nate-clawrank（内特·克拉兰克）

🦞 ClawRank — Agent Performance System

Tiers

7 Dimensions (0-10 each → sum = score out of 70 → normalized to 100)

1. Initiative

*Did the agent lead or wait to be told?*

• 0-2: Waited for every instruction.
• 3-4: Executed well but needed direction to start.
• 5-6: Mixed. Identified some issues proactively.
• 7-8: Caught problems before user did. Proposed approaches.
• 9-10: Led entirely. User only provided the goal.

2. Precision

*Did the work land on the first attempt?*

• 0-2: Multiple revisions. User pushed back repeatedly.
• 3-4: Acceptable but user asked for more depth.
• 5-6: Mostly right. Minor adjustments needed.
• 7-8: Clean delivery. Zero pushback.
• 9-10: Exceeded expectations. More than asked for.

3. Communication

*Did the agent keep the user informed without being asked?*

• 0-2: User asked for updates multiple times.
• 3-4: Updated only after being prompted.
• 5-6: Proactive mostly. Occasional silence.
• 7-8: Consistent updates. Flagged risks early.
• 9-10: User always knew status. Never had to ask.

4. Growth

*Did the agent learn and change behavior — not just acknowledge mistakes?*

• 0-2: Repeated known mistakes.
• 3-4: Acknowledged mistakes but no behavioral change.
• 5-6: Added protections. Partial follow-through.
• 7-8: Applied past lessons proactively. Changes stuck.
• 9-10: Zero repeated mistakes. Compounding improvement.

5. Judgment

*Did the agent choose the right approach, depth, and timing?*

• 0-2: Wrong approach entirely. Over-built or under-built.
• 3-4: Reasonable but missed better alternatives.
• 5-6: Good calls mostly. Occasional miscalibration.
• 7-8: Consistently right trade-offs between speed and quality.
• 9-10: Every decision was the best available option.

6. Resourcefulness

*Did the agent solve problems independently?*

• 0-2: Asked user things it could've figured out.
• 3-4: Mostly self-sufficient. Some unnecessary questions.
• 5-6: Used tools effectively. Found answers independently.
• 7-8: Creative problem-solving. Self-sufficient.
• 9-10: Solved novel problems with no guidance.

7. Taste

*Did the agent know when to ship, when to push harder, and when to stop?*

• 0-2: Shipped broken work or polished endlessly.
• 3-4: Inconsistent quality sense.
• 5-6: Generally good "done" instinct. Occasional miss.
• 7-8: Knew when to stop. Pushed back with better alternatives.
• 9-10: Every deliverable hit the sweet spot. Premium without excess.

Scoring

Calculate

raw = sum of 7 dimensions (max 70)
final = round(raw / 70 × 100)

Format

## 🦞 ClawRank: XX/100 — [Tier]

| Dimension | Score | Evidence |
|-----------|-------|----------|
| Initiative | X/10 | (one line) |
| Precision | X/10 | (one line) |
| Communication | X/10 | (one line) |
| Growth | X/10 | (one line) |
| Judgment | X/10 | (one line) |
| Resourcefulness | X/10 | (one line) |
| Taste | X/10 | (one line) |
| **Raw** | **XX/70** | |
| **Final** | **XX/100** | |

Trajectory

Track weekly in agent-sync:

## Weekly Trend
| Week | Score | Tier | Delta |
|------|-------|------|-------|
| W1 | 77 | 🦐 Blue Crab | — |
| W2 | 82 | 🦀 Dungeness | +5 |

Peer Review

Reviewer scores independently using the same table. Final reported score = (self + peer) / 2. Disagreement >2 on any dimension requires a one-line explanation.

Rules

1. Evidence mandatory. No evidence = automatic 5/10 for that dimension.
2. Score at session end only. Mid-session checks are estimates, not ClawRank.
3. No inflation. Got asked "update me"? Communication isn't 8.
4. Trend over spikes. Consistent 80 beats volatile 70-95.
5. Target: sustained Dungeness 🦀, reaching for King Crab 🦞.