Summary / Benches
B2 piperbench
Twelve coding tasks from her own repositories, each with a gold test.
Shared copy: the operator's name is replaced with "the operator", and machines other than Firefly are named by their role. Nothing else is changed.
What is built now
12 tasks with gold tests in her coding-cortex repository, used as smoke tests during the build.
What is missing
Growth to 40 tasks mined from her recovery snapshots, and wiring as a per-commit smoke test.
Reviewer Claude, 2026-10-04
Reviewer's read: Small; used for smoke only
Her repositories are not git repositories, so "mine the history" has to go through the hourly recovery snapshots. The replay bench (B5) has since done a better version of the same idea, using her own past builds with their original acceptance tests.
How it scores smoke only
Pass/fail per task on each harness commit.
Adoption rules shared by every arm: How the tests work.
Research & attempts
No documents yet.
Score history
No scores yet. The first entry is the baseline for suite v1.
Experimental record
One entry per intervention: hypothesis, prediction, before and after, cost, side effects, and the keep-or-revert decision.
No interventions recorded yet.