SkillOptimizer
Research hub
Null resultShare of skills

Tasks the assistant already solves

Does a stricter task prompt reduce tasks the assistant solves perfectly without the skill?

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

Headline figures

10

Skills drawn

measured-local · 2026-09-19

2

Skills that cannot run offline, no task graded

measured-local · 2026-09-19

16

Tasks graded without the skill

measured-local · 2026-09-19

6of 16

Tasks the unaided host scored 1.000

measured-local · n=16 · 2026-09-19

37.5%

Share already solved unaided

measured-local · 2026-09-19

32.4%

The same share under the earlier prompt, from the beats-base-model study

measured-local · 2026-09-19

Notes

  • A fresh seeded draw of different skills, so the prompt change is confounded with the draw.
  • Floor-perfect has more than one cause: some checks are satisfiable by a competent unaided answer whatever the prompt says.

Tables

Tasks graded without the skill under the stricter prompt

n: Runs without the skill behind the row. 18 rows.

skilltask_idgradedrepetitionsscore_none_medianfloor_perfectnevidence_classdatecommit
ce-polish-betapolish-login-formYes30.6No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
ce-polish-betapolish-product-gridYes30.8No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
design-negotiationtimeline-compression-negotiationYes31Yes3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
design-negotiationscope-reduction-already-builtYes31Yes3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
enrich-lead—No———0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
gh-create-issuefile-bug-report-issueYes30.3333No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
gh-create-issuefile-feature-request-issueYes31Yes3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
github-codespaces-efficiencyaudit-wasteful-devcontainerYes30.2No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
github-codespaces-efficiencyoptimize-machine-and-prebuildsYes30.8No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
memory-reviewer—No———0measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
paperkitios-markup-editorYes30.2No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
paperkitmacos-markup-safe-loadYes30.4No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
speccreate-cache-specYes31Yes3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
specamend-spec-taskYes30.8No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
validate-datasql-query-denominator-validationYes31Yes3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
validate-datamulti-issue-analysis-validationYes30.6667No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
validating-and-publishing-canvasescanvas-capability-declarationsYes30No3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590
validating-and-publishing-canvasescanvas-version-conflict-mergeYes31Yes3measured-local2026-09-1980e4a13fb96ebb96d8efff8690a8c77215657590