nickgallick Strategic Tempo Evaluation — Skill 54
Purpose
Measure not just correctness, but the quality of TIMING decisions. Tempo is WHEN the agent explores, commits, verifies, pivots, escalates, and defers.
Core Concept
Great agents often outperform average ones because they control tempo better — not because they write better code.
Tempo Signals
Negative Tempo Signals
| Signal | Description | Detection |
|---|---|---|
| Over-commitment | Coding before reading the full codebase | First substantial edit < 2 min after start |
| Unnecessary delay | Reading every file when the bug is in one obvious place | > 50% of time in exploration with low information gain |
| Bad verification timing | Writing 500 lines before running a single test | Large diff with no intermediate test runs |
| Failure to checkpoint | No intermediate test runs between major changes | Edit streak > 10 files without any test execution |
| Too-late pivoting | 80% of time on wrong approach before switching | Major direction change after 80%+ time elapsed |
| Reckless acceleration | Submitting without testing after running out of patience | Final submission with no test run in last 20% of time |
Positive Tempo Signals
| Signal | Description | Detection |
|---|---|---|
| Productive patience | Spending time understanding before acting (when warranted) | Exploration phase proportional to problem complexity |
| Well-timed tool use | Using search/debug tools at the right moment | Tool calls that change the agent's approach productively |
| Incremental verification | Testing after each meaningful change | Test runs distributed throughout the session |
| Early pivoting | Recognizing wrong approach quickly and switching | Direction change before 30% of time elapsed |
| Strategic escalation | Moving to more powerful tools/approaches when simpler ones fail | Tool complexity increases with problem difficulty |
Tempo Scoring
Tempo is NOT pure speed. It is the quality of timing decisions under uncertainty.
Example: Two agents, both finish in 30 minutes
Agent A (Reckless):
0:00 - 2:00 → Quick scan
2:00 - 27:00 → Coded continuously
27:00 - 28:00 → Ran tests once
28:00 - 30:00 → Submitted
Tempo score: Low — no incremental verification, no exploration proportional to complexity
Agent B (Disciplined):
0:00 - 8:00 → Read codebase, identified key areas
8:00 - 11:00 → Planned approach, documented hypotheses
11:00 - 23:00 → Coded in chunks, tested 3 times
23:00 - 26:00 → Reviewed, refined based on test results
26:00 - 30:00 → Final verification, submitted
Tempo score: High — exploration matched complexity, incremental verification, hypothesis-driven
Average vs Elite Tempo Patterns
Average agents either:
- Rush and break things (reckless tempo)
- Hesitate and drown in context (paralyzed tempo)
Elite agents:
- Know when to act (calibrated tempo)
- Phase transitions are deliberate and appropriately timed
Measurement
Tempo is primarily evaluated by the Process Judge, using session telemetry:
| Telemetry Source | What It Reveals |
|---|---|
| Timestamps on tool calls | Exploration vs execution ratio |
| File read patterns | Breadth-first vs depth-first exploration |
| Edit-to-test intervals | Verification discipline |
| Direction changes with timestamps | Pivot timing quality |
| Tool call sequence | Escalation pattern |
Tempo-CDI Interaction
Challenges that produce high tempo variance across agents are highly discriminative. If all agents have the same tempo pattern, the challenge doesn't test timing decisions.
Good challenge design should create tempo decision points — moments where:
- The right amount of exploration is non-obvious
- The right time to commit is ambiguous
- The right verification strategy depends on understanding the problem
Integration Points
- Process Legibility (Skill 55): Tempo is one component of overall process quality
- Failure Archetypes (Skill 48): Premature Convergence and False Confidence Stop are tempo failures
- Adaptive Phases (Skill 53): Phase shifts create forced re-planning that tests tempo flexibility
- CDI (Skill 46): Tempo variance contributes to Score Variance Quality