---
name: convention-lean-discovery
type: convention
status: proposed
description: >-
  The lean discovery kit Sonar reaches for when a full engagement-kickoff is too
  heavy for a first pass — a minimum-viable set of four artefacts (actor card,
  who-does-what map, findings + opportunities, lean prioritisation) that moves
  fast to low-hanging fruit and key-impact priorities, and whose outputs double
  as reusable, promptable AI inputs. Read before running a lean first pass, or
  when an engagement needs direction quickly rather than the full
  playbook-engagement-kickoff. Sits beside convention-user-modelling and
  convention-audit (it composes their definitions in a lean form) and feeds
  skill-prioritise. Surfaced by a1.Muse. Proposed first draft — confirm before
  adoption.
---

# Lean discovery

The fast first pass. A full engagement (`playbook-engagement-kickoff`, `convention-user-modelling`, the whole audit) is the right depth eventually, but too heavy for a first look that just needs to find the low-hanging fruit and the one or two key-impact priorities. Lean discovery is the **minimum-viable kit** that gets there quickly — four artefacts, deliberately compact — and whose outputs are written as **reusable AI inputs**, so the same fielded artefacts feed later AI-assisted work rather than being thrown away. It is a lean sibling of the full toolkit, not a replacement: it composes the same definitions (`convention-user-modelling`, `convention-audit`, `standard-prioritisation`) at lower resolution and decomposes into them when depth is warranted.

## When to reach for it

- A first pass on a venture or funnel, where speed to "where do we win" beats completeness.
- A quick read across several actor × product cells, where modelling each in full would be prohibitive.
- The opening move of an engagement that may later graduate to the full `playbook-engagement-kickoff`.

If the object is a funnel specifically, the leak-finding still routes through `skill-conversion-diagnostic`; lean discovery frames the actor and locates the intervention, it does not replace the diagnostic.

## The four artefacts

One pass produces these four, in order. Each is **fielded and confidence-tagged** (`known` / `assumed` / `unknown`), so it reads as a promptable AI input, not prose.

1. **Actor card.** A compact composite of segment + core JTBD + forces + the one or two metrics that matter, for one actor. This is the reusable "this user" input every later artefact and prompt references. It **deliberately overloads** — it folds several `convention-user-modelling` artefacts into one card, breaking that convention's one-artefact-per-job rule **on purpose, for speed**. The cost is acknowledged: decompose the card into proper segment / persona / JTBD / forces artefacts when the engagement deepens.

2. **Who-does-what map.** A lean stage spine × {actor action, touchpoint, served-by}. It locates an intervention by **who and where** — not "fix the website" but "the broker, at the quote stage, on the comparison touchpoint, served by X". It is the lean form of a service blueprint: enough to place a finding, not the full backstage/support lanes.

3. **Findings + opportunities.** Captured on the `convention-audit` card ladder (Finding → Theme → Opportunity), each tagged by stage / theme / impact / confidence. **Cluster only where the grouping is obvious** — lean discovery does not force an exhaustive theme synthesis.

4. **Lean prioritisation.** `standard-prioritisation`'s Impact × Effort × Confidence, with the `known/assumed/unknown` overlay, producing a short ranked list and a **research-first track** for the high-impact unknowns. Feeds `skill-prioritise` when the engagement formalises.

## Principles

- **Lean over complete.** Coverage is traded for speed deliberately; gaps are tagged `assumed` or `unknown`, not filled by invention.
- **Every artefact is a promptable AI input.** Keep a consistent fielded format so an artefact drops straight into a prompt — e.g. *"audit [object] as [actor card] pursuing [JTBD] at [stage]"*. This is the reason the kit exists in this shape.
- **Build-on-as-we-go.** Fields are refined over time; more is `assumed` now, more becomes `known` later. The card is a living input, not a one-shot deliverable.
- **Replicate across the actor × product matrix.** Model one cell fully, then vary the deltas across the matrix rather than re-modelling each from scratch — e.g. broker-solo / team × finance / FX, SME × finance / FX, plus distinct actors that are not funnel users (a lender / originator such as Caisus is its own actor, not a stage in someone else's funnel).

## Relationship to the full kit

| Lean artefact | Decomposes into (full kit) |
| -- | -- |
| Actor card | segment, persona, JTBD layers, forces (`convention-user-modelling`) |
| Who-does-what map | service blueprint (`skill-service-blueprint`) |
| Findings + opportunities | discovery audit / conversion diagnostic (`convention-audit`, `skill-discovery-audit`) |
| Lean prioritisation | full prioritisation (`standard-prioritisation`, `skill-prioritise`) |

Start lean; graduate a cell to the full artefacts only where depth earns its cost.

## Guardrails

- **The overload is intentional and named.** The actor card knowingly breaks one-artefact-per-job; never treat it as a finished model — decompose before relying on it for deep work.
- **Tag, don't invent.** Lean speed comes from honest `assumed`/`unknown` tags, not from guessing as if known.
- **Lean is a starting depth, not a ceiling.** When the stakes rise, graduate to `playbook-engagement-kickoff` and the full toolkit rather than stretching the lean kit past its resolution.
- **Proposed.** First draft from the a1.muse instance (broker-solo × finance as the template cell); confirm the kit and whether it lives as this convention or as a lean-mode section of `playbook-engagement-kickoff` before adoption.

## Tone

Calm, precise, sentence case, British spelling, no exclamation marks, outcome before adjective. (cos.tov.)