Agent Skills: Estimation

Estimation

Provisional pending the first-engagement gate. This wiring is the floor, not the finished standard; refine it after the first real paid engagement is estimated and delivered against.

Overview

A fixed-cost scoping / estimation discipline that de-risks fixed-cost bids. It is deliberately light — a checklist plus a method, not a model. The job is to turn an engagement into a defensible number: a work breakdown, PERT-weighted effort with uncertainty ranges, explicit assumptions & exclusions, and contingency — then hand those to the SoW so the price and the scope boundary rest on the same evidence.

It is independently triggerable (any "estimate this" / "PERT estimate" request) and is also the pricing step of the proposal-writing flow.

It composes by reference: the estimate decomposes against the build breakdown owned by epic-planning, and its sequencing/roadmap framing comes from decision-package. It restates neither.

What it covers (R4)

1. Decomposition (work breakdown). Break the engagement into estimable leaf items — small enough that each can carry a three-point estimate. For build-tier work, decompose against the EPIC/PLAN breakdown from epic-planning; do not invent a parallel breakdown.
2. Uncertainty ranges. Each leaf gets a range, not a point. PERT (below) is how the range becomes a weighted estimate plus a variance you can roll up.
3. Assumptions & exclusions. State what the number assumes (inputs, access, decisions, availability) and what it explicitly excludes. The exclusions are the SoW out-of-scope list — one list, two documents (see proposal-writing).
4. Contingency. A buffer sized from the rolled-up variance and the assumption risk — not a flat guess. The worksheet shows the roll-up.
5. Link to the SoW scope boundary. The breakdown's leaves are the in-scope items; the exclusions are the out-of-scope items; the PERT total + contingency is the fixed-frame price.

Techniques: PERT is the default

Multiple estimation techniques exist; PERT is the favoured system here because three-point estimates make fixed-cost uncertainty explicit and roll up into a defensible contingency.

PERT (three-point) — the default. For each leaf, capture three estimates and weight them:

Expected (TE) = (O + 4·M + P) / 6        SD = (P − O) / 6        Variance = SD²

• O = optimistic (everything goes right), M = most-likely, P = pessimistic (plausible worst case — not catastrophe).
• Roll up: engagement TE = Σ leaf TE; engagement SD = √(Σ leaf variances).
• Size contingency from the rolled-up SD (e.g. a one- or two-SD buffer for the confidence level the fixed frame needs). The worksheet does the arithmetic.

Alternatives (use when they fit better, or to triangulate):

| Technique | Use when | |-----------|----------| | Analogous (top-down, from a similar past engagement) | Early, little detail; a quick sanity anchor | | Parametric (units × rate, e.g. per source system / per ontology module) | The work scales with a countable driver | | Bottom-up (sum of leaf estimates) | Detail exists; pairs naturally with PERT per leaf |

Triangulate when the stakes are high: a PERT bottom-up that disagrees with an analogous top-down is a signal to re-examine the breakdown.

Gantt / scheduling is EXTERNAL

This skill produces the work breakdown + estimates. It does not draw Gantt charts or schedules. Scheduling (dependencies, calendar, resource levelling, the Gantt itself) is rendered in external tools — Smartsheet or MS Project — by importing the work breakdown. Do not attempt to draw Gantts in-repo; produce a clean breakdown that those tools can consume. Sequencing intent (pilot → scale order) comes from decision-package; the calendar rendering of it is the external tool's job.

The checklist

Run before any number leaves the room:

• [ ] Work broken into leaves small enough for a three-point estimate (decomposed against the epic-planning breakdown where build work is involved).
• [ ] Each leaf has O / M / P and a PERT expected value.
• [ ] Variances rolled up; engagement SD computed.
• [ ] Contingency sized from the rolled-up SD + assumption risk (not a flat percentage guess).
• [ ] Assumptions listed; each risky assumption has an exclusion or a stop/extend trigger.
• [ ] Exclusions = the SoW out-of-scope list (the invariant in proposal-writing).
• [ ] Triangulated against at least one alternative technique when the bid is large.
• [ ] Work breakdown exported clean enough for Smartsheet / MS Project to schedule.

The fillable worksheet (PERT table + roll-up + contingency) is in references/pert-worksheet.md; the full checklist is in references/estimation-checklist.md.

Boundary & Related Skills

This skill OWNS: the estimation / scoping discipline — PERT + work breakdown + uncertainty ranges + assumptions/exclusions + contingency, and the clean breakdown that external scheduling tools consume.

This skill DELEGATES:

• The proposal / SoW wrapper (and the scope-boundary discipline the exclusions feed) → proposal-writing.
• The build breakdown the estimate decomposes against → epic-planning.
• Sequencing / roadmap intent (pilot → scale) → decision-package.
• The calendar/Gantt rendering → external tools (Smartsheet, MS Project) — out of repo.

Related: proposal-writing, epic-planning, decision-package.

Tone & style

British English. Terse and numeric. Every number traces to a leaf, an assumption, and a technique; no unexplained round figures. A point estimate without a range is a smell — show the range and the contingency that covers it.