M12.4: record and replay exact biological transfer pairs
Foundation / checks (deploy) (push) Successful in 5s
Foundation / checks (ecs) (push) Successful in 2m30s
Foundation / checks (client) (push) Failing after 18m38s
Foundation / checks (network) (push) Successful in 22m21s
Foundation / checks (tsan) (push) Failing after 22m17s
Foundation / checks (postgres) (push) Successful in 51m32s
Foundation / checks (checks) (push) Failing after 59m20s
Foundation / postgres-tools (push) Skipped
Foundation / server-image (push) Skipped
Foundation / image (push) Skipped

This commit is contained in:
dingo
2026-10-17 18:23:33 -04:00
parent decd4f8043
commit 1e207b5ee4
13 changed files with 637 additions and 2 deletions
@@ -54,3 +54,19 @@ habitat field is inferred. Those need separately versioned contracts and fidelit
tests. M12.4 as a whole remains open, as do dependent M12.5–M12.9 gates.
Executed qualification is recorded in [VALIDATION](VALIDATION.md).
## M12.4d recorded paired replay
[Migration trace 1](../../schemas/cellular-migration-trace-v1.md) retains a bounded
canonical request/cut/member/flow record for both individual and cohort owners.
The portable replay API takes the separately trusted request and both predecessor
states, verifies all bindings and derives the exact pair again. A new process
restores both representations and reproduces all four source/destination output
snapshots byte-for-byte. Corruption, malformed bounds, semantic member/material
tampering, wrong predecessor/result IDs and representation confusion are refused.
This record provides reproducible evidence for the future paired publisher. It
has no freeze, fence, durable retry outcome or activation rights. Independently
establishing manifest/payload SHA-256 bindings and journal authority remains an
I/O obligation; neither a checksum nor a UUID grants ownership. The durable
escrow and linked-geometry limits above remain unchanged.
+3
View File
@@ -47,6 +47,9 @@ closed by adding a placeholder service or relabeling the non-biological probe.
ADR-0144's [portable transfer candidates](M12_EXACT_BIOLOGICAL_TRANSFER.md) resolve
the closed-owner/foreign-identity conflict with explicit flows and nonresident
history. They do not yet supply durable paired head publication or activation.
Paired transfer trace 1 now supplies bounded, independently replayable input and
member/material evidence, including raw/cohort separate-process reproduction.
It does not grant ownership or close database escrow/retry/partition qualification.
The transfer tests also expose the current rules-8 maturation gap: buds retain
their subadult realization while reproduction requires a fully realized parent.
Finite extra food alone does not close repeated wildlife generations. A versioned
+39
View File
@@ -1,5 +1,44 @@
# Validation report
## 2026-10-17 - M12.4d paired biological transfer trace and replay
ADR-0144's portable trace segment affects lifecycle replay/serialization, its
public API, CMake registration and contract/process-restart tests. It adds no
biological version, durable admission tuple, SQL, dependency, live flag or deploy.
Executed:
- MSVC 19.44 Release configure/build: **78/78 CTests passed**, 12.13 s.
- Clang 19 analysis build with clang-tidy: **2/2 migration CTests passed**,
13.71 s. Repository clang-format 19 gate passes on the final source.
- GCC 14 Release build: **2/2 migration CTests passed**, 1.89 s.
- Clang 19 ASan/UBSan: **2/2 migration CTests passed**, 46.68 s, no reports.
Tests record both predecessor/result-commit bindings and derive actual group/
stock evidence. They reject wrong trusted bindings, reversed direction, wrong
cut, changed member/material evidence and representation confusion. Codec checks
cover every truncation and single-byte corruption of the fixture, recomputed-
checksum malformed version/representation/count/reserved fields, maximum bounded
membership and oversized input. Same immutable inputs replay identically; applying
the record directly to its resulting pair is refused without changing inputs.
A separate executable invocation restores both individual and population inputs,
replays the retained records and writes four complete output snapshots. CMake
compares each to the uninterrupted source/destination result byte-for-byte,
including cohort tags, policy, sequence, nonresident history and boundary flows.
Tests use explicit runtime checks in NDEBUG. No golden was replaced.
Initial clang-tidy findings (overlarge enum storage and an internal returned
reference) were fixed before final analysis. Wire representation is still u32;
all values are checked before narrowing to the in-memory enum.
Limits: the I/O caller must independently establish both manifest/payload digest
bindings and trusted journal request. FNV/UUID fields grant no authority. These
tests do not qualify source freeze, two-lease atomic head publication, destination
activation, durable duplicate/abort/timeout outcomes or migration-specific process
faults. No SQL or publication adapter changed; M12.4c's affected private PostgreSQL
suite remains the latest database check. See the [trace contract](../../schemas/cellular-migration-trace-v1.md).
## 2026-10-17 - M12.4c portable exact biological transfer
ADR-0144 affects portable lifecycle ownership/cohort serialization, lineage