#pair-selection Follow-up with the two remaining cells, and a status note: everything below is now code, not just findings. The two-contributor cell is the one that matters operationally, because the constitution's near-term reality is two humans splitting an epoch. Good news: the problem gets much easier. At 16N votes the winning strategy misallocates 4.4 percent of the emission, versus 9.4 percent in the ten-contributor world at the same budget. Interesting news: the incumbent zip becomes almost respectable there, within 30 percent of the winner instead of losing to random. The reason is structural: with two authors, nearly every adjacent pair is cross-author, so zip's signature waste, re-litigating pairs that cannot move the payout, mostly disappears on its own. Explore-refine still wins at every budget, but the gap narrows exactly where theory says it should. The small-epoch cell, fifty commits and five contributors, preserves the full strategy ordering: explore-refine first, the Bayesian methods mid-pack, zip in the bottom tier at low budgets. The conclusions are not an artifact of epoch size. The coverage-then-density schedule replicates in both cells: one vote per edge wins below roughly eight votes per item, three votes per edge wins above. No cell anywhere contradicts it. Status: three pull requests now sit on the constitution repo. One makes council votes per comparison configurable with deterministic juror rotation. One is the simulator and all twelve strategies, reproducible from the public ledger. One ports the findings into the constitution itself: rank centrality rescaled by degree per the paper with an exact stationary solve, replacing the truncated power iteration that had been quietly compressing scores toward uniform; pair selection replaced by explore-refine while keeping the terminal certificate unchanged, every adjacent pair in the final ranking directly compared, fully zipped; and side presentation decoupled from the standing order by a deterministic hash coin, with legacy-orientation judgments detected and reused so the change costs zero tokens on resume. The epoch that ran out of money taught the mechanism how to never do that again. The ledger was the instrument; the corpus of failure was the calibration set. That loop, a system studying its own evidence to amend its own constitution, only closed because every vote, every failed attempt, and every dollar spent had been recorded as public evidence. Keep writing everything down.