SkillOptimizer
Research hub
Null resultAI usage (tokens)

The mission rule, out of sample

Does refusing a rule check that forbids the skill's own purpose hold on skills it never saw?

measured-local2026-09-1980e4a13
  • docs/experiments/DETECTION-2026-09-19.md · 80e4a13 · 2026-09-20

Headline figures

35

Unseen skills the rule could see

measured-local · 2026-09-19

2

Rule check refused by the rule before measurement

measured-local · 2026-09-19

18

Comparisons the rule let through

measured-local · 2026-09-19

5

Let through, and the mission regressed

measured-local · 2026-09-19

0of 5

Of those, accepted anyway

measured-local · n=5 · 2026-09-19

Notes

  • Precision is unobservable here: a refused rule check never ran its improved arm. Re-measuring the two refusals with the rule off would recreate the mission-breaking runs lane K already paid for, so it was not run.
  • The five regressions it let through are a different failure, the agent thrashing against the rule check; the net-call and cheaper-than-original conditions rejected all five.

Tables

The mission rule on the unseen bands

n: Runs per arm behind a comparison; 0 when the skill was not compared. 35 rows.

skillbandstatusrule_outcomeacceptednevidence_classdatecommit
activitykitabsolutecomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
app-clipsabsolutecomparedpassed_mission_regressedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
arize-traceabsoluteunmeasurablenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
auto-repo-setupabsolutecomparedpassed_mission_preservedNo6measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
axiom-dataabsolutecomparedpassed_mission_preservedNo6measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
computer-useabsoluteno_baseline_advantagenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
cpp-coding-standardsabsoluteno_baseline_advantagenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
cpp-testingabsolutecomparedpassed_mission_regressedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
fix-merge-conflictsabsolutecomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
no-commentsabsoluteno_baseline_advantagenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
pr-screenshotsabsolutecomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
quantum-qiskitabsolutecomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
redis-expertabsoluteno_baseline_advantagenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
shepherd-prabsoluteunmeasurablenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
shopify-review-triageabsolutecomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
supersede-prabsoluteunmeasurablenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
terraform-skillabsolutecomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
webiny-admin-architectabsolutecomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
webiny-full-stack-architectabsolutecomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
webiny-mailer-smtpabsolutetasks_not_discriminatingnot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
atmos-secretsconditional-restartifact_refusedrefused_by_ruleNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
dmr-from-django-ninjaconditional-restcomparedpassed_mission_regressedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
publish-project-to-githubconditional-restartifact_refusedrefused_by_ruleNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
review-prconditional-restcomparedpassed_mission_preservedNo6measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
terraformconditional-resttasks_not_discriminatingnot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
baoyu-url-to-markdownpath-scopedno_baseline_advantagenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
code-splittingpath-scopedcomparedpassed_mission_preservedNo6measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
harborpath-scopedunmeasurablenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
maple-onboarding-stylepath-scopedcomparedpassed_mission_regressedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
presentationpath-scopedno_baseline_advantagenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
prp-spikepath-scopedcomparedpassed_mission_regressedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
research-litpath-scopedcomparedpassed_mission_preservedNo3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
tech-searchpath-scopedunmeasurablenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
triagepath-scopedno_baseline_advantagenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
workflowpath-scopedno_baseline_advantagenot_reachedNo0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590