≈720 tokens on demand. Your agent reads this file only when SKILL.md points to it.
Reporting Anti-Patterns
Purpose: Use this reference when Harvest must ensure reports stay actionable, contextual, and resistant to metric gaming.
Contents
- Report-design anti-patterns
- Goodhart and gaming
- Reporting cadence
- Audience layers
- Quality checklist
Report-Design Anti-Patterns
| ID |
Anti-pattern |
Guardrail |
RA-01 |
Too many metrics |
Keep 3-5 core metrics for the audience |
RA-02 |
No context |
Include previous period, target, or trend |
RA-03 |
Cherry-picking |
Show negative and positive signal together |
RA-04 |
Individual ranking |
Prefer team aggregates; keep personal detail private |
RA-05 |
Over-reporting |
Match frequency to decision cadence |
RA-06 |
Snapshot bias |
Add trend lines or period comparisons |
RA-07 |
No next action |
Attach actions to important findings |
RA-08 |
Blind trust in automation |
Review anomalies before publishing |
Goodhart And Gaming
Typical patterns to watch:
| Signal |
Possible gaming |
| PR sizes suddenly become uniform |
Artificial PR splitting |
| Review times collapse with no comments |
Rubber-stamp approvals |
| Friday evening merges spike |
Weekly metric chasing |
| Coverage jumps without assertion growth |
Hollow test additions |
Reporting Cadence
| Cadence |
Best for |
| Daily |
Build status, open PR count, urgent operations |
| Weekly |
Cycle time, merge volume, review responsiveness |
| Monthly |
Quality trends, DORA-style patterns, effort summaries |
| Quarterly |
Tech debt, long-term architecture or process health |
Audience Layers
| Layer |
Focus |
L1 Executive |
Business impact and delivery status |
L2 Manager |
Team performance and bottlenecks |
L3 Engineer |
Detailed PR-level technical feedback |
Quality Checklist
- Is the audience explicit?
- Are the core metrics limited and contextualized?
- Does the report include both signal and caveats?
- Does it avoid personal ranking?
- Does it end with next actions?
2026 Caveat Additions
- AI period flagging: When the report covers any window in which the team's AI-assistant adoption rate changed materially, flag the change inline. DORA 2025 reports AI now positively correlates with throughput but negatively with delivery stability — so a "throughput up" headline without the stability counter-context is misleading.
- DORA 2025 vocabulary: Use percentile language ("Top 15%", "Top 15-30%") and 7 team archetypes instead of Elite/High/Medium/Low when sourcing DORA-style commentary.
- Copilot Code Review split: When review-time metrics include the Copilot reviewer, separate human and AI review timestamps; otherwise rubber-stamping detection and pickup-time benchmarks misclassify automated comments as human review activity.
