wishket-feedbacklisted

# wishket-feedback — Profile Calibration via Application Outcomes ## Flow ```mermaid flowchart LR A[pipeline store (state.db)] --> B{Sufficient sample?} B -- Under 5 items --> Z[Notify insufficient data] B -- 5+ items --> C[Cross-analyze win rates by grade/stack] C --> D[Detect anomalies] D --> E[Propose adjustments] E --> F{User approval} F -- Approved --> G[Edit profile.yaml] ``` ## Step 1: Gather Data - Read applications from `~/.wishket-radar/state.db` (SQLite — `sqlite3 ~/.wishket-radar/state.db 'select data from applications'`, 행마다 Application JSON). 구 `applications.yaml`은 자동 이관된다. If there are fewer than 5 closed items (`체결`, `진행 중`, `완료`, `미체결`, `탈락`), terminate analysis and advise collecting more outcome data first. - Analyze only closed items; exclude in-progress states (`관심`, `지원`, `상담`, `미팅`). - Check step-by-step conversion rates before overall win rates. ## Step 2: Anomaly Detection Identify discrepancies between scout fit grades and actual outcomes: - **Grade A win rate < Grade B win rate**: Matching score misaligned. Over-weighted keywords in Grade A announcements may be false positives. Suggest lowering weights or splitting keywords. - **Consistent rejections on specific stack**: Weak experience or competitive market rate issues. Suggest lowering weight or removing from `roles`. - **Consistent wins on specific keyword**: Core strength. Suggest increasing weight or adding related synonyms. - **Applications stalled before mana