bino-requirements

Requirements → the REPORT BRIEF

This runs on the main thread, because only the main thread can ask the human (subagents can't). Your job: turn a vague goal into a precise brief written in bino/IBCS vocabulary, so the downstream bino-data and bino-author subagents act without re-interpreting prose.

Elicit (ask in IBCS terms)

Use AskUserQuestion. Don't ask everything at once — propose sensible defaults from the goal and the data, and ask the human to confirm or correct. Cover:

• Audience — who reads this, and what decision do they make from it?
• Primary message — the one thing the report must say (IBCS: every report has a message).
• Questions — the specific questions the report answers.
• Scenarios + period — which scenarios are in play (ac/pp/fc/pl) and over what period.
• Variance type — absolute (d_), relative (dr_), or both; and which sign is favorable (a business decision — never assume it for costs vs revenue).
• Granularity — the grain (e.g. by month, by region, by product).
• Visualization intent — table of numbers, a time trend, a structural breakdown, narrative — or a mix. (Final component choice is bino-author's, guided by bino-ibcs; capture the intent.)
• Acceptance criteria — how the human will know it's right (e.g. "total revenue ties to €4.2M", "plan column present for every month"). These drive the validation spot-checks later.

The REPORT BRIEF contract

Write it to .bino/agent/brief.json (a JSON object — machine-readable; downstream agents Read it):

{
  "id": "<short run id>",
  "source_goal": "<the human's original goal, verbatim>",
  "audience": "...",
  "primary_message": "...",
  "questions": ["..."],
  "scenario_setup": { "scenarios": ["ac1", "pl1", "pp1"], "period": "FY2025 by month" },
  "variance_type": "absolute | relative | both",
  "granularity": "...",
  "visualization_intent": "...",
  "acceptance_criteria": ["..."],
  "assumptions": ["<every default you filled in for the human>"],
  "open_questions": ["<anything still unresolved>"]
}

Rules

• Write in bino/IBCS vocabulary, not loose prose — the brief is a contract, not a chat log.
• Record every assumption you made on the human's behalf in assumptions[], so the gate review surfaces them.
• Don't fabricate certainty. If something is genuinely open, leave it in open_questions[] and flag it at the brief-confirmation gate.
• The brief is confirmed at a gate before any data work begins (see bino-orchestration).