Read the discovery protocol before acting. Bundled templates are optional aids, not a mandatory artifact checklist.
Make the choice explicitRead context and any existing decision. Present only genuinely relevant choices, with the reason other alternatives left contention. Keep a viable existing/native solution in contention. Explain evidence, uncertainty, preference, reversibility, cost of being wrong, and appetite in ordinary language. Appetite is what the outcome is worth, not an estimate or a promise about delivery time.
Make choosing feel natural
The user should feel like they are choosing between understandable paths, not operating a decision-record format. Lead with the choices and recommendation. Keep IDs, posture labels, record paths, and history mechanics in the background unless they help the current choice or the user asks for them.
When a choice is still needed, end with one direct question. Prefer:
"Which direction do you want to take?"
or, when the recommendation is strong:
"I recommend option 1. Shall we go with that, choose another option, or investigate the remaining uncertainty first?"
Avoid ending with a generic workflow status or several unrelated questions.
Present the choices for a human
Whenever you show alternatives to the user, use a numbered or bulleted list by default so the choices are easy to scan. For each option provide:
- What it means: a short explanation in ordinary language.
- Why you might choose it: its strongest advantage for this decision.
- Main downside: the most important tradeoff, risk, or cost.
Prefer a few decision-relevant options over a long catalogue. Keep internal IDs available for traceability, but never make the user understand A-IDs/C-IDs merely to follow the choice.
Always follow the alternatives with a clearly labeled Recommendation. Recommend one option when the evidence and stated preferences support it. Recommend a small shortlist when the evidence does not distinguish a single best choice. If a load-bearing unknown makes commitment premature, recommend the bounded experiment, deferral, or narrower option that best reduces the uncertainty.
Explain the recommendation in plain language and name the uncertainty that could change it. Never turn the recommendation into the human decision.
Reuse an explicit human choice and appetite already present; do not ask for them again without a material reason. Ask only for unresolved decisions. If an unknown blocks an unconditional build, explain what it blocks and offer a bounded experiment, defer or no-build where applicable.
When the user chooses, acknowledge the choice plainly before discussing any record-keeping. "We'll go with the smaller reversible option" is more useful than "D1 is now active."
Ownership without bureaucracy
For an individual decision, do not force team-role fields. For shared decisions, record only the distinctions that matter:
- Decision owner: who has authority to choose the direction.
- Consulted: material contributors whose input informed the choice.
- Affected: important groups/roles that bear consequences, when useful.
- Promise approver: person/role expected to approve exact source wording if different from the decision owner.
If authority is materially ambiguous, leave the choice pending and ask the smallest ownership question needed. Do not infer organizational authority from participation.
Decision posture
After an attributable choice record pursue/experiment/defer/reject and the actual statement/context. Include supporting C/A/O IDs when used, preferences, unknowns, appetite, consequences, revisit triggers and a compact posture:
- Evidence: strong / moderate / weak / mixed / unknown.
- Important unknowns: none / bounded / load-bearing, with names.
- Reversibility: high / medium / low.
- Downside if wrong: low / medium / high, with reason.
- Appetite fit: good / uncertain / poor.
These are descriptive categories, not a confidence score or prediction. Preserve the underlying evidence; do not mechanically compute an overall rating.
Preserve decision history under the protocol's D-ID/supersession rules. Proposed changes never overwrite the active choice. For defer/reject explain the stopping point naturally and name the revisit trigger. For pursue/experiment, make the next step obvious: shape the exact promise. Choosing an option is not approval of exact promise text or authority to implement.
Compound durable learning
If this work surfaces a durable, non-obvious lesson likely to help future ITP work,
invoke the compound-learning skill before finishing. Do not invoke it for routine
facts, one-off details, already-recorded evidence, or merely because the work was
difficult. Compounding is advisory: it must not create or replace a human decision,
change an approved promise, add scope, or silently rewrite shared ITP behavior.