Estimate

Environment

Read references/environment.md from the ba-toolkit directory to determine the output directory.

Calibration interview

Follow the Interview Protocol: ask one question at a time, present a 2-column | ID | Variant | markdown table of up to 4 options plus a free-text "Other" row last (5 rows max), mark exactly one row (recommended), render variants in the user's language (rule 11), and wait for an answer.

Inline context (protocol rule 9): if the user wrote text after /estimate (e.g., /estimate t-shirt, post-srs phase, +20% buffer), parse it and skip the matching questions.

Before estimating, ask the following (skip questions where the answer is already clear from context):

1. Scale: Story Points (Fibonacci: 1 / 2 / 3 / 5 / 8 / 13 / 21), T-shirt sizes (XS / S / M / L / XL), or ideal person-days? Recommended: Fibonacci Story Points.
2. Reference stories: Do you have known reference stories to anchor the scale? (e.g. "US-003 is a 3" or "login is an S") If yes, calibrate against those.
3. Team assumptions: Full-stack developer pair? Or separate frontend/backend? (affects day estimates only)
4. Out of scope for this estimate: Any stories to skip (e.g., already estimated, on ice)?
5. Estimation phase (Cone of Uncertainty band) — what's the maturity of the underlying artifacts? Recommended: select the band that matches the latest completed pipeline step.
   • Discovery / Brief only — variance ±400% (4× spread); confidence label "Order of magnitude".
   • Post-SRS — variance ±100%; confidence label "Rough order of magnitude (ROM)".
   • Post-AC + Use Cases — variance ±50%; confidence label "Budget estimate".
   • Post-Wireframes + Datadict + APIcontract — variance ±25%; confidence label "Definitive estimate".
   • Post-Implementation Plan — variance ±10%; confidence label "Control estimate".
6. Risk multiplier — apply a buffer for high-risk stories (those linked to 🔴 Critical or 🟡 High risks in 00_risks_*.md)? Recommended: +20% on Critical-linked stories, +10% on High-linked.
7. Technical debt allocation — % of velocity reserved for tech debt, bug-fixes, support? Recommended: 20% (industry baseline).

If the user types /estimate without additional input and prior context is sufficient, proceed with defaults and state assumptions clearly.

Estimation model

For each User Story, assess the following complexity factors:

Apply the reference stories as anchors. If no references are given, calibrate internally: the simplest story in the set = 1 SP or XS.

Do not pad estimates — model what a reasonably skilled team would take, not a worst-case scenario.

Estimation discipline — what this skill is and is not

This skill produces a single-estimator analytical pass. A senior BA running estimation in production would augment this with:

• Planning poker — team-based estimation where each developer commits independently and outliers discuss before re-voting. The canonical agile-team estimation technique. /estimate cannot do this — it must be run with the dev team after the analytical pass.
• Wideband Delphi — multi-round expert estimation with anonymous re-voting. Used when the team is distributed or when politics distort consensus. Same caveat — requires real humans.

Cone of Uncertainty. Boehm's classic principle: estimates made early in the lifecycle (post-Brief, post-Discovery) carry up to 4× variance from the eventual reality, narrowing to 1.25× by the time implementation is detailed. This skill always emits a confidence label based on the user-selected estimation phase (calibration question 5). A 5 SP estimate at the post-Brief phase is "5 SP ± 400%"; the same estimate at the post-Implementation-Plan phase is "5 SP ± 10%".

Single-point estimates without a confidence band are dangerous — they are misread as commitments. Every estimate produced by this skill carries an explicit confidence band tied to the Cone of Uncertainty.

Output

Slug: read the **Slug:** line from the managed block of AGENTS.md (project root, or ../AGENTS.md if cwd is output/) and use it verbatim. See ../references/slug-source.md.

If scope ≤ 20 stories — update 03_stories_{slug}.md in-place

Add a **Estimate:** field to each US block and append the Estimation Summary table at the end of the file.

If scope > 20 stories — create 03_stories_estimates_{slug}.md

Standalone estimation report with the full table.

Estimation Summary table (always produced in chat as closing output)

## Estimation Summary — [PROJECT_NAME]
Scale: [Story Points / T-shirt / Days]
Estimation phase: [Discovery / Brief / SRS / AC / Wireframes / Implementation Plan]
Confidence band: [±400% / ±100% / ±50% / ±25% / ±10%]
Confidence label: [Order of magnitude / ROM / Budget / Definitive / Control]

| US | Title | Epic | Estimate | Confidence | Key drivers |
|----|-------|------|----------|------------|-------------|
| US-001 | [Title] | E-01 | 3 SP ± 25% | Definitive | 2 AC scenarios, simple UI |
| US-002 | [Title] | E-01 | 8 SP ± 25% | Definitive | external payment API, 4 AC scenarios |
| US-003 | [Title] | E-02 | 13 SP ± 50% | Budget | new domain area, complex state machine, RISK-02 +20% |
...

| Metric | Value |
|--------|-------|
| Total stories estimated | [N] |
| Total Story Points (point estimate) | [N] SP |
| Total with confidence band | [N – N] SP ([low] – [high]) |
| Largest story | US-[NNN] ([N] SP) — [reason] |
| Stories ≥ 13 SP (consider splitting) | [N]: US-[NNN], … |
| Average per story | [N] SP |
| Risk-adjusted total (high-risk +20%) | [N] SP |
| Tech debt reservation ([N]%) | [N] SP |
| **Net feature capacity** | [N] SP |

⚠️ Single-estimator caveat. This is an analytical estimation pass by a single estimator. For a real team commitment, run planning poker with the dev team using these numbers as the starting anchor and re-vote on any item where the team's estimate differs by more than one Fibonacci step.

Splitting recommendations

For any story estimated at 13 SP or higher (or XL), suggest a concrete split:

⚠️ US-[NNN] ([N] SP) is large. Consider splitting:
  → US-[NNN]a: [sub-story suggestion]
  → US-[NNN]b: [sub-story suggestion]
Run /split US-[NNN] to perform the split.

Closing message

After saving, present the following summary (see references/closing-message.md for format):

• Saved file path (if stories artifact updated) or new estimate file created.
• Scale used and total estimate.
• Number of stories estimated.
• Stories flagged for splitting (if any).
• Assumptions made (if defaults were applied without a calibration interview).

Available commands: /estimate [US-NNN] (re-estimate a story) · /split [US-NNN] (split a large story) · /analyze · /done

Style

Be explicit about the rationale for each estimate. Use concise bullet drivers, not paragraphs. Generate output in the language of the artifact — see references/language-rule.md for what to translate and what stays in English.