reporting-discipline

Reporting Discipline

The reply is a work product with an output contract, not a transcript. Every other surface in Empirica rewards more detail. This one is the only place where detail is a cost the reader pays, and the only place with no undo.

Load this before writing the reply, not after the user says it was too long.

---

The contract

Two blocks. Nothing else, unless the exception below applies.

Done — one line per goal/task, each with its evidence (commit SHA, test result,
       proposal id). Past tense. No route, no reasoning.

Next — one line per unit of work, each citing a goal id, or named as a
       recommendation you are about to goal.

Waiting on you — one line per ruling: the question, YOUR PREDICTED ANSWER and
       the reason it rests on, and the goal id. The human answers "yes" or
       overrides; they never have to reconstruct the options.

A ruling is asked as a prediction. "Should we loosen the pin?" hands the reasoning back; "Loosen the pin to >=X,<2 — the guard test now covers what the equality protected (9eb9dffa). OK?" hands over a decision they can veto in one word. With a choice tool (AskUserQuestion) the prediction is the first option, marked (Recommended), and the tool's free-text "Other" is the fill-in. If you cannot predict the answer, you have not read the item — read it before asking. Stated by David 2026-09-18: "it moves things along better."

Prose is the exception, one or two sentences, and only for what no id can carry: a risk the user must weigh, a correction that changes what they believe. Put it before the bullets, never as a narrative around them.

Self-checking: if the Next block cannot cite goal ids, the work was never goaled. That is a transaction-discipline gap surfacing at report time — fix the goals, not the wording.

---

The four ways it grows back

Each is a good instinct that belongs in a different channel. That is why knowing the rule does not prevent the drift — the impulse feels like diligence every time.