docs(compliance): re-count quality gates; test(grpc): add Layer 2 conformance cases

[devin-ai-integration]

Summary

Two things the last batch left behind: stale numbers in the compliance docs, and a §5.2 Layer 2 row for the gRPC wrapper that had been an unmoving to-do. Both are now settled against actual runs on this branch.

Re-counted, not re-asserted. Every number below came from a run or from counting checked-in fixtures, replacing figures that predate the last seven merged PRs:

Claim	Was	Is
Execution conformance	39/39	43/43
Training corpus	80/100 (TRAINING_EXAMPLES.md said "80% clean")	81/100 — 19 files, 37 errors
Robustness subtests	25	27
Golden AST fixtures / negative parser subtests	23 / 17	24 / 19

The Coverage by Feature Type list also double-counted: its Evaluation: 3 conformance cases are three of the ten calc_* cases, and the per-type numbers summed to 41, not the actual 43. It now sums to 43 by fixture prefix. TRAINING_EXAMPLES.md additionally still listed implicit stdlib typing as an open cause of corpus errors — that was fixed by the implicit-typing work; the two remaining resolution causes (implicit redefinition, and members contributed by perform/references) replace it.

Added the caveat that the corpus gate needs ./scripts/download-training-examples.sh and therefore skips in CI, so a green check is not evidence it passed.

gRPC Layer 2 (❌ → ✅), by adding the cases rather than declaring a deviation. internal/grpc/testdata/conformance/ already contained evaluate_arithmetic and instantiate_part fixtures that no test read — dead data. internal/grpc/conformance_test.go now discovers every .expected.json there and drives the real handlers end-to-end:

parseResp := srv.ParseFile(ctx, &pb.ParseFileRequest{Source: content(model), ContentHash: caseName})
// any diagnostic of severity "error" fails the case
switch tc.RPC {
case "Evaluate":      srv.Evaluate(...)      // expected_result
case "Instantiate":   srv.Instantiate(...)   // expected_slots, type_symbol_id
case "ExecuteAction": srv.ExecuteAction(...) // inputs -> expected_outputs
case "ExecuteState":  srv.ExecuteState(...)  // events -> expected_states_visited
}

Value assertions compare the pb.Value oneof arm as well as the payload, so a right-looking number returned as the wrong type fails. expected_error pins in-band failure modes; execute_action_no_initial uses it to lock the AGENTS.md §4 contract that an empty action errors at initialize(), through the RPC boundary. Schema is documented in that directory's README.md, so new cases are a data-only change.

The two adopted fixtures needed import ScalarValues::*; to parse clean: unlike the runtime-tier harness, ParseFile loads the stdlib and runs the semantic passes, and there is no implicit library import yet.

The doc's gRPC handler table also pointed at line ranges that had drifted by ~40 lines; corrected.

Verification

gofmt -l .                        # empty
go build ./... && go vet ./...    # clean
go test ./...                     # pass
go test -race ./internal/grpc/... # pass
make lint                         # staticcheck + gosec, zero findings
go test ./internal/core/model/ -run TestTrainingExamples   # 81/100 clean, unchanged

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/fd6e8fba638c403f836a5dcab81205e5
Requested by: @HuiJun

[devin-ai-integration]

Original prompt from jason.han

@JPL-Devin/Systemica We've recently resynced with the open source branch. Reassess the branch and then continue on with theses next steps, using subsessions for each task that can be completed atomically.
ATTACHMENT:"https://nasa-jpl-demo.devinenterprise.com/attachments/b0443b98-63d4-4bf9-9eb2-6e053a420317/NEXT_SESSION%281%29.md"

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR that start with 'DevinAI' or '@devin'.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.