Piper · Coding harness
Log in

Summary / The 27 components

15 Reflection after failure

Extracting a reusable explanation from a failure ("I misunderstood the invariant") instead of retrying blind.

Latest—not scored
Baseline— 
Live use—last 7 days

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

Nothing. The lessons table is empty.

What is missing

Lesson distillation, admission rules and retrieval.

Reviewer Claude, 2026-10-04; close-out verdicts 2026-10-23

Reviewer's read: Closed 10-22: built as the hx reflect stage; no detectable gain

Reconciled into 20 (one store, one call). An evidence point from elsewhere in her system argues for caution: a 09-27 review of her acceptance failures found 95 of 100 were novel, so remembered lessons have little to bite on. The gains are in solving, not remembering.

Close-out, 10-22 (pre-registered, boxed, Claude): built for real as stage S4 (reflection after a failed repro) of the close-out bundle arm hx (A12): h0b + localize pre-pass + repro-first with failure digest + repro-triggered reflect/repair (2 rounds) + critic with one revision. hx against h0b, verified-hard 38 + pro-hard 20, 3 seeds: 103 against 94 cells (verified 65/114 against 55, pro 38/60 against 39). Gain +3.0 tasks of 58, paired bootstrap CI [−1.0, +7.0]; 4 regressions on tasks h0b always solved (limit 2); 2.9× h0b cost per cell. No detectable gain at this model. The pre-registered rule closes every bundled component together, with no halving round. Terminal: not re-run.

How it is measured no detectable gain (hx, 10-22)

See 20.

Adoption rules shared by every arm: How the tests work.

Research & attempts

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.