From 37f002930e11b7566a4397ecf1f385c66ee00b6c Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Thu, 6 Aug 2026 16:31:19 -0500 Subject: results: test SDIL on measured physical surrogate --- TWO_STATE_BIAS_PROGRAM.md | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'TWO_STATE_BIAS_PROGRAM.md') diff --git a/TWO_STATE_BIAS_PROGRAM.md b/TWO_STATE_BIAS_PROGRAM.md index 1020b11..c7a238f 100644 --- a/TWO_STATE_BIAS_PROGRAM.md +++ b/TWO_STATE_BIAS_PROGRAM.md @@ -297,6 +297,14 @@ Stop this paper direction if: `30/50/80/100 mV` before the observed gate plateaus; see `results/physical_bias/p0_state_dependence.json`. This establishes a measured locally predictable component suitable for P1, not an SDIL learning result. +- Physical P1 measured-surrogate development: mixed. With the same upfront + neutral observations, frozen affine SDIL beats frozen constant calibration + at all seven switching periods and removes essentially the complete modeled + raw-to-oracle gap. Online constant recalibration also reaches the oracle but + uses `240--6000` further neutral observations. The ideal leading-order + overclamping analogue reaches zero error without neutral observations and + beats SDIL everywhere. Therefore the mechanism test passes, but the physical + strong-baseline gate does not; see `results/physical_bias/p1_surrogate.json`. - Dual Prop same-path confirmation: active/supporting, not a passed result. - EP/CpL adapters: not implemented under this bias model. - Current score for this new paper framing: 5/10 until P1 passes. -- cgit v1.2.3