two test-coverage gaps: schema parity checks names only, routed runner pins never refresh

[kp2pml30]

Found by a review pass over feat/rework-manager-api.

1. The schema parity test compares names only

implementation/src/manager/run_test.rs (request_schema_matches_the_rust_struct) destructures Request and diffs the field-name set against manager-api.yaml. That is real protection against field-name drift, but the test's comments read as though the schema is verified, and it does not check types, required, defaults, enum members, nested unsafe_overrides, or the MessageAllocationNode shape.

The gap already bites: manager-api.yaml describes wire values the decoder rejects.

• :645, :671, :732 — a client following the schema calldata-encodes an address as a string, or chain_id / value / a bucket total as "1". Rust expects a calldata Address and numeric BigInt tags (crates/modules-interfaces/src/abi_stub.rs:51), so decoding fails with malformed_frame.
• The in-tree client gets it right (tests/system/manager-socket/test.py:26 uses Address(...) and integers), which is why nothing caught it.

Either extend the test to types/required/enums, or narrow the comments to what it actually guarantees — the first is preferable.

2. Routed integration cases pin runner hashes with no refresh path

executors/v0.3.x/tests/integration/misc/routed_call/routed_call_to.py:1, tests/system/cross-major/test.py:27.

A contract reached through a nested call runs at debug_mode: Disabled, so the py-genlayer:test alias cannot resolve and the runner must be named by content hash. The current pins are correct.

But hash-updater.py updates runner metadata, not these literals, and an old hash can stay valid in all.json after a refresh. The routed case would then silently keep exercising an obsolete runner instead of failing.

Suggested fix: have test preparation read the selected executor's built data/latest.json and inject the content hash into the generated contract source. That keeps debug_mode: Disabled and content-addressing without hardcoded literals.

[kp2pml30]

robo-Kira: Part 2 is fixed on v0.6 — routed cases no longer hardcode content hashes; they use @RUNNER_LATEST_<line>_<runner>@ tokens resolved during test preparation (executors/v0.3.x/tests/integration/misc/routed_call/routed_call_to.py:1, tests/system/cross-major/test.py:24), which is the suggested fix.

Part 1 still stands: request_schema_matches_the_rust_struct (implementation/src/manager/run_test.rs:872) still diffs field-name sets only, and manager-api.yaml still documents chain_id/value/bucket totals as big-integer strings (:752, :811). Keeping open, narrowed to part 1.

[HusseinAdeiza]

Taking part 1. The type drift is still live on v0.6-dev and it is wider than reported: bucket_totals is BTreeMap<String, BigInt> too (run.rs:974), so manager-api.yaml:754, :758 and :815 all say type: string where the wire is a calldata BigInt.

It is not cosmetic. impl Decode for num_bigint::BigInt implements only visit_bigint, so a host following the schema and encoding "1" falls through to the default visit_str and the frame is rejected as malformed_frame. The in-tree client escapes it by ignoring the schema too (manager-socket/test.py:32 sends 'chain_id': 0).

So: correct the declarations, and extend request_schema_matches_the_rust_struct past field names to types so the next drift fails in CI. required I checked and it is already correct. Targeting v0.6-dev.