From 8927d2f1ae8df9a77f2ee8571df68fc8ed375d79 Mon Sep 17 00:00:00 2001 From: sylph-decoder Date: Fri, 4 Sep 2026 13:17:26 +0000 Subject: [PATCH] re: pre-register F6's numbers for an out-of-sample boot Every F5/F6 figure comes from captures taken for other purposes and re-analysed after the reader and three labels were corrected. None has been tested out of sample. Six predictions registered before the capture, five of them checkable by check_labels.py, which now takes --cap NAME=path. Prediction 6 is the one already got wrong once (0.057 against ptcopyright instead of 0.14 against the real plate), and is stated so a landing near 0.057 would show the correction itself was wrong. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_01Jc4pciRArGHfxGGhEbwp5t --- docs/re/f6-out-of-sample-prereg.md | 30 ++++++++++++++++++++++++++++++ tools/re-capture/check_labels.py | 8 ++++++++ 2 files changed, 38 insertions(+) create mode 100644 docs/re/f6-out-of-sample-prereg.md diff --git a/docs/re/f6-out-of-sample-prereg.md b/docs/re/f6-out-of-sample-prereg.md new file mode 100644 index 00000000..226bcfaa --- /dev/null +++ b/docs/re/f6-out-of-sample-prereg.md @@ -0,0 +1,30 @@ +# Pre-registration — F6's numbers on a boot that did not produce them + +Committed **before** the capture is taken. Every F5/F6 figure I hold comes from +`f6`, `f6b`, `f5` or `f5-early` — captures taken for other purposes, several +re-analysed after I fixed the draw reader and corrected three element labels. +**None has been tested out of sample.** A number re-derived from the same log +that suggested it is not independent evidence. + +One fresh no-input boot. Five predictions, all phase-free, all failing loudly if +wrong: + +| # | quantity | predicted | +|---|---|---| +| 1 | `pteff03a` period / `pteff03` period | **1.200** ±5 % | +| 2 | strip length ratio (`sy` 800/600) | **1.333** ±6 % | +| 3 | pulse period / sweep loop | **0.100** ±5 % | +| 4 | pulse amplitude vs `ptbtn00f.rat`'s declared 8-key curve | **≤3 alpha levels** mean error | +| 5 | `ptcopyright` ramp / parent ramp | **0.733** ±8 % | + +And one that is not in the checker, stated here so it cannot be quietly dropped: + +| 6 | sweep enters **before** `ptbtn00f`, by | **0.138–0.141** of a sweep loop | + +⚠️ Prediction 6 is the one I got wrong once already — I measured it against +`ptcopyright` and published 0.057. If the fresh run lands near 0.057 rather than +0.14, the correction was wrong and the original figure was right. + +**A pass is the boring outcome.** The value is that five of the six are checks a +wrong reader or a wrong label would fail, and the reader and two of the labels +have been wrong inside the last day. diff --git a/tools/re-capture/check_labels.py b/tools/re-capture/check_labels.py index 6c55cbe3..22844771 100755 --- a/tools/re-capture/check_labels.py +++ b/tools/re-capture/check_labels.py @@ -38,6 +38,14 @@ def declared_alpha(t): CAPS = {'f6': '/sylph-home/re/f6/xenia_re_ui_draws_01.log', 'f6b': '/sylph-home/re/f6b/xenia_re_ui_draws_01.log'} +# `--cap NAME=/path/to/log` replaces the set, for testing a fresh capture +# out-of-sample against predictions registered before it was taken. +_ov = [a for a in sys.argv if a.startswith('--cap')] +if _ov: + i = sys.argv.index(_ov[0]) + spec = sys.argv[i+1] if _ov[0] == '--cap' else _ov[0].split('=',1)[1] + k, v = spec.split('=', 1) + CAPS = {k: v} SWEEP_PAGE = '8154' PTCOPY_X = (-0.54, 0.54, 0.54, -0.54)