SSM-Clock—Acceptance Dataset & Core Results (3.1–3.2)

The v1.1 inverse (stacked snapshots + multistart + Brent refine) meets or exceeds the release gates on the 4-cycle human/astro set, delivering minutes-level error with PASS >= 95% even at 6 deg noise. All results are deterministic, ASCII-only, and reproducible with the commands below.

3.1 Test dataset and knobs (frozen for v1.1)

  • Period set: 1, 7, 29.5306, 365.2422
  • Noise levels (deg): 2, 6
  • Seeds: 200 per noise point
  • Grid: 2 min
  • Stack: S = 5 snapshots, delta = 0.5 day (12 h)
  • Multistart: K = 7 (top-7 coarse minima refined)
  • Refine: Brent (--refine_steps 80)
  • Gating: alpha_kz = 0.0 (gentle/off for acceptance)
  • Horizon: T_search = max(periods) (non-integer periods present)
  • Pass policy: error <= 28.8 min with pass_rate >= 95% over all seeds

Repro command (exact knobs)

python ssm_clock_longbench_v2.py ^
  --period_sets "1,7,29.5306,365.2422" ^
  --seeds 200 ^
  --noise_list "2,6" ^
  --grid_step_min 2.0 ^
  --alpha_kz 0.0 ^
  --stack 5 --stack_dt_days 0.5 ^
  --multistart_k 7 --refine brent --refine_steps 80 --bracket_mult 2

3.2 Core 4-cycle set results (v1.1)

  • All gates PASS.
  • Interpretation: 2 deg noise produces a tight error band (mean ~2.7 min). 6 deg noise remains robust (mean ~8.2 min; P95 ~ 19.6 min).
Period Set                    Noise   Pass(X/Y)  Pass%   Mean(min)  P95(min)  Gate(>=95%)
1,7,29.5306,365.2422         2       200/200    100.0   2.7        6.5       PASS
1,7,29.5306,365.2422         6       199/200     99.5   8.2        19.6      PASS

One-line claim. Minutes-level “clock time” recognition holds across days to thousands of years (symbolic horizon repeats), without wall clock or ephemeris.

Navigation

Back: SSM-Clock—CSV Modes & Packaging (2.10–2.11)
Next: SSM-Clock—Sensitivity & Error Distributions (3.3–3.4)

Explore further:

https://github.com/OMPSHUNYAYA/Symbolic-Mathematical-Clock