Skip to content

[design] Define cognitive-load proxy metric to measure A3 anti-pattern doc + Phase 0.4.5 fail-fast effectiveness (residue from #119) #131

Description

@kiki830621

Problem

#119 residue:NSQL §4.6 殘留 item 1。原 Spectra path 評估時提出的未解 acceptance metric 需求 — 「reviewer cognitive load 爆表」是主觀體感,沒可量化指標證明 chain anti-pattern (A3) 用法是否真的有改善。

Quoted residue(from #119 diagnosis):
「reviewer cognitive load 爆表」是主觀體感 — 沒可量化 acceptance(14 diffs vs 1 diff,認知負荷沒測量單位)。Spectra discuss 階段建議定 proxy:每 PR diff 行數中位數、或每 PR 跨 module 數

Post-#119 implementation 已 ship Phase 0.4.5 fail-fast + workflows.md A3 doc 教育。但我們不知道這個 reframing 是否真的工作 — anecdotally 看 user 不再撞 14-issues case,但沒量化證據。

Type

design / metric(observability)

Expected

定義 1-2 個 proxy metric 追蹤 reframing 效果:

  • Anti-pattern A3 fire 次數(/idd-all-chain Phase 0.4.5 觸發 abort 的 count over time)
  • Per-cluster-PR / per-chain-PR diff 行數中位數(cluster vs atomic 是否 reviewer load 真的有差?)
  • 或:每月跑 /idd-all-chain 的 invocation 次數 vs /idd-all batch 比例(user 是否真的 prefer batch path 在 multi-root case?)

不需要完整 observability stack,初版可手動 git log --grep + gh api search 跑 ad-hoc query。

Actual

無 metric;Phase 0.4.5 fail-fast 是否真有抑制 user 誤用,完全 anecdotal。

Source

residue from #119 at /idd-close time (Step 3.6)


Current Status

Phase: diagnosed
Last updated: 2026-07-17 by /idd-diagnose (attended batch)

Key Decisions

  • 維持 deferred(H-cluster 裁決)— Simple when triggered(實證失效訊號)

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions