Issues

• Direction: what the judgment-model use cases imply for bouncer (gate, Stop hook, skill router)

  clownware/bouncer#156 · Filter by author chrispezza opened on Oct 13, 2026

• Measure Clef as a backend on the always-on host (deferred; Jev stays)

  clownware/bouncer#155 · Filter by author chrispezza opened on Oct 13, 2026
  1

• Drift: hard-rule count (19 vs 22), README status line 0.2.9 vs plugin 0.2.20, repo description claims ~100 ms

  clownware/bouncer#153 · Filter by author chrispezza opened on Oct 12, 2026

• calibrate: one timed-out fixture loses the whole run, and the local preflight does not wait for a sleeping server

  clownware/bouncer#152 · Filter by author chrispezza opened on Oct 12, 2026

• ADR-011: open-weights models in the research corpus that the candidate list does not mention

  clownware/bouncer#149 · Filter by author chrispezza opened on Oct 12, 2026
  1

• calibrate: propose per-question thresholds and the uncertainty band from labelled answers

  found-work
  clownware/bouncer#140 · Filter by author chrispezza opened on Oct 12, 2026

• ADR-011: measure the resident general model through the logit path as a candidate

  found-work
  clownware/bouncer#139 · Filter by author chrispezza opened on Oct 12, 2026
  2

• State: send the size of a commit message or PR body, not its text

  found-work
  clownware/bouncer#138 · Filter by author chrispezza opened on Oct 12, 2026

• The hook pays for CLI-only code: bundle size is now a measurable per-call cost

  found-work
  clownware/bouncer#134 · Filter by author chrispezza opened on Oct 12, 2026

• outside_repo: a Write or Edit outside the project prompts wherever the file is; the state cannot say

  found-work
  clownware/bouncer#130 · Filter by author chrispezza opened on Oct 12, 2026
  2

• ADR-011: build the expanded calibration gate and the attested local backend

  clownware/bouncer#127 · Filter by author chrispezza opened on Oct 12, 2026
  2

• Should a server-reported truncation refuse an allow in the hook, like bouncer's own cap?

  enhancement
  clownware/bouncer#118 · Filter by author chrispezza opened on Sep 27, 2026
  1

• Optional: LangChain AutoModeMiddleware as a row over the gate fixtures

  clownware/bouncer#115 · Filter by author chrispezza opened on Sep 27, 2026
  

• PRD §12: a 0.05 mean Brier gap is loose against Jev's 0.059, decide the bar

  clownware/bouncer#114 · Filter by author chrispezza opened on Sep 27, 2026
  

• export: run on the real decision log and read the "did not rebuild" count

  clownware/bouncer#113 · Filter by author chrispezza opened on Sep 27, 2026
  1

  

• docs: README calibration table and un-placeholder the announcement once A, B and D exist

  clownware/bouncer#112 · Filter by author chrispezza opened on Sep 27, 2026
  1

  

• local adapter: first live run of the logit path (llama.cpp or vLLM, ADR-005 as written)

  clownware/bouncer#111 · Filter by author chrispezza opened on Sep 27, 2026
  

• calibrate: first live run of the LLM one-token-logprob arm (jev vs chat@)

  clownware/bouncer#110 · Filter by author chrispezza opened on Sep 27, 2026
  

• calibrate: first live run of the openjev-sglang arm (jev vs jev@modal)

  clownware/bouncer#109 · Filter by author chrispezza opened on Sep 27, 2026
  

• Ship a versions.json so clownware.org can render a version fact strip

  enhancementfound-work
  clownware/bouncer#106 · Filter by author chrispezza opened on Sep 27, 2026

• Key delivery for the desktop app without a plaintext settings.json entry

  clownware/bouncer#99 · Filter by author chrispezza opened on Sep 23, 2026

• Consumer: a correctness and rework policy set, observe-only over real sessions

  enhancement
  clownware/bouncer#95 · Filter by author chrispezza opened on Sep 19, 2026

• Consumer: should the agent take this step — action gating beyond safety

  enhancement
  clownware/bouncer#94 · Filter by author chrispezza opened on Sep 19, 2026

• Consumer: a commit and push readiness set, judged before the agent pushes

  enhancement
  clownware/bouncer#93 · Filter by author chrispezza opened on Sep 19, 2026
  1

• Consumer: validate an agent's output against criteria instead of calling a reasoning model

  enhancement
  clownware/bouncer#92 · Filter by author chrispezza opened on Sep 19, 2026