psychology

chevp-ai-framework
Lifecycle Gates Guidelines Commands Agents Templates
looks_two Step 2 of 3

Exploration

Think first, see first — then build.

Plan concrete features, prototype them, validate the approach. No production code yet.

Two Sub-modes

Declared in the plan frontmatter as exploration-mode: A | B

explore

Exploration-A — Problem

Understand the problem in motion. Low fidelity (sketches, throwaway scripts, paper, single-screen demos).

Goal: A confirmed framing of the problem — and a retired hypothesis or two.
Output: Low-fidelity prototype, refined hypotheses.
tune

Exploration-B — Solution

Decide between concrete solutions. High enough fidelity to compare — ≥2 candidates side-by-side.

Goal: A chosen approach with documented trade-offs.
Output: Comparable prototypes, ADR for the chosen direction.

Skipping A and going straight to B is allowed only when the hypotheses artifact already records the problem framing as confirmed.

Mandatory Deliverables

Six artifacts — some always required, others depending on sub-mode

1
Feature Plan/Spec A & B

Always — written for features/complex changes, verbal for trivial.

2
ADR usually B

Only for new decisions arising during exploration. Fundamental ADRs belong in Context.

3
UX Prototype (low-fidelity) A

Whenever the problem framing is uncertain.

4
UX Prototype (comparable) B

Mandatory for visual/physical output, ≥2 candidates side-by-side.

5
Challenger Output

Always before G2 — top-3 failure modes, ≥2 alternatives, strongest counter-argument.

6
insights.md

Always at the end of Exploration — records what was learned.

The 6 Decisions Toward G2

Each surfaced via AskUserQuestion

  1. Choose exploration-mode — A (plan-first) or B (prototype-first)
  2. Approve the Feature Plan/Spec
  3. Pick among Challenger alternatives (≥2 options + counter-argument)
  4. Confirm the prototype validation (works / partial / fails → kill)
  5. Confirm Kill Criteria and insights.md are recorded
  6. Pass G2 → Production (approve / reject / revise)

Quality Gate G2

All criteria must be satisfied before Production starts

check_circleFeature plan/spec exists with exploration-mode: A | B declared
check_circleSteps are concrete enough for direct implementation
check_circleScope and non-scope clearly defined
check_circleKill Criteria defined — when do we abandon?
check_circleRisks and alternatives documented
check_circleAcceptance criteria defined
check_circlePrototype exists and is visually confirmed (where applicable)
check_circleinsights.md exists and is non-empty — ≥1 hypothesis confirmed/killed
check_circleChallenger output present: top-3 failure modes, ≥2 alternatives, strongest counter-argument
check_circleEvidence block filled (hypothesis / result / reasoning)
check_circleGatekeeper-G2 verdict recorded
check_circleHuman has approved the plan/spec (evidence-based, not rubber-stamp)
block BLOCKER: Do not proceed to Production until every checkbox is satisfied.

Roles Active in This Step

Five roles — Challenger is mandatory before G2

SDLC

Spec + prototype governance, when to require what.

AI-Plans

Feature plan/spec creation, acceptance criteria.

UX-Tooling

Prototype creation, preview feedback loop.

Software-Architecture

ADR for new decisions, design alternatives.

Challenger — mandatory before G2

Internal sceptic. Produces top-3 failure modes, ≥2 alternatives, strongest counter-argument, product-coherence check. Read more →

arrow_back Step 1: Context Step 3: Production arrow_forward
menu_book

Read the source

02-exploration/README.md open_in_new