Skip to content

chore: post-pivot cleanse + loop-engineering README + G3 cold-gap receipt line - #56

Merged
PDgit12 merged 3 commits into
mainfrom
chore/cleanse-readme
Jul 6, 2026
Merged

chore: post-pivot cleanse + loop-engineering README + G3 cold-gap receipt line#56
PDgit12 merged 3 commits into
mainfrom
chore/cleanse-readme

Conversation

@PDgit12

@PDgit12 PDgit12 commented Jul 6, 2026

Copy link
Copy Markdown
Owner

What

  • Cleanse (-617 lines, evidence-based): orchestrate.ts (hidden duplicate of loop.ts), 3 orphan research scripts, measure.ts → scripts/ (bench-only consumer). fan/dashboard/hub/evals/profile/learn all KEEP with recorded evidence.
  • README remap: loop-engineering-first (state + optimization + enforcement substrate — can't lie / can't run away / can't go broke); 5-platform enforcement table with honest per-host ceilings; receipt section; commands table fixed (adds onboard/terse, drops orchestrate); npm description + keywords repositioned.
  • G3 receipt line: idle gaps >5min counted and priced — the cold-cache sink optimization can't touch, found from a real session's quota burn.

Verification

  • rm -rf dist && npm run verify EXIT=0 (562 tests)
  • Production audit 50/50 PASS on built artifact; npm audit 0 vulnerabilities
  • Leak audit: published tarball = dist+LICENSE+README only; remote tree has zero internal docs; secret hits = scrubber patterns + FAKE fixtures only
  • Live-verified a–h on real binaries incl. packed tarball (internal docs/VERIFICATION.md)

PDgit12 added 3 commits July 6, 2026 23:13
Cleanse (evidence-based, two exploration passes):
- remove src/orchestrate.ts + its CLI dispatch + test — hidden from help,
  self-labeled P3, duplicate of loop.ts (kill-list: in-session loop plumbing)
- remove orphan scripts research.mjs / research-measure.mjs / shape-profile.mjs
  (unwired from package.json, grep-clean)
- move src/measure.ts -> scripts/measure.ts (bench-only consumer)
- fan.ts: header settles the worktree-kill tension (uses git worktrees,
  is not a worktree manager — the parallelism leg of the loop)
- CLAUDE.md points at the binding product map; stale internal docs dropped

README remap (GitHub = npm):
- reposition loop-engineering-first: state + optimization + enforcement
  substrate — can't lie / can't run away / can't go broke
- 5-platform enforcement table (Claude/Codex/Cursor/Gemini/VS Code) with
  honest per-host capability deltas; Windsurf/Cline stated as advisory
- new receipt section with honest-math rules; ambient/goal/loop/headless
  loop story; commands table gains onboard + terse, drops orchestrate
- honest ceiling added to Numbers: optimization cannot offset cache-cold
  re-reads or subagent spawns
- package.json description + keywords repositioned (loop-engineering,
  agent-loop, hooks, enforcement, codex, gemini-cli)

Every README claim traces to the live-run log (docs/VERIFICATION.md,
internal): real binaries, real MCP stdio driver, packed-tarball re-test.
verify EXIT=0 (560 tests) + pack clean of removed surfaces.
…n can't touch

Idle gaps >5min between session events mean the next turn re-read the whole
context uncached (provider cache TTL). The receipt now counts those gaps and
states the estimated re-read cost + the cheaper move (handoff + fresh session).
Labeled estimate; silent when no gap or no usage data. Found from a real user
session where quota burned 81->90% on cold returns while the receipt said
nothing about the biggest sink.
Copilot AI review requested due to automatic review settings July 6, 2026 17:44

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

@PDgit12
PDgit12 merged commit 73a0e0a into main Jul 6, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants