ADR-0109 — Output economy: a readable, economical agent-output convention (a floor, not a hook)

Context

Agent output drifts verbose: persuasive prose, restated context, ceremony. Verbosity costs reader attention and tokens, and — the non-obvious part — long explanatory prose can make a reader trust the output more, not check it more ([[OVERRELIANCE-REVIEW]]: explanations often persuade rather than help evaluation; more detail can worsen over-reliance). Suspec already implies the cure piecemeal — the vocabulary tiers (ADR-0057), the ~100-line always-on AGENTS context budget, and the format-tax evidence ([[FORMATFREE]]: format restriction during reasoning degrades it; structure the frame, reason free-form, then emit the structured artifact). What is missing is a single, stated output convention an agent (and a worker definition) can follow.

A maximal version of this exists as a third-party tool (the "caveman" output-compression plugin, hook- enforced with intensity levels). Suspec deliberately does not copy that mechanism: a hook that rewrites output is enforcement, and the framework is reconcile-only / honest-levels — it does not enforce style, and an over-terse default risks the misread the caveman tool itself carves out (it drops to normal prose for safety/confirmations/multi-step sequences). Suspec wants the economy, set at a gentle floor, with the dial left optional.

Decision

A convention floor, plus an optional dial — never a hook.

1. Output convention (floor). Agent-produced output (run summaries, reviews, findings, decisions, chat) SHOULD be: evidence-first (the finding/result + its evidence before any prose); structure over prose (tables/lists where they carry the same signal in less space); signal-dense (no filler, no restating the prompt, no persuasion); free-form reasoning, structured emission (reason in whatever form thinks best, then emit the lean artifact — [[FORMATFREE]]); and justification sized to make checking cheap, not to convince ([[OVERRELIANCE-REVIEW]]). Level: convention.
2. Clarity outranks brevity. The floor never compresses at the cost of correctness or safety: security notes, irreversible-action confirmations, and multi-step sequences where order matters stay in full, unambiguous prose. Brevity is the default, not a mandate.
3. An optional concision skill is the dial (suspec-skills) for teams/agents that want stronger economy. It is opt-in conditioning, not a Suspec requirement and not a runtime hook.

Consequences

• A stated floor every worker definition can point at, consistent with the vocab tiers and the context budget — readability and token economy without a new mechanism.
• Not enforced. No hook, no checks.yaml rule, no contract change. A reconcile pass MAY later note egregious verbosity as an advisory, but style is not gated (ADR-0063; ADR-0077 D8).
• The convention shapes the consolidated review skill's output (the review packet is the densest, most evidence-first artifact Suspec produces).

Affected obligations / constraints

• Refines: ADR-0057 (vocabulary tiers — adds an output- economy floor). Grounded by: [[FORMATFREE]], [[OVERRELIANCE-REVIEW]].
• Does NOT change: the artifact formats, the verdict model, or the checks contract; introduces no enforcement (ADR-0063) and no provider-specific mechanism (provider-neutral, principles).