Register timeline forecast entry 20 (supersedes 19)

[MaxGhenis]

Summary

Append-only supersession per docs/forecasts/README.md: entry 19 flips to superseded (dates untouched); entry 20 registers p50 2026-09-24 / p80 2026-10-29 on the identical claim and criterion. Rule-5 material evidence — measured, not modeled:

1. Both arms cleared their reliability bars with production-rate measurements. Missing-reason dual-review pipeline: ranks 1–340 reconciled at 98–100% exact agreement (zero unresolvable), frontier at 640/990 after seven ~2–3h three-lane waves. Purpose arm: the rubric-v2 cure measured 96.67% blinded dual-pass agreement with cured q_p 107.66/lane-hour (~6× the alert-era rate priced into entry 19), lifting the hold; production adjudication has begun.
2. At measured rates the adjudication stages compress from months to weeks: the remaining ~49k missing-reason rows and the 21,102-prompt purpose residue both project to 2–3 calendar weeks of continuous waves.
3. Ceremony latency dropped: CI pytest is now 4-way sharded per Python version (Shard CI pytest 4 ways per Python version (fan-in keeps protected contexts) #402), compounding across the A20 ceremony, revision-22 activation, tier-2 R04/R05, R06, and the reconstruction-to-publication chain.
4. Structure unchanged; the p80 buffers the observed ≥1-unplanned-ceremony-per-stage rate; dates conditional on lawful total settlement under the enacted kill criteria.

Ledger test 5 passed; black + ruff clean; indent=1 preserved.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
social-security-model	Ready	Preview	Aug 17, 2026 11:49am