frumu-ai/development-estimation

Development Estimation

Purpose

Create consistent, defensible estimates by:

• Triaging and clarifying scope (especially for bug lists)
• Decomposing work into deliverable components
• Accounting for testing, QA, and release overhead
• Identifying risks and unknowns with explicit mitigation
• Producing confidence-based ranges (P50/P80/P90)

Optimized for reviewing an existing project or backlog and producing a realistic delivery forecast.

When to Use

• Estimating feature or project scope and delivery timelines
• Reviewing a backlog and producing a time estimate
• Producing risk-aware estimates with confidence intervals
• Converting a list of problems into an execution plan

Avoid When

• The task is trivial and a gut-check is sufficient
• Requirements are unknown and discovery is the real work
• You lack access to the codebase or requirements

Inputs Required

Provide as much as possible; estimate quality depends on clarity.

Minimum

• List of items (bugs/features) with short descriptions
• Target platform/environment (e.g., iOS/Android/Web, staging/prod)
• Constraints (deadline, must-fix vs can-slip)

Helpful

• Repro steps, logs, screenshots
• Architecture notes or codebase access
• Deployment pipeline and QA process
• Team capacity model (engineers, focus hours/day, meeting load)
• Definition of done (tests required, QA sign-off, release scope)

Estimation Frames

Always confirm the estimate frame up front:

• Effort: engineer-days or engineer-weeks
• Calendar: derived from capacity (effort / throughput)
• Cost: optional, requires rate assumptions
• Complexity: t-shirt sizing or points (only if requested)

Default Capacity Model

Unless provided, assume:

• 5–6 hours/day of effective build time per engineer
• Review and coordination overhead included per item
• QA/release overhead is stated explicitly as an assumption

Triage Rubric (Bug Backlog)

For each bug, capture:

Repro

• A: Always
• B: Sometimes / flaky
• C: Cannot repro / unclear

Clarity

• A: Clear cause or clear fix path
• B: Partial (needs investigation)
• C: Unknown (needs discovery)

Impact

• Blocker / High / Medium / Low

Fix Type

• UI / Backend / Infra / Data / Dependency / Config / Unknown

Risk Flags

• Concurrency / migrations / auth/security / payments / third-party API
• Performance / cross-platform behavior / state persistence
• Touches core path / unknown blast radius

Rule: If Repro=C or Clarity=C, include a discovery slice.

Workflow

1) Define Scope and Estimate Type

• Confirm what is in/out of scope
• Confirm estimate type: effort, calendar, cost, complexity
• Confirm confidence target(s): P50 / P80 (default), optionally P90
• Confirm constraints: deadlines, must-fix items, release boundary

2) Intake and Normalize the Backlog

• De-duplicate issues
• Merge near-duplicates and note uncertainty
• Group items by subsystem (auth, billing, UI, sync, infra)
• Identify unknown-unknown areas (legacy zones, flaky tests, poor observability)

3) Per-Item Breakdown

Estimate each item using this structure:

• Triage / reproduce
• Investigation / diagnosis (if needed)
• Implementation
• Tests (unit/integration/e2e as appropriate)
• Code review and iteration
• QA verification
• Release / deployment steps (if applicable)

Use hours only for work expected to be under one day; otherwise use engineer-days.

4) Dependencies and Sequencing

• Identify dependencies (internal modules, data migrations, external APIs, app-store releases)
• Mark blockers and parallelizable streams
• Flag sequencing needs (must land before X)

5) Risks, Unknowns, and Buffers

Maintain an Unknowns Register:

• Unknown → discovery task → decision/validation method → timebox

Risk buffer guidance:

• Low risk: +0–10%
• Medium risk: +10–25%
• High risk: +25–50%
• Unknown-heavy: include discovery and a re-estimate gate

6) Roll-Up Totals and Confidence

Produce:

• Per-item ranges (min/most-likely/max or P50/P80)
• Totals by bucket (must/should/nice)
• Overall confidence:
  • P50: likely if things go normally
  • P80: realistic commitment range
  • P90: conservative range for high-stakes delivery

7) Validation and Re-Estimation Plan

Provide:

• Fastest validation steps to reduce uncertainty (logs, repro harness, spike)
• Re-estimation checkpoints (after discovery spikes, after first tranche)

Output Format (Required)

A) Executive Summary

• Scope statement (included/excluded)
• Estimate frame (effort vs calendar + capacity assumptions)
• Total estimates: P50 / P80 (and P90 if requested)
• Key risks and biggest unknowns

B) Backlog Estimate Table

For each item:

• ID / Title
• Bucket (Must/Should/Nice)
• Subsystem
• Triage scores (Repro/Clarity/Impact)
• Estimate (P50/P80)
• Risk flags
• Notes / assumptions

C) Totals by Bucket

• Must-fix total (P50/P80)
• Should-fix total (P50/P80)
• Nice-to-have total (P50/P80)

D) Assumptions and Exclusions

• Explicit assumptions (env access, test maturity, release cadence)
• Explicit exclusions (UX redesign, perf overhaul, refactors unless listed)

E) Risks and Unknowns Register

• Unknown → discovery plan → timebox → impact if true

F) Next Steps

• Top 3 actions to validate quickly
• Proposed sequencing / milestones
• Re-estimation trigger points

Common Mistakes

• Estimating a bug list as if it is already well-specified
• Skipping repro/triage and undercounting diagnosis time
• Ignoring testing, QA verification, and release overhead
• Mixing effort and calendar without a capacity model
• Overstating precision (single-number estimates with no confidence)
• Not separating must-fix from nice-to-have

Quick Reference

• Reference template: skills/development-estimation/references/estimate.md
• Preferred output: P50/P80 effort plus derived calendar via stated capacity
• Default unit: engineer-days (hours only for sub-day tasks)

Quality Bar

A good estimate is:

• Transparent about assumptions
• Broken down into components
• Risk-aware with explicit unknown handling
• Delivered as a range with confidence levels
• Paired with a plan to validate and tighten the range