestep, and nearby high-velocity cells at the longer horizon, without a large delta-v-per-second or late-failure penalty. ## 4. Full Lock Readback Command: ```bash npm run threebody:phase13 ``` Slate: 3 mass ratios x 3 timesteps x 3 radii x 4 velocity scales x 4 modes x 8 seeds at `duration=16`, for 3,456 trials. Output: - `results/threebody/phase13-long-horizon-lock/` - 3,456 trials - Terminal outcomes from the runner: 1,154 bounded, 2,030 escape, 272 close approach - `trial-outcomes.csv` contains 2,592 paired non-passive rows: 864 each for naive, guarded TRACK, and oracle - Candidate envelope rows: 88 / 324 - Candidate rows by mode: guarded TRACK 77, oracle 11, naive 0 - Best-cell class balance: 81 promising, 17 mixed, 5 risky, 5 negative - Best-cell controller: guarded TRACK in 100 / 108 cells, oracle in 8 / 108 ## 5. Full Lock Read Phase 13 passes the long-horizon lock for the scoped high-velocity pocket. The result is not merely an 8-second delay: guarded accelerometer TRACK still dominates the favorable band at `duration=16`. Velocity-class read: - `velocityScale=1.05`: 27 / 27 best cells are promising. - `velocityScale=1.15`: 27 / 27 best cells are promising. - `velocityScale=1.1`: 19 / 27 best cells are promising, 8 / 27 are mixed, and none are risky or negative. - `velocityScale=0.95`: 8 / 27 promising, 9 / 27 mixed, 5 / 27 risky, and 5 / 27 negative. This remains the long-horizon boundary. Mode read: - Naive local acceleration remains a failed baseline: 0 candidate rows and 0 bounded paired trials out of 864, while its matched passive rows bounded 144 times. - Guarded TRACK bounded 749 / 864 paired trials; its matched passive rows bounded 144 / 864. - Oracle bounded 261 / 864 paired trials; useful as a privileged heuristic reference, but still not an optimal controller. Cost-rate read: - Guarded TRACK candidate rows have mean delta-v `3.665` over 16 seconds, or about `0.229` delta-v per simulated second. - Phase 11 guarded TRACK candidate rows had mean delta-v `1.741` over 8 seconds, or about `0.218` delta-v per simulated second. - The total control cost roughly doubles with horizon, but the cost rate does not materially explode. Late-failure read: - Guarded TRACK has 29 / 864 paired trials that survive past 8 seconds but fail before 16 seconds. - Of those late failures, 17 are `controller_destabilized_or_shortened_passive`, 6 are `control_effort_or_saturation`, and 6 have no classified mechanism. - This is not a large late-failure cliff; most guarded TRACK failures still occur before the original horizon or in the known low-velocity boundary. Boundary read: - All 5 negative best cells occur at `velocityScale=0.95` and mass ratio `1`. - The `velocityScale=0.95` band also contains all risky best cells. - The long-horizon result therefore strengthens the positive high-velocity pocket while preserving the low-velocity/equal-mass warning boundary. ## 6. Claim Impact Pre-registered branch: **pass, with boundary sharpening**. Updated earned wording: > In the tested planar restricted setup, a guarded accelerometer-proxy TRACK > controller improves survival over passive and naive local baselines across a > mapped high-velocity near-escape pocket through a 16-second tested horizon. > The result is not global: the low-velocity boundary, especially equal-mass > cells near `velocityScale=0.95`, still exposes controller harms. Do not upgrade this to indefinite stability, orbit-family maintenance, or general three-body control. Phase 13 only supports a longer tested horizon in a restricted planar operating envelope.