The version archive
Liberty Bee ships as fully re-run baselines, never as edits to a live one. When a new baseline takes over the front page, the one it replaces is preserved here — its numbers, the issues it surfaced, and how to reproduce it — exactly as it shipped.
The discount grid — how far could it push?
Rows: up-front discount depth. Cells: % of runs surviving 20 years. The finding: past 40% off, no funding level tested guaranteed survival — at half-off-forever, even $11M lost 28% of its runs. Measured at V2’s founder-subsidy pay basis; at real wages every price runs higher. 10% row n=500, 25% n=150, others n=25 per cell.
The honesty release: the survival threshold moved $4.5M → $5M, community-kept at $11M moved $13.5M → $11.9M, scarcity's share of retention 44% → 48% — every headline re-measured and published even where it got worse. The discount grid ran at nine depths (no guaranteed funding level past 40% off). Corpus assets remain on the v2.0.0 release (17 checksummed corpus backups; the seed universe was retired 2026-08-23 when real-address data left the project). Superseded by V3’s clean-provenance rebuild and three-basis record.
20-year survival by starting capital. The frozen 1.0 baseline drew a clean threshold at $4.5M — below it, an organization usually runs out of money before year twenty; at or above it, all 800 runs survived. The living farm kept the same engine running to 5,903 runs and drew the fuller picture.
Left running, the farm reached 5,903 runs and 1,136 failures — nearly all of them the deliberately under-capitalized rungs, where running out of money is the finding, not a fault. At the funded rungs it reads 99.6%, not the frozen 100%: a few organizations failed in their opening weeks for purely institutional reasons — no evictions, no rent increases — the acquisition-gate defect now logged as KD-042.
It also sampled the deep-discount-25 scenario — a standing 25% below-market rule — that the frozen set only referenced. There the threshold sits a full rung higher, near $5M: the measured price of deeper affordability, and a dimension the shipped corpus couldn't show on its own.
Known issues. Every defect the 1.0 run surfaced stays in the open defect ledger — the lead-deleading gap, the acquisition gate, plan-dependent determinism — each with how to reproduce it and how the next baseline resolves it. Reproduce: every run here regenerates from its seed on the published 0.5.0 engine and Gold backup in the repository.
Three baselines now stand — 1.0, V2, and the V3 front page — each preserved exactly as shipped. Development is mothballed after V3; if a future baseline is ever cut, this archive gains another entry and nothing here moves.