Experiments · E248c

Does a separate size factor for separation fix the reading?

Partly. It reads a fresh 300 K separation at 304 K, but a 95 K separation still passed: its peak sat on the 60 K floor of the temperature grid.

In the log: rung 1 corrects separation with its own factor (pre-registered 2026-09-27 23:09)

mixedDate 2026-09-27 23:09, as written in the logrung 1 · ordering0 predictions · 1 result paragraphEXPERIMENTS.md lines 17456–17467, lines 17469–17477

Pre-registration

The pre-registration, as written

E248c — rung 1 corrects separation with its own factor (pre-registered 2026-09-27 23:09)

Rule. When the cold state clusters (nearest-neighbour α₁ > +0.1 in the coldest sampled state of every seed), rung 1 divides its peak by r = 0.67 ± 0.03 (E248/E248b) instead of 0.95 ± 0.03. Ordering readings are unchanged. The caller computes α₁ from the legs' own stored occupations.

Out-of-sample test, fresh seeds 8 and 9 (the factor was fitted on seeds 6 and 7):

  • clustering Ising at exact 300 K: the read T is within 10 % of 300 K;
  • clustering Ising at exact 95 K: p_rung1 ≤ 0.1 (the case that would pass without the rule);
  • the seven E247 known answers (ordering) re-read unchanged.

Prediction: all three hold.

Results

EXPERIMENTS.md · line 17469

E248c RESULT (2026-09-27 23:21). Fresh seeds 8 and 9, read with the separation factor:

  • 300 K: holds. Peak 204 K, corrected to 304 K, within 1.5 % of exact. The factor holds out of sample.
  • 95 K: FAILS, a false pass. p_rung1 = 1.0, verdict "no significant peak, and the window is quiet (1.2 meV/atom)". The sampler did not freeze (swap acceptance 7 % at 60 K). The separation is still under way at the grid's 60 K floor: E falls from −1.15 meV/atom at 93 K to −2.56 at 60 K. At 95 K the peak sits near 64 K (×0.67), too close to the floor to be read as a peak. The fallback "quiet window" test uses an absolute 3 meV/atom, larger than this whole transition. A decomposing alloy that separates inside the window passes rung 1. The 0.67 factor is right, but it cannot act on a peak the grid does not reach.

The full record

This entry is written in 2 separate places in the log, shown here in log order.

EXPERIMENTS.md · lines 17456–17467

E248c — rung 1 corrects separation with its own factor (pre-registered 2026-09-27 23:09)

Rule. When the cold state clusters (nearest-neighbour α₁ > +0.1 in the coldest sampled state of every seed), rung 1 divides its peak by r = 0.67 ± 0.03 (E248/E248b) instead of 0.95 ± 0.03. Ordering readings are unchanged. The caller computes α₁ from the legs' own stored occupations.

Out-of-sample test, fresh seeds 8 and 9 (the factor was fitted on seeds 6 and 7):

  • clustering Ising at exact 300 K: the read T is within 10 % of 300 K;
  • clustering Ising at exact 95 K: p_rung1 ≤ 0.1 (the case that would pass without the rule);
  • the seven E247 known answers (ordering) re-read unchanged.

Prediction: all three hold.

EXPERIMENTS.md · lines 17469–17477

E248c RESULT (2026-09-27 23:21). Fresh seeds 8 and 9, read with the separation factor:

  • 300 K: holds. Peak 204 K, corrected to 304 K, within 1.5 % of exact. The factor holds out of sample.
  • 95 K: FAILS, a false pass. p_rung1 = 1.0, verdict "no significant peak, and the window is quiet (1.2 meV/atom)". The sampler did not freeze (swap acceptance 7 % at 60 K). The separation is still under way at the grid's 60 K floor: E falls from −1.15 meV/atom at 93 K to −2.56 at 60 K. At 95 K the peak sits near 64 K (×0.67), too close to the floor to be read as a peak. The fallback "quiet window" test uses an absolute 3 meV/atom, larger than this whole transition. A decomposing alloy that separates inside the window passes rung 1. The 0.67 factor is right, but it cannot act on a peak the grid does not reach.

Related entries

Built with PRISMWebsite and visualizations made using Claude