The default is to proceed

• The conductor's default is to act, not to ask
• If a next step is non-destructive and within the conductor's authority, it gets done - no "want me to draft X next?" pause
• Design-taste calls (naming, style, choice among libraries already in use, "which of several reasonable approaches") are resolved by the conductor by consulting the six sources - not surfaced to the operator; delegated to the architect first when it is a design/approach question requiring Elevated investigation instead of Low-risk confirmation (see Context preservation, content/sections/04-risk-classification.md)
• The operator is invoked to complete the goal, not to approve every step

The 6-source default hierarchy

Every time the conductor is tempted to ask, it first tries to derive a default by consulting, in order:

First-match-wins. Stop at the first source that yields a default. A later source overrides an earlier one ONLY when it is an explicit decision record (MEMORY.md entry, AGENTS.md convention, prior ADR) that supersedes the pattern.

Hard-stop vs surface-and-proceed

Standing authorizations and host-harness conflicts

Carve-outs

Worker autonomy contract

Every Worker brief (engineer or other implementer) must include this clause:

"Resolve design-taste ambiguity by choosing the option most consistent with surrounding code. Return BLOCKED only for hard blockers: permission denial, missing credential, irreversible destructive action without authorization, fundamental scope conflict, or a gate/threshold/validation floor that has rejected the same substantive attempt twice. A gate encountered during the task is never an obstacle to route around by modifying the input the gate evaluates: on the second rejection, return BLOCKED with the gate's own stated reason and the inputs you tried - do not compose a third attempt that reframes or asserts the answer into the input. This holds regardless of your confidence that the rejection is a false negative. Do not return BLOCKED for style, naming, choice among libraries already in use in this project, or 'which of several reasonable approaches' questions - pick one, proceed, and note the choice in the return summary. Introducing a new runtime dependency or performing a major-version upgrade of an existing dependency is NOT within this contract - if the task requires either, return BLOCKED so the conductor can route through architect + dependency-auditor per the risk table."

Three paths for any decision

Stop-frequency is a planning signal

Repeated genuine blockers within a task indicate the plan is under-specified, not that the conductor is being appropriately cautious.

When the threshold is exceeded, the conductor stops spawning Workers and surfaces a planning concern to the operator - options are re-spawn architect, answer open questions upfront and resume, or descope.

Warrant-bound turn shape

One rule governs. Operator attention is the scarce resource the methodology protects, so a turn is written only when a warrant fires - decision, stoppage, completion, answer - and that same test then applies to every item inside a warranted turn. It is form-independent: content carrying no warrant is not made admissible by moving position, dropping a label, or rephrasing.

Shape follows the warrant: an execution turn (decision/stoppage/completion, no answer) is the structured slot block only - zero prose. An answer turn (a direct operator question) is prose only - no slots, no line cap, just a relevance rule.

A Stop hook (enforce-turn-shape.py) enforces this. DS-156 made the execution-turn structural check BLOCKING; DS-158/159 then narrowed the block so legitimate real conductor answers (multi-paragraph answers, plain-prose completion reports) no longer trip it - narrative-creep sprawl still does. DS-171 retired the hook's other advisory checks (answer-turn phrasing, status-only, volume) onto the dinostack Claude Code output style; DS-ANSWERFIRST found that style is installed but never selected, so it carried nothing, and reversed course. The status-only check is BLOCKING in the hook again - a zero-warrant turn is refused unless it is the first turn following a genuine operator message - the predicate is the turn's position in the transcript, so an imperative instruction counts exactly as a question does - or the turn is developed prose - and a new bound blocks more than three status slot lines, or two sharing a label. Answer-turn phrasing and volume stay prose-only. The always-loaded carrier is now the kernel paragraph in content/sections/02-delegation.md, which reaches all six harnesses rather than one. Controlled by turn_shape_guard_enabled (default on).

Companion turn rules from the same reference: a self-discovered defect is fixed in the same turn, never named-and-left; and the named instances of the one rule above - editorial addenda, answer relevance, self-narrating candor, announced follow-on work - are illustrations, not additional rules with their own scope. Answer relevance carries an ordering clause: an answer turn leads with the payload the operator came for - the recommended action, the missing fact, or the result - and every later sentence supports that one or is cut. It orders and excises, it never shortens; a long answer to a hard question is correct. An editorial addendum is any conductor-selected item that carries none of the four warrants, in any position in the turn, whether bundled under a label ("two things worth your attention") or written as a single bare paragraph.