Live local E2E evidence — malformed safety content

Tested on the exact malformed webhook fixture:

[automaton]
name = "Order"
states = ["Draft"]
initial = "Draft"

[[webhook]]
name = "callback"
path = "callbacks/result"
method = "POST"

Baseline binary SHA-256:

f577b620751d78ac76384233da6a111077f77037a62e2cfbe97ed8257ba581aa  /tmp/arn214-temper-before

Current binary built from PR head e31b6c406941d6f60e18fa7972792c2152b8a246:

fe13ae7a41597301d75cb444269e1fa07a6b6f8ca888a6b1d51327b170a348d9  target/debug/temper

Before: malformed content was silently omitted

Command:

/tmp/arn214-temper-before verify --specs-dir /tmp/arn214-e2e/specs

Exit: 0

Output:

Running verification cascade...
  Specs directory: /tmp/arn214-e2e/specs

Running IOA verification cascade...

  Verifying Order...
proptest: FileFailurePersistence::SourceParallel set, but no source file known
    [PASS] L0 Symbolic PASSED: 0 guards satisfiable, 1 invariants inductive, 0 unreachable
    [PASS] L1 Model Check PASSED: 1 states explored, all properties hold
    [PASS] L2 Simulation PASSED: 5 seeds, 0 transitions, 0 dropped msgs
    [PASS] L3 Property Tests PASSED: 100 cases, 30 max steps

IOA verification cascade: ALL PASSED

Verification Report
==================================================

Specification Summary:
  Entity types:    8
  Actions:         12
  Functions:       5
  State machines:  0

Warnings (3):
  WARN: Order: has TlaSpec annotation but no specification source was provided
  WARN: Payment: has TlaSpec annotation but no specification source was provided
  WARN: Shipment: has TlaSpec annotation but no specification source was provided

==================================================
Result: PASS -- all cross-validation checks passed.

Note: Full model checking (Stateright) is not yet integrated.
      Run TLC separately for exhaustive state space exploration.

Live server command:

RUST_LOG=error TURSO_URL=file:/tmp/arn214-e2e/before-evidence-quiet.db /tmp/arn214-temper-before serve --no-observe --specs-dir /tmp/arn214-e2e/specs --port 3218

Startup output:

  Loading app: default from /tmp/arn214-e2e/specs
    Loaded spec: Order (verification pending, lint clean)
Starting Temper platform server...
  Temper Data API: http://localhost:3218/tdata
  [verify] Starting verification for Order...
  [verify] Order: [PASS] L0 Symbolic PASSED: 0 guards satisfiable, 1 invariants inductive, 0 unreachable
  [verify] Order: [PASS] L1 Model Check PASSED: 1 states explored, all properties hold
  [verify] Order: [PASS] L2 Simulation PASSED: 5 seeds, 0 transitions, 0 dropped msgs
  [verify] Order: [PASS] L3 Property Tests PASSED: 100 cases, 30 max steps
  [verify] Order: all levels passed
Listening on http://0.0.0.0:3218

Health command and exact response:

curl -sS -i http://127.0.0.1:3218/healthz

HTTP/1.1 200 OK
content-length: 0
date: Tue, 14 Jul 2026 16:31:55 GMT

After: parser rejects before verification or bind

Command:

target/debug/temper verify --specs-dir /tmp/arn214-e2e/specs

Exit: 1

Output:

Running verification cascade...
  Specs directory: /tmp/arn214-e2e/specs
Error: Failed to parse IOA spec for Order

Caused by:
    TOML parse error: TOML parse error at line 6, column 1
      |
    6 | [[webhook]]
      | ^^^^^^^^^^^
    missing field `action`

Live server command:

TURSO_URL=file:/tmp/arn214-e2e/after-evidence.db target/debug/temper serve --no-observe --specs-dir /tmp/arn214-e2e/specs --port 3216

Exit: 1 before binding.

Output:

OTEL export configured: endpoint=https://logfire-us.pydantic.dev source=LOGFIRE_TOKEN logfire_auth=true
  Storage: turso (file:/tmp/arn214-e2e/after-evidence.db)
  Loading app: default from /tmp/arn214-e2e/specs
Error: failed to parse IOA spec for Order

Caused by:
    TOML parse error: TOML parse error at line 6, column 1
      |
    6 | [[webhook]]
      | ^^^^^^^^^^^
    missing field `action`

Behavioral result: the same malformed safety declaration that previously passed all levels with zero transitions and reached a healthy live server is now rejected durably at the parser boundary.

Independent GPT-5.6 open-PR diff review

Reviewed head SHA: e31b6c4
Reviewed base SHA: a28fdb2

Findings: No actionable findings.

Review coverage included correctness and fail-closed behavior of the canonical whole-document schema parser; durability and source-located rejection of malformed, incomplete, unknown, and duplicate declarations; legacy guard/effect and integration-config compatibility; synthesized-trigger canonical round-trip handling; repository-spec migrations; test adequacy; and deterministic-simulation constraints. The parser change is pure and deterministic, adds no clocks, randomness, concurrency, or simulation-visible I/O.

Independent validation completed on the reviewed SHA:

• cargo test -p temper-spec: 241 unit tests passed, the 130-spec canonical corpus round trip passed, and 3 migration differential tests passed
• cargo fmt --all -- --check: passed
• git diff --check for the GitHub base-to-head range: passed
• GitHub completed checks were green at posting time; the aggregate Tests job was still in progress

Verdict: PASS

@greptile review

Greptile remediation audit

Greptile completed successfully on head e31b6c406941d6f60e18fa7972792c2152b8a246.

• Check: https://github.com/nerdsane/temper/runs/87146501337
• Started: 2026-07-14T16:44:23Z
• Completed: 2026-07-14T16:52:04Z
• Result: success
• Scope: 29 files reviewed
• Check annotations: 0
• GitHub issue comments authored by Greptile: 0
• GitHub pull-request reviews authored by Greptile: 0
• GitHub inline review comments authored by Greptile: 0
• GraphQL review threads: 0

The check summary reports 1 comments added, but direct reads of all three GitHub comment surfaces plus thread-aware GraphQL state found no Greptile-authored item. There are therefore no actionable Greptile findings, no remediation commit, and no thread requiring a reply or resolution. The PR head and diff remain unchanged from the pre-Greptile independent PASS.

Final post-Greptile independent GPT-5.6 review

Reviewed head SHA: e31b6c4
Reviewed base SHA: a28fdb2

Diff-after-Greptile check: unchanged. The current GitHub head matches both the Greptile check-run head and the SHA recorded in the Greptile remediation audit. Greptile completed successfully on this SHA with zero annotations, and all current-head required CI checks are complete and green.

Findings: No actionable findings.

Review coverage included the complete live GitHub PR diff across all 29 files; fail-closed whole-document TOML parsing; unknown, incomplete, duplicate, and malformed declaration rejection; source-located parser failures; compatibility for legacy guards, effects, action parameters, composite metadata, integration configuration, and synthesized action triggers; canonical parse/serialize/parse stability; repository-spec migrations; syntax-hook behavior; test adequacy; regression risk; and deterministic-simulation constraints.

Greptile summary reconciliation: the embedded concern about duplicate authored integration trigger values is not actionable for this PR. The Temper platform IntegrationRegistry intentionally models one trigger to a Vec and tests multiple integrations on the same trigger. Adding parser-level trigger uniqueness would reject that supported fan-out. The separate legacy server lookup behavior predates this patch and is unchanged here.

The parser path remains pure and deterministic: it adds no clock, randomness, concurrency, unordered collections, or simulation-visible I/O.

Verdict: PASS

Greptile concern reconciliation — integration trigger fan-out

The final reviewer surfaced an embedded Greptile concern about duplicate authored integration trigger values. I traced it through the production behavior instead of dismissing it.

Concrete evidence on the reviewed head:

• Production IntegrationRegistry intentionally maps each trigger to Vec<IntegrationConfig> and appends every matching integration:

  temper/crates/temper-platform/src/integration/registry.rs

  Lines 11 to 28 in e31b6c4

  	pub struct IntegrationRegistry {
  	/// Maps event name to list of integrations triggered by that event.
  	by_trigger: BTreeMap<String, Vec<IntegrationConfig>>,
  	}
  	impl IntegrationRegistry {
  	/// Build a registry from a list of integration configs.
  	pub fn from_configs(configs: Vec<IntegrationConfig>) -> Self {
  	let mut by_trigger = BTreeMap::new();
  	for config in configs {
  	by_trigger
  	.entry(config.trigger.clone())
  	.or_insert_with(Vec::new)
  	.push(config);
  	}
  	Self { by_trigger }
  	}

• The unit test registers both fulfillment and analytics on SubmitOrder, then requires both to be returned in deterministic order:

  temper/crates/temper-platform/src/integration/registry.rs

  Lines 70 to 87 in e31b6c4

  	}
  	#[test]
  	fn lookup_returns_matching_configs() {
  	let registry = IntegrationRegistry::from_configs(vec![
  	test_config("fulfillment", "SubmitOrder"),
  	test_config("payment", "ConfirmOrder"),
  	test_config("analytics", "SubmitOrder"),
  	]);
  	let submit_integrations = registry.lookup("SubmitOrder");
  	assert_eq!(submit_integrations.len(), 2);
  	assert_eq!(submit_integrations[0].name, "fulfillment");
  	assert_eq!(submit_integrations[1].name, "analytics");
  	let confirm_integrations = registry.lookup("ConfirmOrder");
  	assert_eq!(confirm_integrations.len(), 1);
  	}

• Current-head CI, including all workspace tests, is green: https://github.com/nerdsane/temper/actions/runs/29349702282

Conclusion: parser-level uniqueness for authored integration trigger values would regress the supported one-to-many fan-out capability. No code change is correct here. This PR continues to reject malformed, incomplete, unknown, and duplicate declarations where identity must be unique, while retaining multiple named integrations that intentionally share one trigger.

There is no Greptile GitHub thread to reply to or resolve; the completed thread-aware audit found zero review threads. Final independent reconciliation and verdict: #390 (comment)

ARENA SHIPPABLE · GPT-5.6 · 2026-07-14T17:05:14Z

PR #390 is open on reviewed head e31b6c406941d6f60e18fa7972792c2152b8a246. MERGE NOTHING.

Mandatory gate evidence, in timestamp order:

1. Live RED/GREEN server and CLI E2E: fix(spec): reject malformed IOA safety content (ARN-214) #390 (comment)
2. Pre-Greptile fresh independent GPT-5.6 PASS: fix(spec): reject malformed IOA safety content (ARN-214) #390 (comment)
3. Subsequent @greptile review request: fix(spec): reject malformed IOA safety content (ARN-214) #390 (comment)
4. Greptile completed successfully, 29 files reviewed, zero annotations: https://github.com/nerdsane/temper/runs/87146501337
5. Greptile review/thread/remediation audit: fix(spec): reject malformed IOA safety content (ARN-214) #390 (comment)
6. Concrete trigger fan-out concern reconciliation: fix(spec): reject malformed IOA safety content (ARN-214) #390 (comment)
7. Final fresh post-Greptile GPT-5.6 PASS on the unchanged open PR diff: fix(spec): reject malformed IOA safety content (ARN-214) #390 (comment)
8. Current-head CI, every required check green: https://github.com/nerdsane/temper/actions/runs/29349702282

Greptile created zero GitHub reviews and zero GraphQL review threads, so there was no remediation commit, inline reply, or resolution mutation to make. The embedded trigger-uniqueness concern was traced to production BTreeMap<String, Vec<IntegrationConfig>> behavior and its two-integrations-on-one-trigger regression test; enforcing uniqueness would remove supported deterministic fan-out.

Commit chain:

• ADR: ade7461
• Isolated RED: ff66dfdd
• GREEN: e31b6c4

The PR does not touch crates/temper-actor-runtime. It remains open and unmerged.