-
Notifications
You must be signed in to change notification settings - Fork 0
Anti Patterns
Common estimation mistakes and how the skill guards against them.
The mistake: "This will take 4 hours."
Why it's harmful: Creates false certainty. Sets up failure — any deviation looks like a miss.
The guard: The skill always outputs ranges + PERT expected value + confidence bands. You can't get a single number without context.
The mistake: "1 point = 1 day, so 5 points = 5 days."
Why it's harmful: Defeats the purpose of relative sizing. Points measure complexity; hours measure time. They're different dimensions. Once converted, teams game the points to hit hour targets.
The guard: Story points are optional and come with an explicit warning: "Story points measure relative complexity, not time."
The mistake: Only the developer estimates. QA, ops, and PM don't participate.
Why it's harmful: Misses edge cases, deployment concerns, and requirement ambiguities that other roles would catch.
The guard: For L/XL tasks, the skill suggests team review. The detailed questionnaire surfaces integration, risk, and review concerns that force broader thinking.
The mistake: Treating the estimate as a deadline promise.
Why it's harmful: Teams pad defensively. Trust erodes when "estimates" are missed. Planning becomes adversarial.
The guard: The skill separates "expected" (what we think) from "committed" (what we promise) at explicit confidence levels. An expected value of 4 hours with a 90% committed value of 7.2 hours makes the uncertainty visible.
The mistake: "Team A delivers 40 points per sprint, Team B only does 25. Team A is better."
Why it's harmful: Teams game velocity — inflating points, splitting tasks artificially, cutting quality. Cross-team comparisons are meaningless because point scales aren't calibrated.
The guard: The skill never presents velocity as productivity. It focuses on estimate accuracy (PRED(25)), not throughput.
The mistake: Pre-estimating the entire backlog with detailed hour breakdowns 6 months before the work starts.
Why it's harmful: Stories become outdated. The Cone of Uncertainty means early estimates are wildly inaccurate. Time spent estimating stale work is wasted.
The guard: Quick batch mode uses T-shirt sizing (S/M/L/XL) for rough backlog sizing. Detailed estimation is reserved for imminent work.
The mistake: Estimating "rewrite the backend" as a single XL task.
Why it's harmful: XL tasks have the widest uncertainty ranges and lowest agent effectiveness (30%). A single estimate for 2-6 weeks of work is barely more useful than a guess.
The guard: The skill triggers a warning: "XL estimates are directional only — the round-based model cannot capture emergent complexity at this scale. Break into L or smaller tasks before committing to deadlines." Agent effectiveness at XL (0.3) automatically inflates human time, making the cost of not decomposing visible. XL numbers should be treated as rough sizing, not commitments.
| Pattern | Warning |
|---|---|
| Total > 2 weeks | "High uncertainty. Consider phased delivery with re-estimation." |
| Spread ratio > 3× | "Wide range. Consider investigation spike first." |
| Definition = concept | "Concept-phase. Estimates can be off by 2-4×." |
| Task = investigation | "Timebox this. Output is a plan, not code." |
Getting Started
Core Concepts
- How It Works
- Task Types
- Agent Effectiveness
- Confidence Levels
- Cone of Uncertainty
- PERT Statistics
- Small Council
Reference
Accuracy
Contributors