Files
MeltMind/artifacts/s2-g1-copy-calibration-2026-11-14/threshold-ruling-source.txt
T

31 lines
1.9 KiB
Plaintext
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
(2 total) ===
--- Reply 1 of 2 ---
From: ClaudeWorker (local) <botuser-T0C8JMA61UY-B0C7YUHNDFC@slack-bots.com> (U0C7SSQLBJ7)
Time: 2026-11-14 21:19:16 CST
Message TS: 1794662356.246479
Accepted; working on: “MeltAstra 正在实际启动下一阶段校准,已确认无既有校准产物/进程,固定 det/2000面积PPS/全三角双向P95 协议不改。常规…”
--- Reply 2 of 2 ---
From: ClaudeWorker (local) <botuser-T0C8JMA61UY-B0C7YUHNDFC@slack-bots.com> (U0C7SSQLBJ7)
Time: 2026-11-14 21:19:40 CST
Message TS: 1794662380.787319
*Claude: confirmed, with one correction to the physical route*
• *Two separate thresholds, as you propose:* τ_phys and τ_sf. Each = 3 × the largest positive-pair residual on its own route, where a pair's residual is min(det+, det−) P95. A pair links if either route, in either branch, is at or below its threshold.
• *Correction: express the physical route without units as well.* Part volumes span about 5 orders of magnitude, so a single τ in raw units would mean entirely different tolerances for small and large parts. So:
• *physical residual = P95 in raw units ÷ the mean of the two parts' R_g*;
• *the physical route also requires an R_g ratio ≤ 1.05.* That keeps it a same-size route.
The scale-free route stays as it is (each part normalised by its own R_g).
• *Positives per route,* as you set them out:
• physical: rigid copies, mirror copies, and the 2 natural pairs;
• scale-free: all three copy types (rigid, mirror, 2× scale) plus the 2 natural pairs.
• *Negative control:* the 2× scaled copies must fail the physical route because of the R_g-ratio gate. Report that they do. If any passes, stop.
• *Also report:*
• the full residual distribution of positives per route and branch;
• for the mirror copies, which branch fits best. det− should win; if it doesn't, flag it, as a check on the implementation.
Everything else proceeds as you described under the reproducible contract.