[MSD21091969]

Refs #77
Stacked on #82 (d57a341): merge after #82. moos-kernel merges with merge commits, so this diff narrows by itself. If #82 is squashed or rebased instead, rebase this onto master first.

• The unit test failed 4 of 6 subtests without the fix and passed 6 of 6 with it.
• go build ./..., go vet ./... and go test -count=1 ./... all pass, and gofmt -l . prints nothing.
• Integration test with MOOS_INTEGRATION=1:
  • with MOOS_ONTOLOGY_PATH set to mtdc-lab's operad/moos-mtdc-2.2.0.json, the stratum test passes
  • on ffs0's 4.0.7, it fails and lists exactly the 5 types (event_notice, governance_proposal, message_packet, tool_call, tool_result), as expected.
• Replaying the logs (Z440 kernel logs and scratch log): 0 LINKs' verdicts change with the fix.

[Copilot]

🟡 Changes recommended

The new integration assertion knowingly fails the repository’s real-ontology CI job until the sibling ffs0 data is updated.

1 open finding

• Update ffs0 ontology before enabling the integration assertion · New

What changed in this PR

Updates LINK validation so malformed strata skip only M5 direction checks while type-list rules still run.

Changes:

• Continues stacked wildcard/type-list enforcement from #82.
• Enforces type constraints despite unparseable strata.
• Adds unit and integration coverage plus formatting cleanup.

File	Description
internal/​transport/​server_port_colors_test.go	Updates ontology fixture type lists.
internal/​transport/​auth_test.go	Formatting only.
internal/​reactive/​predicate_test.go	Formatting only.
internal/​reactive/​engine_test.go	Formatting only.
internal/​operad/​wildcard_acceptance_t342_test.go	Tests wildcard and list validation.
internal/​operad/​validate.go	Separates M5 parsing from type validation.
internal/​operad/​validate_link_pair_test.go	Tests malformed-stratum behavior.
internal/​operad/​registry.go	Clarifies pair type-list semantics.
internal/​operad/​ports_generate_t342_test.go	Updates wildcard fixtures.
internal/​operad/​occupancy.go	Formatting only.
internal/​operad/​occupancy_test.go	Formatting only.
internal/​operad/​loader.go	Validates declared type lists.
internal/​operad/​loader_test.go	Updates loader fixture.
internal/​operad/​loader_color_source_t342_test.go	Updates color fixtures.
internal/​operad/​loader_closure_t342_test.go	Updates closure fixture.
internal/​operad/​integration_ontology_test.go	Checks all ontology strata.
internal/​operad/​color_equality_t342_test.go	Adds wildcard lists to generated fixtures.
internal/​mcp/​protocol.go	Formatting only.
internal/​kernel/​sweep.go	Comment formatting only.
internal/​kernel/​liveness_test.go	Formatting only.
internal/​hdc/​fiber.go	Removes trailing whitespace.
internal/​hdc/​crosswalk.go	Removes trailing whitespace.
internal/​graph/​state.go	Comment formatting only.
internal/​graph/​node.go	Formatting only.
internal/​fold/​errors.go	Formatting only.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[MSD21091969]

t342 · Z440 seat · review of #83 (G5, #77 kernel half; delta d57a341..c38a8c2). Verdict: correct and ready to merge after #82.

The review covered the spec, real data, and code with mutation tests. Every finding was re-checked by a skeptic. All copies and caches were on C: and have been deleted.

Verified

• The fix is operad: ValidateStrataLink returns early on an unparseable stratum and skips the type-list rules (rules 2 and 3) #77's diff in behaviour.
  • Both strata are parsed without an early return. Rule 1 (M5) runs only when both parse. Rules 2 and 3, with operad: * in type lists matches any type (Refs #79) #82's typeAllowed, always run.
  • An absent stratum and "S0-S2" are handled alike, and error messages are unchanged.
  • The rule-1 doc comment is amended.
• Both tests are byte-identical to the issue's text (diffed).
• The 4/6 → 6/6 check holds. With only the validate.go hunk reverted, TestValidateStrataLink_UnparseableStratumStillTypeChecked fails exactly 4 of 6: two pair: cases and two wf: cases. The wildcard and S4 controls pass. At the head, 6 of 6 pass.
• Build and format. go build, go vet and go test ./... pass in all 9 packages. gofmt -l on the committed bytes of the 3 changed files prints nothing.
• Integration, TestIntegration_EveryNodeTypeStratumParses:
  • fails on ffs0 4.0.7, 4.0.9-B and mtdc-2.1.0, with exactly event_notice, governance_proposal (S0-S2), message_packet, tool_call, tool_result;
  • passes on mtdc-2.2.0;
  • is the only integration failure at the head.
• Real logs. Every LINK in the primary, menno, lola, moos and scratch logs (707) was replayed through ValidateStrataLink on both commits, over five ontologies.
  • 0 verdicts change. No logged LINK touches one of the five types: 8 governance_proposal nodes are ADDed, but never LINKed.
  • The existing TestIntegration_T342_ShadowLog cannot see this change, because it only calls ValidateLINK.
  • Replay at boot does not call ValidateStrataLink, so restarting on today's logs is unaffected.
• Future effect on 4.0.7. Measured over every (declared pair × source type × target type):
  • 18,169 combinations go from admitted to refused, all of them touching the five types. Nothing goes the other way, and nothing else changes.
  • That matches operad: ValidateStrataLink returns early on an unparseable stratum and skips the type-list rules (rules 2 and 3) #77's own figures (21,713 → 3,544 admitted).
  • On mtdc-2.2.0: 0 changes.

Low: two test gaps (optional; add as separate tests so the 4/6 count stays valid)

Four of ten plausible wrong fixes survive the suite: an || guard, a source-only guard, no guard at all, and keeping the early return when both ends are unparseable. Two small tests in validate_link_pair_test.go would catch them:

• (a) a LINK where both ends are unparseable (governance_proposal → tool_call) must still be type-checked;
• (b) an S4 source → an unparseable target must be admitted. Rule 1 is skipped, not applied.

On Copilot's CI point

The new integration test fails on ffs0's 4.0.7, as #77 intends. Today's green ontology-integration check proves nothing: its log says FFS0_READ_TOKEN is not configured and no integration test ran. Once the token is set, or for anyone running MOOS_INTEGRATION=1 locally on 4.0.x, the suite is red until ffs0 gives the five types a stratum. Sam's call:

• (i) Keep the red flag on purpose. The operad half stays visibly owed.
• (ii) Give the five types S2 in ffs0. This could be in the planned 4.0.9 bump (as mtdc-2.2.0 did) or a separate 4.0.x.
  • The cost is 18,169 shapes refused for future LINKs, which also applies on the current 23ba9ce build (the stratum skip ends for those types).
  • No logged LINK is affected.
  • The R1 runbook's gate then stays green on later builds too.

Merge order: #81 → #82 → #83. If #82 is squashed or rebased, rebase this onto master first.

— Z440 seat (urn:moos:agent:claude-cowork.hp-z440), harness-session 71a0b53d

[MSD21091969]

t342 · Z440 seat · re-checked 06ca234 (G5b). It adds commits on top of c38a8c2, with no rewrite. Both new tests are separate functions, so the 4/6 count on the verbatim test still holds.

The four wrong fixes that survived before now each fail:

wrong fix	killed by
`
source-only guard	TestValidateStrataLink_S4ToUnparseable_Admitted
no guard	TestValidateStrataLink_S4ToUnparseable_Admitted
early return when both ends are unparseable	TestValidateStrataLink_BothUnparseable_StillTypeChecked

gofmt -l is clean, and go vet plus go test ./... pass in all 9 packages. Ready to merge after #82, from the Z440 review side. The integration flag on ffs0 4.x stays red until the five strata land, as agreed. — Z440 seat, harness-session 71a0b53d