F6 — cluster or scatter? Three more independent no-input boots to explain the out-of-sample residue #25

New Issue

sylph-decoder · 2026-09-11T21:26:20Z

sylph-decoder commented

2026-09-11 21:26:20 +00:00

Follow-on from #9's shaping pass:
docs/re/f6-residue-shaping.md.

f6-out-of-sample-RESULT.md found the sweep→plate lead and the
ptcopyright/parent ramp ratio both miss the same direction on a fresh boot
(0.75×, 0.71× of predicted). The shaping pass connects this to
f6-unit10, which
already established the sweep and plate families are gated by separately
triggered parents — so there is no structural guarantee their relative
phase is a boot-to-boot constant, only an empirical question nobody has
enough data to answer. Two live hypotheses, not yet distinguished:

the cross-group phase genuinely varies boot to boot (a real game fact worth
reporting as "measured, with variance" rather than a constant), or
an artifact in how check_labels.py derives frame-based ratios from a
differently-shaped capture.

Instrument: ⟨capture⟩ × 3 more independent no-input boots (total n = 6
with f6/f6b/f6c), with check_labels.py's repaired per-run diagnostics
(lag, sample count, per-cycle variance) saved alongside each pass/fail line
rather than discarded.

Pass: the three new runs either cluster (supports a real, possibly
discrete, boot-order dependency) or scatter without structure while the
amplitude error tracks pulse_period's own noise (supports an instrument
explanation) — either is an answer.

Does NOT reinstate the withdrawn 0.138–0.141 sweep→plate lead regardless
of outcome; at best it explains why no single number applies.

New exploratory dynamic RE, not a continuation of #9's own (static) scope —
filed for approval rather than started.

Follow-on from #9's shaping pass: [`docs/re/f6-residue-shaping.md`](../../docs/re/f6-residue-shaping.md). `f6-out-of-sample-RESULT.md` found the sweep→plate lead and the `ptcopyright`/parent ramp ratio both miss the same direction on a fresh boot (0.75×, 0.71× of predicted). The shaping pass connects this to [`f6-unit10`](../../docs/re/f6-unit10-parent-alpha-gates-the-sweep.md), which already established the sweep and plate families are gated by **separately triggered** parents — so there is no structural guarantee their relative phase is a boot-to-boot constant, only an empirical question nobody has enough data to answer. Two live hypotheses, not yet distinguished: * the cross-group phase genuinely varies boot to boot (a real game fact worth reporting as "measured, with variance" rather than a constant), or * an artifact in how `check_labels.py` derives frame-based ratios from a differently-shaped capture. **Instrument:** ⟨capture⟩ × 3 more independent no-input boots (total n = 6 with `f6`/`f6b`/`f6c`), with `check_labels.py`'s repaired per-run diagnostics (lag, sample count, per-cycle variance) saved alongside each pass/fail line rather than discarded. **Pass:** the three new runs either cluster (supports a real, possibly discrete, boot-order dependency) or scatter without structure while the amplitude error tracks `pulse_period`'s own noise (supports an instrument explanation) — either is an answer. **Does NOT reinstate** the withdrawn 0.138–0.141 sweep→plate lead regardless of outcome; at best it explains why no single number applies. New exploratory dynamic RE, not a continuation of #9's own (static) scope — filed for approval rather than started.

sylph-decoder referenced this issue

2026-09-11 21:26:23 +00:00

re: land the F5/F6 title-clock corpus (docs/re, reference data, sylpheed-formats) #23

sylph-decoder added the agent/decoder kind/item state/proposed labels 2026-09-11 21:26:27 +00:00

sylph-decoder referenced this issue

2026-09-11 21:26:35 +00:00

f6-out-of-sample-RESULT: three of six predictions failed — shape the residue #9

fabi referenced this issue from a commit

2026-09-12 13:24:31 +00:00

re: fix two stale rows -- NEW GAME's destination, and my own repeat claim

fabi referenced this issue from a commit

2026-09-19 20:29:49 +00:00