From 6f3982a4dafa9968cb257919c194df170f06b7f6 Mon Sep 17 00:00:00 2001 From: Theodore Song Date: Fri, 21 Aug 2026 15:08:42 -0400 Subject: [PATCH] Differentiate forward learning by agent --- README.md | 11 ++++++ index.html | 110 ++++++++++++++++++++++++++++++++++++++++++++++------- sw.js | 2 +- 3 files changed, 108 insertions(+), 15 deletions(-) diff --git a/README.md b/README.md index e99ae49..c342482 100644 --- a/README.md +++ b/README.md @@ -318,6 +318,17 @@ legs atomically when a positive gap appears. Mixed explicit/implicit years, invalid dates, changed wording, non-Yes/No labels, stale quotes, and non-positive margins are rejected. +Build 79 splits the forward directional learner by agent strategy. Every new +observation stores the agent IDs whose actual acceptance rules matched that +candidate. Graded outcomes retain the scope, and each agent builds an +event-clustered calibration from only its eligible opportunity universe. A +positive Momentum Chaser cohort can therefore promote for Momentum Chaser +without enabling the same trade for Value Hunter or the other agents; a losing +cohort can also veto one strategy without freezing all ten. The Suggestions tab +lists the agents that earned an aggregate promotion. Explicit empty scopes do +not leak to any agent, legacy unlabeled outcomes remain readable for migration, +and eligibility metadata survives local/offline and cloud-state compaction. + Build 77 separates the directional learner's evidence lineage from the global strategy release. Code-history verification found the same trend/reversal generator and 24-hour/72-hour grading policy in Strategy 51 through Strategy diff --git a/index.html b/index.html index 09543d4..82f0d1a 100644 --- a/index.html +++ b/index.html @@ -341,7 +341,7 @@ footer{margin-top:34px;padding-top:22px;border-top:1px solid var(--border);color