Create Estimation

Overview

Create a structured three-point estimation using PERT-weighted averages. Breaks work into concrete tasks, estimates each with optimistic/likely/pessimistic bounds, identifies risk factors, and produces a total range with a recommended commitment point.

Workflow

1. Read project context -- Read .chalk/docs/ for PRDs, user stories, acceptance criteria, and architecture docs. Understanding scope and integration points is essential before estimating.

1. Determine the next estimation number -- List files in .chalk/docs/engineering/ matching the pattern estimation.md. Find the highest number and increment by 1. If none exist, start at 1.

1. Understand the scope -- From $ARGUMENTS, conversation context, and project docs, identify:

- What exactly is being estimated (feature, epic, spike, migration) - Acceptance criteria or definition of done - Known integration points with existing systems - Whether this is greenfield or modification of existing code - If scope is unclear, ask the user before estimating -- estimation without scope understanding is guessing

1. Break into tasks -- Decompose the work into concrete, estimatable tasks. Each task should be:

- Small enough to estimate with reasonable confidence (ideally 0.5-3 days of effort) - Independent enough to estimate separately - Inclusive of all work types: implementation, testing, code review, documentation, deployment

1. Three-point estimate each task -- For every task, provide:

- Optimistic (O): Best case if everything goes smoothly, no surprises, familiar territory - Likely (L): Most probable duration given normal conditions, some minor friction - Pessimistic (P): Worst reasonable case -- unfamiliar code, complex bugs, unclear requirements (not catastrophic, just hard)

1. Calculate PERT estimate -- For each task: Expected = (O + 4L + P) / 6. Sum all task estimates for the total.

1. Identify risk factors -- List specific risks that could push the estimate toward the pessimistic end. Assign each a probability (low/medium/high) and impact in time.

1. Calculate risk buffer -- Add a percentage buffer based on cumulative risk:

- Well-understood work, few unknowns: 10-20% buffer - Some unknowns, moderate integration complexity: 20-40% buffer - Significant unknowns, new technology, complex integrations: 40-70% buffer

1. Determine recommended commitment -- The number the team should commit to externally. This is the PERT total + risk buffer, rounded to a sensible unit (half-days for small work, days for medium, weeks for large).

1. Write the file -- Save to .chalk/docs/engineering/<n>estimation<feature_slug>.md.

1. Confirm -- Tell the user the estimation was created with its path, the recommended commitment, and the confidence range.

Filename Convention

<number>_estimation_<snake_case_feature>.md

Examples:

• 2estimationuser_authentication.md
• 5estimationpayment_integration.md
• 9estimationmigratetov2_api.md

Estimation Format

# Estimation: <Feature or Work Item Title>

Last updated: <YYYY-MM-DD>

## Scope

<Brief description of what is being estimated. Link to PRD or user story if available.
State any scope boundaries explicitly: "This estimate covers X but not Y.">

## Assumptions

- <Assumption 1, e.g., "API contracts are finalized and won't change">
- <Assumption 2, e.g., "Design mockups are complete before development starts">
- <Assumption 3, e.g., "One engineer working full-time on this">

## Task Breakdown

| # | Task | Optimistic | Likely | Pessimistic | PERT Estimate |
|---|------|-----------|--------|-------------|---------------|
| 1 | <Task description> | <O> | <L> | <P> | <(O+4L+P)/6> |
| 2 | <Task description> | <O> | <L> | <P> | <(O+4L+P)/6> |
| 3 | <Task description> | <O> | <L> | <P> | <(O+4L+P)/6> |
| ... | ... | ... | ... | ... | ... |
| | **Subtotal** | **<sum O>** | **<sum L>** | **<sum P>** | **<sum PERT>** |

All values in **engineer-days** unless otherwise noted (1 engineer-day = 1 person working 1 full day).

## Commonly Forgotten Tasks

The following are included in the breakdown above. If any were intentionally excluded, note why.

- [ ] Unit and integration tests
- [ ] Error handling and edge cases
- [ ] Code review iterations (typically 1-2 rounds)
- [ ] Documentation updates (API docs, README, runbook)
- [ ] Migration scripts or data backfill
- [ ] Deployment and rollout (feature flags, staged rollout)
- [ ] Monitoring and alerting setup
- [ ] Performance testing under realistic load

## Risk Factors

| Risk | Probability | Impact | Mitigation |
|------|------------|--------|------------|
| <Specific risk> | Low / Medium / High | +<N> days | <How to reduce likelihood or impact> |
| <Specific risk> | Low / Medium / High | +<N> days | <How to reduce likelihood or impact> |

## Summary

| Metric | Value |
|--------|-------|
| **Optimistic total** | <sum of all O values> |
| **PERT total** | <sum of all PERT values> |
| **Pessimistic total** | <sum of all P values> |
| **Risk buffer** | <X%> (<rationale>) |
| **Recommended commitment** | **<PERT + buffer, rounded>** |
| **Confidence range** | <Optimistic> to <Pessimistic + buffer> |

## Recommended Commitment

**<N> engineer-days** (<approximately N/5 weeks> for a single engineer)

This is the number to communicate externally. It includes the PERT-weighted estimate plus
a <X%> risk buffer for <brief rationale>.

## Revisit Triggers

- Scope changes that add or remove acceptance criteria
- Discovery of unknown integration complexity during implementation
- Actual velocity tracking shows >20% deviation from estimates after first 30% of work

Content Guidelines

Estimation Units

Use engineer-days as the base unit. This measures effort, not calendar time.

• 1 engineer-day = 1 person working 1 full day
• To convert to calendar time, divide by the number of engineers and add coordination overhead (typically 10-20% per additional engineer)
• For very small tasks (< 0.5 days), use hours but convert to days in the summary

Calibrating Optimistic vs. Pessimistic

The spread between O and P reveals confidence level:

If any single task has a ratio worse than 1:5, flag it as needing a timeboxed investigation spike before estimation is meaningful.

Breaking Down Tasks Well

Good task breakdown follows these rules:

• Each task has a clear deliverable (not "work on X" but "implement X endpoint with validation")
• Tasks include the full cost: code + tests + review, not just the coding part
• Integration tasks are explicit, not hidden inside implementation tasks
• "Glue work" is accounted for: config changes, environment setup, CI pipeline updates

Setting Risk Buffer

The buffer is not padding -- it is a quantified acknowledgment of uncertainty:

Effort vs. Duration

Always clarify the distinction:

• Effort: Total engineer-days of work (what you estimate)
• Duration: Calendar time to complete (depends on team size, parallel work, interruptions)
• Two engineers don't halve the duration -- coordination overhead adds 10-20% per person
• Context switching between projects adds 20-30% to effective duration

Anti-patterns

• Single-point estimates: "This will take 5 days" is a guess, not an estimate. Always provide the three-point range. The spread itself is valuable information about uncertainty.
• Anchoring bias: The first number mentioned becomes the anchor. Break into tasks first, estimate each independently, then sum. Never start with a total and work backward.
• Not accounting for integration/testing time: Implementation is typically 40-60% of total effort. The rest is testing, integration, review, deployment, and documentation. If your estimate only covers coding, it is 40-60% of the real number.
• Estimating without understanding scope: If you don't have acceptance criteria or a clear definition of done, you cannot estimate. Push back and get clarity first, or estimate the spike to get clarity.
• Confusing effort with duration: 10 engineer-days is not "2 weeks." It is 2 weeks for 1 engineer, but could be 1 week for 2 engineers (plus coordination overhead), or 3 weeks for 1 engineer with context switching.
• Planning fallacy: People consistently underestimate by 30-50%. If the team has historical data, calibrate against it. If not, lean toward the pessimistic end for commitments.
• Precision theater: Reporting "7.3 engineer-days" implies false precision. Round to the nearest 0.5 for small estimates, nearest whole day for medium, nearest week for large.
• Estimating in isolation: If the work integrates with systems owned by other teams, account for coordination time, API negotiations, and blocked-waiting time. Cross-team dependencies are the most common source of estimation misses.