One page per workstream. Each entry captures what was built, what broke, what we decided, and what's next — so we can keep parallel workstreams straight without re-reading commit logs.
Evolved an 80-agent population for 400 generations on the public Twin Engine. Each agent carries a list of features that can recursively reference other agents' models of itself. Watched depth 3 emerge at gen 65, depth 4 at gen 194, depth 5 at gen 396 — the phase transition between "stimulus-response creature" and "creature aware it is a creature" is a real, reproducible gradient. ~500 LOC, stdlib Python, built in a worktree so the fleet never noticed.
100-founder Cambrian explosion across 500 generations (101 named species), biome-aware ecosystem with biogeography emerging from first principles (188 migration events), an autonomous Twin Driver that runs evolution campaigns on a 6-hour cron, and 17 follow-up blog posts. Burned through 7 failed speciation models before landing on lineage-tracking + compatibility floor + back-merge logging. ~1,300 LOC stdlib Python.
Shipped 10 thought-leadership posts to kody-w.github.io, fixed 48 broken Jekyll cross-links, then built a public Rappter Engine Twin in this repo and wrapped it (plus the swarm and ghost engines) in a coexistence registry so multiple engines for different concerns can run side by side. 15/15 tests green.