Slice-plan refinement

Refine an existing executable plan in place. Do not create another plan, implement product code, or change its selected story or bounded correction outcome.

Require a refinable plan

Require an existing executable plan and this project's context required by dough-slice-planning, especially any supplied target, hard limit, exceptions, overrun policy, and the plan lifecycle.

Identify whether the plan's source is a selected feature story or a bounded retrospective correction under planning scope and lifecycle. For a complete correction, use the plan's outcome, findings, scope, proof, decisions, and slices directly; do not require or create a seed. Name whichever required correction field is missing and stop. A complete plan permits refinement, not execution; preserve the invoking instruction's authorization boundary.

• If no plan exists, use dough-slice-planning.
• If a plan is marked as awaiting story refinement after resplitting, use dough-story-refinement on its mapped story and realign the plan before treating it as refinable or executable.
• If the selected story's goal, scope, or examples must change, use dough-story-refinement.
• If a correction's beneficiary, bounded outcome, scope, or proof must change, stop through the shared execution-decision handoff; do not turn it into a feature story. A disputed product constraint uses the plan-conflict handoff below and remains human-owned.
• If the parent problem, candidate selection, or sibling ordering must change, use dough-story-decomposition.
• After the assessment below, if the cumulative design is supported and all remaining slices are cohesive and single-proof-loop, meet any supplied target, and have no unexplained path beyond a supplied hard limit, report ready for direct execution; no further refinement is required. Execution still requires separate authorization from the invoking workflow.

Refine the plan

Read the plan and only the code and tests needed to judge execution boundaries. Read and apply:

• slice decomposition, including its cumulative design assessment, sizing, and escalation rules; and
• active-plan refinement, including executable proof ownership.

When refinement reconsiders the plan's selected existing solution or the architectural direction supporting it, reapply architectural thinking, including its authoritative PFE handoff. Otherwise carry the decision forward; do not repeat PFE or create a new direction topic merely because slices are being refined.

Preserve completed slices, applicable evidence, and the selected story's goal and scope or the correction plan's bounded outcome and scope.

Assess the cumulative design, including how remaining examples build on completed slices, then classify each remaining slice:

Result	Decision
Ready	One Behavior/Structure gate, one proof loop, a cohesive path consistent with the cumulative model, and a plausible hypothesis under any supplied target
Refine	Same story or bounded correction, but the slice has multiple beats, unsupported special-case design, low confidence, or a supplied target or hard-limit concern
Escalate	Learning requires review of the source outcome, selected story, or parent problem

A suspected accidental contractual restriction uses the shared plan-conflict handoff before a conflicting plan edit; classify it as Escalate, not an automatically correctable special case. Route other Escalate findings through the input gate. Refine every remaining Refine slice.

After an overrun, confirm attempt-owned work is safely parked or reverted before editing the plan. Stop for human judgment when ownership is unclear. Do not commit, push, implement, or verify product behavior unless the invoking workflow separately authorizes it.

Count the resulting plan's slices, including completed slices but excluding obsolete replaced slices. If the count is greater than 15, report story resplit recommended: <count> slices; use dough-resplit-story and link to dough-resplit-story. This is a recommendation, not an automatic invocation or a new execution-readiness gate. Do not resplit the story or change its backlog placement during slice-plan refinement.

Report the plan path, replaced slices, resulting slice count, sizing exceptions, any resplit recommendation, and whether execution can resume. End with:

## SLICE PLAN REFINED