From 0b1d53508f497ab1cb2b9f2f079e5ecbb957eea7 Mon Sep 17 00:00:00 2001 From: Adam Wright <668955+adamjohnwright@users.noreply.github.com> Date: Fri, 9 Oct 2026 09:12:58 -0400 Subject: [PATCH] RESULTS: canonical on generator main (F7 set nodes + cap pools): held-out in release 87.93%, all 84.95% (+1.61pp vs MPB), experimental 69.96% (-5.8pp) Co-Authored-By: Claude Opus 5.5 (1M context) --- docs/RESULTS.md | 33 ++++++++++++++++++++++++++++++++- 1 file changed, 32 insertions(+), 1 deletion(-) diff --git a/docs/RESULTS.md b/docs/RESULTS.md index f6c3101..8bb7722 100644 --- a/docs/RESULTS.md +++ b/docs/RESULTS.md @@ -1,6 +1,6 @@ # DeltaSignal: current results -**Last measured (canonical box below)**: 2026-10-03. Section 1 onwards: 2026-09-21, Reactome Release97, catalog build `cat_prod` +**Last measured (canonical box below)**: 2026-10-05. Section 1 onwards: 2026-09-21, Reactome Release97, catalog build `cat_prod` (generator `f2842bc`), solver `5032771` with the current code defaults (`hill_sat`, assembly-limiting on, `DS_INHIBITOR_EPS=1e-12`, `DS_DEPLETION_H_MIN=0.1`). @@ -19,6 +19,37 @@ commit. --- +> **2026-10-05: canonical numbers on generator main with F7 set nodes and cap +> pools (specs/044, 045).** Build `20261004-1959_6990015` (generator 6990015: +> boundary layer split, F7 set nodes, `LNG_CAP_POOLS=1`), solver `5979e48` +> (`DS_SCC_ORDER=flow`), `results/5979e48`. +> +> | | cases | accuracy | macro-F1 | +> |---|---|---|---| +> | **curator held-out, in release** (70 pathways) | 18,573 | **87.93%** | **0.8443** | +> | curator held-out, every case (71 pathways) | 19,000 | 87.34% | 0.8361 | +> | curator, all pathways, in release (81) | 23,625 | 85.50% | 0.8239 | +> | curator, all pathways, every case | 24,100 | 84.95% | 0.8168 | +> | experimental, every case (10 pathways) | 849 | **69.96%** | **0.6007** | +> +> **What moved since the previous box:** +> - **Cap pools: experimental +14, curator tuning +21, held-out 0, nothing +> broken.** All of it is in RAF: past the variant cap, a set is one OR pool +> instead of one all-required bundle. +> - **F7 set nodes:** experimental −5 and curator −11, all in Mitotic G1, as the +> F7 arm predicted. The cause is one faithful new join into the CDK2 loop. +> +> Converged: 1,664 / 1,725 curator and 218 / 244 experimental solves. +> +> **Against MP-BioPath's published figures:** +> - Curator: 84.95% / 0.8168 against 83.34% / 0.8063, **+1.61pp**. +> - Experimental: 69.96% (594/849) against 75.74% (643/849), **−5.8pp**. +> +> Next on the network side is specs/046 (variant nodes end to end). Today 2,504 +> required inputs read a root subunit of a complex the pathway produces. +> +> **Previous box**, kept below for the record. + > **2026-10-03: canonical numbers, now reproducible across rebuilds (specs/013 > amendment 3).** The model is unchanged except the order cyclic components > are swept in: `DS_SCC_ORDER=flow` (default since this date) replaces an order