AI-assisted effort scale (Fibonacci)

Official router: effort-scales.yaml (ai_fibonacci).

Use this scale when sizing plan steps for AI-augmented work. Points reflect anticipated token churn, coordination complexity, and whether work fits a single branch/PR — not wall-clock developer time.

Standard range: 1 → 8. Use 13 only when work clearly exceeds ~two 8s combined.

Points	Meaning
1	Fixer-upper — tweak a config value, copy, env var, or other small targeted change.
2	A bit more — small addition to something that already exists (one file or a tight pair).
3	A few focused edits across a handful of files; one branch; a simple model can carry it.
5	Moderate diff (~10 files or equivalent scope); still one branch, but more context to hold.
8	Large context window, multiple PRs, or a validation gate before the next step.
13	Epic — cross-cutting, security-sensitive, or multi-surface work that would otherwise stack multiple 8s.

Sizing rules

• Assign one Fibonacci value per step — never a range.
• Phase or plan rollups are sized independently (they are not sums of child steps).
• Do not mental-arithmetic totals — run scripts/summarize-effort.mjs instead.
• in_progress steps may include effort_remaining (also Fibonacci, ≤ effort) when meaningful work is already done.

Distinction from wall-clock T-shirt sizing

remediation-plan task sizing (wall_clock_tshirt in taxonomy) uses wall-clock effort for human team capacity. This scale is for WIP progress tracking under AI assistance. See effort-scales.yaml for lifecycle handoff and indicative mapping when converting from a remediation draft.