Files
aevolution/docs/engineering/M12_EXACT_BIOLOGICAL_TRANSFER.md
T
dingo decd4f8043
Foundation / checks (deploy) (push) Successful in 5s
Foundation / checks (ecs) (push) Successful in 2m18s
Foundation / postgres-tools (push) Canceled after 0s
Foundation / server-image (push) Canceled after 0s
Foundation / image (push) Canceled after 0s
Foundation / checks (checks) (push) Canceled after 5m33s
Foundation / checks (postgres) (push) Canceled after 5m29s
Foundation / checks (tsan) (push) Canceled after 5m28s
Foundation / checks (client) (push) Canceled after 5m37s
Foundation / checks (network) (push) Canceled after 5m39s
M12.4: preserve exact biology across ownership candidates
2026-10-17 18:17:59 -04:00

3.6 KiB
Raw Blame History

M12.4c exact biological transfer candidates

ADR-0144 resolves the closed-owner/accounting conflict before implementing biological ownership transfer. The portable lifecycle layer now prepares a detached source/destination pair at an exact compatible cut. This segment adds owner format 7 and population envelope 2, without a live migration flag, database schema, durable manifest bundle, dependency or deployment.

The pair transfers actual zero-account wildlife, including every member of the selected genome group. Opening stocks remain immutable; explicit cumulative incoming/outgoing flows explain each owner's current stock. Bodies, reserves, age, growth/damage, ancestry, identity, coordinates and cohort tags are exact. Departures retain nonresident archive facts instead of recording death. Complete genome ancestry and organism revisions merge without erasing destination history, rewinding revisions or resurrecting dead organisms. Archive known-alive counts include nonresidents and must never be used as an owned census/extinction proof.

The primitive refuses incompatible cuts/worlds/frames/geometry, vulnerable controlled cells, identity conflicts, missing groups, placement conflicts, capacity exhaustion, archive conflicts and counter overflow. Both detached candidates must validate before success; failures leave the two originals exact. Local offspring use the destination's identity namespace, with sorted insertion even below imported IDs. Existing regional twin randomness retains its meaning; the migration cut does not promise identical future environments or outcomes.

The tests use real funded founders and real budding, not fabricated wildlife or the non-biological probe. They exercise repeated no-tick return trips, exact per-member equality and per-channel combined closure, canonical restore, imported death/decay, lower-namespace local reproduction among imported residents, whole-group capacity refusal, body conflicts, malformed/truncated ledgers, overflowing cumulative counters and cohort tag/policy/sequence preservation. Lineage tests include revision conflicts and a maximum-depth 4,096-genome ancestry using an iterative traversal.

The first reproduction fixture exposed an existing rules-8 limitation: a bud is created below full realization, growth is retained, and fund_bud requires full realization. Additional food does not mature that bud under the current kernel. This transfer segment does not invent growth or alter reproduction rules. Its ordering regression uses a real fully grown local founder reproducing alongside imported higher-namespace residents. Self-sustaining wildlife generations still need an explicit versioned maturation contract and ecological qualification.

This is not the durable escrow gate. No prepared candidate may become an active destination owner before a future fenced transaction commits both heads and immutable transfer evidence. The source remains the sole owner until that commit; a staged ghost owns nothing. Durable preparation/freeze, two-owner fencing, ambiguous outcomes, abort/restart, partition/duplicate delivery, process kills and stock-in-transit audits remain required. Existing profile-4/5 publishers do not admit this new owner tuple, and retained trace 1 explicitly refuses envelope 2.

Numeric coordinates are region/frame-local and current movement is clamped, without wrap. No atlas, route, edge crossing, cross-region contact or linked habitat field is inferred. Those need separately versioned contracts and fidelity tests. M12.4 as a whole remains open, as do dependent M12.5–M12.9 gates.

Executed qualification is recorded in VALIDATION.