UK compute campaign, run 2: reckoner full-population compute (24 expressible measures × 3 forecast years)

The reckoner lane (#66, approved) triaged all 75 HMRC ready-reckoner families and built the offline pipeline; nothing has computed. This is the execution pass — the densest mode-2 battery per unit of sim time (one-parameter deltas, 24 expressible × 3 years = 72 sims plus baselines).

Steps

• --dry-run first on the managed machine: the parameter-path resolution gate settles the paths_verified: false rows (NICs, child benefit, rate legs) — fix or demote to partial per what the installed pe-uk tree actually carries.
• Full compute on the certified bundle, per-run artifacts + staged JSONL under the campaign ingest contract (the #56-established shape, now the merged DB's shape).
• Ingest + comparison artifact: measure × year, HMRC vs PE, signed ratio, bin, basis axis (pre/post-behavioural where HMRC splits it) — benchmark_class = different_model, no summary score.
• Feed #59: the rate-row pre/post-behavioural splits computed here are the behavioural-axis sizings the divergence registry carries as unsized.

Sequencing: after #66 merges; can share the sim session with #76 (same bundle, same lock).