tbd

TBD -- 5-Tier JTBD Methodology

A progressive methodology that moves from "why do users need this" to "what should we build" to "did we cover everything." Each tier's output feeds the next.

Tier	Name	Input → Output	What's Unique Here
1	JTBD	Product description → Job statements	Solution-agnostic; bold the motivation
2	Mini-Jobs	JTBDs → Sub-jobs with alternatives (DAG)	25% must have alt paths; a-suffix convention
3	Job Map	Mini-jobs → Cross-cutting patterns + emotional landscape	Ulwick 8 steps × 4 emotional forces (Push/Pull/Anxiety/Habit); output is synthesized observations, not raw matrix
4	Solution	Patterns + mini-jobs → Epics + Stories	Regroup by solution concern, NOT by JTBD; every story carries emotional force tag + trace
5	Traceability	All tiers → Coverage matrices + gap analysis	5 automated checks (5A-5E); gaps framed as opportunities

Announce on activation: "Running TBD analysis — 5-tier JTBD methodology."

Intent Classification

Keywords	Op	Reference	Prereqs
"JTBD", "jobs to be done", "analyze [product]", "start TBD"	discover	references/tier1-jtbd.md	None
"mini-jobs", "decompose", "break down", "alternatives"	decompose	references/tier2-mini-jobs.md	01-jtbd.md
"job map", "Ulwick", "emotional forces", "map the jobs"	map	references/tier3-jobmap.md	02-mini-jobs.md
"solution", "epics", "user stories", "regroup"	solve	references/tier4-solution.md	03-jobmap-observations.md
"traceability", "coverage", "gap analysis", "close the loop"	trace	references/tier5-traceability.md	04-solution.md
"continue", "resume", "next tier"	resume	Auto-detect highest NN-*.md file, load next tier
"full TBD", "run all tiers"	full	All refs sequentially; still pause at checkpoints
"show example", "golden example"	example	Load specific tier from examples/golden-example/

Dispatch Protocol

1. Classify op from user intent. If ambiguous, ask which tier.
2. Check workspace for prerequisite files (ls *-*.md or equivalent).
3. If prereqs missing, offer to run prior tiers first.
4. Load the reference file for the identified tier.
5. Execute the tier procedure.
6. Write output to NN-name.md.
7. Suggest next tier (or present results if Tier 5 complete).

Output Conventions

• Files: 01-jtbd.md, 02-mini-jobs.md, 03-jobmap-observations.md, 04-solution.md, 05-traceability.md
• IDs: Jobs = J1, J2; Mini-jobs = J1.1, J1.2, J1.3a (a = alternative); Epics = E1; Stories = S1.1, S1.2
• Story format: statement + traces-to (mini-job + job map step) + key emotional force + 3 acceptance criteria

Workflow

Tiers are sequential — each tier's input is the previous tier's output. Cannot skip tiers.

2 hard checkpoints (must pause and get user confirmation before proceeding):

1. After Tier 2: "N mini-jobs across M JTBDs. Everything downstream depends on this — review before continuing."
2. After Tier 3: "Here are the cross-cutting patterns. Do these resonate? Missing any?"

Tiers 1, 4, 5 have soft checkpoints (present results, offer adjustments, but don't block).

Resume: Scan workspace for highest-numbered NN-*.md file. Read it to restore context. Load next tier's reference.

Parallelism

• Tier 3: When 8+ mini-jobs, dispatch one subagent per JTBD group. Each maps its mini-jobs through 8 steps + 4 forces, returns structured observations.
• Tier 4: After epic identification, dispatch one subagent per epic to write stories.
• Tier 5: Run 5A/5B/5C/5D as 4 parallel checks. 5E synthesizes sequentially.

Quality Gates

Tier	Pass Criteria
1	Min 5 JTBDs, each has situation + motivation + outcome
2	Every JTBD has 2+ mini-jobs. At least 25% of JTBDs have an alternative
3	Every mini-job mapped to 4+ steps. Every mini-job has 1+ emotional force
4	Zero orphan mini-jobs (every mini-job → 1+ story). Zero orphan stories
5	This IS the quality gate — present coverage % and gap analysis

Golden Example

When calibrating quality for any tier, read the corresponding file from examples/golden-example/. The example is "Team Kanban Board": 8 JTBDs, 29 mini-jobs (8 alternatives), 6 cross-cutting patterns, 11 epics, 48 stories, 61% job map step coverage. Load specific tier examples on demand — not all at once.