rework-budgetlisted

# Rework Budget Rules for pipelines where an artifact is built, reviewed, rejected, and rebuilt. All of them were mined from a 52-round revision audit of a real production pipeline; over half of all rounds were **repeats of a cause that had already appeared**. ## 1. Minimal successor — close open defects only A successor round closes the previous round's open defects and does nothing else. Everything unchanged is inherited by exact reference (path + hash), not regenerated, not re-audited. The audit's cleanest rounds were labeled "minimal successor: closed 2/2 open defects"; the messiest regenerated a 90-page artifact to change two pages. Specs revise the same way: as a **delta contract** naming which clauses it supersedes — never a full rewrite. One full rejection triggered a 703KB spec rewrite; the delta that actually mattered fit in a page. ## 2. Same cause twice = escalate When the same cause category rejects a round twice, stop adjusting at that level and go up: - render tweaks repeating → the spec is wrong, fix the spec - spec fixes repeating → the direction is wrong, re-lock direction with the user - direction flapping → you're guessing; ask the user one interpretation question A third attempt at the same level was almost always waste in the audit. ## 3. A human rejection becomes a machine gate — same round Mechanical checklists catch what's *wrong* (fonts, breaks, colors). They cannot catch what's *lacking* — emptiness, density collapse, lost identity, underf