Files
aevolution/schemas/cellular-migration-trace-v1.md
T
dingo 5e168d67c3
Foundation / checks (ecs) (push) Successful in 2m26s
Foundation / checks (deploy) (push) Successful in 4m12s
Foundation / checks (network) (push) Successful in 20m57s
Foundation / checks (client) (push) Successful in 22m9s
Foundation / checks (tsan) (push) Successful in 28m49s
Foundation / checks (postgres) (push) Successful in 1h5m2s
Foundation / checks (checks) (push) Successful in 1h17m42s
Foundation / postgres-tools (push) Successful in 1m10s
Foundation / server-image (push) Successful in 3m51s
Foundation / image (push) Successful in 4m54s
feat(lifecycle): qualify portable maturing wildlife transfers
2026-10-19 09:09:15 -04:00

70 lines
3.9 KiB
Markdown

# Exact paired biological transfer trace 1
Follow-on [trace 2](cellular-migration-trace-v2.md) supports portable rules-9
transfers. Trace 1 remains rules 8 with the exact layout and bytes below.
Authority: ADR-0144. Portable replay evidence only; no durable manifest admission,
ownership receipt or activation authority. This is separate from the M10 probe,
normal cellular tick traces and population batch 1. It is not a migration route.
All integers are little-endian without padding. Maximum size is 2,312 bytes.
| Offset | Bytes | Field |
|---:|---:|---|
| 0 | 8 | ASCII magic `AEVMIG01` |
| 8 | 4 | Version, exactly 1 |
| 12 | 4 | Representation: 1 individual owners, 2 population owners; reject all other u32 values before narrowing |
| 16 | 8 | World |
| 24 | 8 | Common frame |
| 32 | 8 | Common completed biological tick, positive; no tick advances |
| 40 | 8 | Whole wildlife genome group |
| 48 | 16 | Stable workflow request UUID, nonzero |
| 64 | 72 | Source endpoint |
| 136 | 72 | Destination endpoint |
| 208 | 40 | Exact transferred inventory: energy, C, N, P, other; five u64s |
| 248 | 4 | Member count, 1..256 |
| 252 | 4 | Reserved zero |
| 256 | 8 each | Sorted unique wildlife life IDs |
| final | 8 | FNV-1a64 over all preceding bytes |
Each endpoint is region u64, predecessor UUID (16), predecessor manifest SHA-256
(32), resulting commit UUID (16). Regions are distinct positive bud-capable IDs.
Both predecessor digests are nonzero; the four commit/predecessor UUIDs are
nonzero and pairwise distinct. Member IDs have valid positive birth-origin region
and sequence within the bud namespace. Transferred inventory is nonzero.
Require exact total length `256 + 8*count + 8`, all header/member bounds before
allocation and complete canonical re-encoding. The two representation variants
are distinct even when materialized biology is equal. Individual inputs must be
owner 7/rules 8; population inputs must contain those owners in envelope 2.
Preparation derives members and material from the actual paired migration
candidate. Replay takes an independently trusted request plus both complete
predecessor states. It checks request/endpoint/result-commit bindings, world,
frame, tick and representation, recomputes exact transfer candidates and requires
the recorded member set and every inventory channel to match. Replaying again
from the same immutable inputs gives the same results; applying it to the output
pair is refused. No controller, random sampler, biological tick, movement,
conversion or identity allocator runs as part of this trace.
The I/O adapter must establish that trusted request against both retained
predecessor manifests and digest-checked payloads before invoking replay. Taking
the trace's own request as a live ownership grant is invalid. FNV only detects
accidental corruption; a future durable publisher must bind this trace's SHA-256,
both result payloads and both fenced heads in one transaction, retaining the
source/destination histories and transfer journal. These UUIDs and SHA fields
alone do not establish current authority or transaction outcome.
The separate-process test records raw and cohort inputs/expected output pairs,
terminates, restores in a fresh process, replays and compares all four complete
output snapshots. This qualifies portable continuation, not database atomicity,
source freeze, ghost staging, abort, duplicate-delivery settlement or crash-safe
activation. No schema 45 publisher admits this trace or the new owner tuple.
M12.4h/ADR-0147/schema 47 subsequently retains trace 1 as durable source-freeze
preparation evidence. Both predecessor cuts are SHA-256 bound and replayed from
the source's retained capsule; SQL also checks canonical shape and FNV. The
source still owns every cell and the destination owns no ghost. Explicit abort
releases the freeze without changing either head. This is not result-manifest
admission or a paired ownership receipt. See [the freeze contract](ecology/cellular-migration-freeze-v1.md).