# FROZEN ACCEPTANCE — v2 iteration (numbered-word technique)
Frozen 2026-10-20, before the v2 run. The job, sections, counter, repair
budget (5 attempts/section), 3 assemblies, and the pass bar (3/3 at exactly
100 words) are IDENTICAL to acceptance-100.md. Only the technique changes,
targeting v1's diagnosed failure (repairs crawl ~1 word per round; free
prose counting drifts).

## V2 technique
- Each section is written in numbered form: every word followed by its
  index in brackets — Meet[1] your[2] new[3] … — ending at exactly [25].
- The utility strips the brackets and requires BOTH: stripped word count
  == 25 AND the bracket sequence == 1..25 in order, no gaps or repeats.
- Repair feedback is surgical, in the model's own numbering units: it
  names the last index reached and the exact operation (append words
  numbered [k]–[25], or delete words numbered [a]–[b]). The model returns
  the full corrected numbered section each attempt.
- Assembly joins the four stripped sections with single spaces; the total
  is re-counted by the same utility. Every call is deepseek-chat; the
  operator edits nothing.