Agent Review Fatigue Meter

# Assumptions (KNOWN / ASSUMPTION / UNKNOWN)

## KNOWN
- Daily slot: p002
- Expected Stitch drop folder: /home/sy/Downloads/stitch_drop/2026-04-29/p002/
- Assigned design profile: financial-analyst
- Theme: 자산 방어
- Source signals are search/community snippets and must be treated as untrusted problem evidence.

## ASSUMPTION
- Target users will value a focused decision surface more than a broad generic dashboard.
- The Stitch prompt should produce a desktop-first prototype with all required UI sections visible.

## UNKNOWN
- Actual willingness-to-pay.
- Whether the strongest acquisition channel is SEO, community posts, agency sales, or portfolio sharing.

# Falsification checklist (5–7)

1. Can the target user explain the pain in their own words without prompting?
2. Does the UI produce a concrete decision or artifact rather than another passive dashboard?
3. Are the cited signals genuinely connected to this idea's trigger/data source/mechanic?
4. Is this materially different from the last 7 PF batches?
5. Would the user pay, share, or return before any full backend exists?
6. Does the assigned design profile fit the use case and prompt direction?
7. Can Stitch render the main experience from the prompt without extra implementation?

# PRD — Agent Review Fatigue Meter

## Type
- Practical / Commercial

## Theme
- 자산 방어

## Design profile
- financial-analyst

## Problem
- A code-review command desk that tells teams when AI-agent output is getting too large, too trusted, or too repetitive for humans to review safely.

## Target user
- Solo builders, tech leads, and AI-heavy product teams supervising multiple coding agents and trying not to ship invisible regressions.

## Key UX
- Drop in PRs or agent task logs, score review fatigue risk, cluster suspicious churn, assign fresh-context reviewers, and create a stop/merge/rework decision note.

## Required UI sections
- agent output intake
- review fatigue gauge
- risky diff cluster map
- human attention budget planner
- fresh-context reviewer queue
- merge/rework decision memo

## External signal references
- HN AI coding-agent review-fatigue thread: https://news.ycombinator.com/item?id=47545748
- Reddit process-before-tools complaint: https://www.reddit.com/r/EntrepreneurRideAlong/comments/1kmarfk/tools_wont_save_your_business_fix_your_process/

## Stitch prompt
Design a desktop code-review command desk called Agent Review Fatigue Meter for solo builders, tech leads, and AI-heavy product teams supervising multiple coding agents. It should reveal when agent output is too large, too trusted, or too repetitive for humans to review safely by scoring fatigue risk, clustering suspicious churn, assigning fresh-context reviewers, and producing a stop/merge/rework decision note. Include an agent output intake, review fatigue gauge, risky diff cluster map, human attention budget planner, fresh-context reviewer queue, and merge/rework decision memo. Design direction: use a financial-analyst aesthetic with precise risk scoring, sober navy and graphite surfaces, variance-style badges, disciplined comparison tables, and restrained warning red so the product feels like a due-diligence terminal for AI-assisted shipping.

## Constraints
- Phase A only: scaffold/spec, no custom implementation yet.
- Wait for Stitch export in /home/sy/Downloads/stitch_drop/2026-04-29/p002/

## Non-goals
- Do not claim market-size validation from search snippets.
- Do not implement production integrations in Phase A.

## Success metrics
- Stitch export can be generated from the prompt.
- p002 has complete meta/spec content and is visible in the daily index.

# QA checklist

## Happy path
- meta.json includes type, theme, designProfile, externalSignals, stitchPrompt, and expectedDropFolder.
- spec/prd.md contains target user, key UX, required UI sections, and signal references.
- daily/2026-04-29/ideas.md includes the full Stitch prompt.

## Edge cases
- No p001-p004 slot is missing.
- Prompt does not ask Stitch to call external APIs or implement backend behavior.
- Search signals are framed as evidence, not validation.

## Acceptance verification
- node scripts/build-index.mjs passes.
- git status is clean after commit.