review(2/3): runtime contracts, diagnostics and workflow qualification

.github/workflows/quality-gates.yml

@@ -190,12 +190,108 @@ jobs:
       - name: Check scheduler standard-trait accessor lint
         run: cargo clippy -p pantograph-scheduler --all-targets -- -D clippy::should_implement_trait
 
+      - name: Run runtime-host contract and accessor compatibility tests
+        run: cargo test -p pantograph-runtime-host-contracts
+      - name: Check runtime-host standard-trait accessor lint
+        run: cargo clippy -p pantograph-runtime-host-contracts --all-targets -- -D clippy::should_implement_trait
+      - name: Run diagnostics ledger contract tests
+        run: cargo test -p pantograph-diagnostics-ledger
+      - name: Run media conversion contract tests
+        run: cargo test -p pantograph-media-conversion
+
+      - name: Run inference compatibility and contract tests
+        run: |
+          cargo test -p inference --lib backend::compatibility::tests
+          cargo test -p inference --lib model_contracts::tests
+          cargo test -p inference --lib server::tests::runtime_matchers_preserve_path_component_comparison
+
+      - name: Run inference payload serialization and planner tests
+        run: |
+          cargo test -p inference --lib managed_runtime::contracts::tests
+          cargo test -p inference --lib image_generation_planner_tests
+          cargo test -p inference --features backend-pytorch --lib pytorch_worker_image_contract_tests -- --list > "$RUNNER_TEMP/pytorch-image-contract-tests.list"
+          python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith("backend::pytorch::pytorch_worker_image_contract_tests::") and line.endswith(": test")]; print("PyTorch image contract tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/pytorch-image-contract-tests.list"
+          cargo test -p inference --features backend-pytorch --lib pytorch_worker_image_contract_tests
+
+      - name: Run gateway attribution and warmup regression tests
+        run: |
+          cargo test -p inference --lib gateway::tests::private_lifecycle_context
+          cargo test -p inference --lib gateway::tests::test_runtime_lifecycle_snapshot
+          cargo test -p inference --lib gateway::tests::test_failed_restart_clears_active_runtime_config_and_attempted_modes
+
+      - name: Run typed error boundary regression tests
+        run: |
+          cargo test -p inference --lib managed_binaries::tests
+          cargo test -p inference --test model_contracts task_registry_resolution_
+
+      - name: Run named PyTorch text request contract tests
+        run: |
+          cargo test -p inference --features backend-pytorch --lib pytorch_named_text_request -- --list > "$RUNNER_TEMP/pytorch-named-text-tests.list"
+          python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith("backend::pytorch::tests::pytorch_named_text_request") and line.endswith(": test")]; print("Named PyTorch text tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/pytorch-named-text-tests.list"
+          cargo test -p inference --features backend-pytorch --lib pytorch_named_text_request
+          cargo test -p inference --features backend-pytorch --lib test_pytorch_generate_text_
+
+      - name: Run llama server startup and reuse contract tests
+        run: |
+          for test_filter in server::tests backend::llamacpp::tests; do
+            cargo test -p inference --features backend-llamacpp --lib "$test_filter" -- --list > "$RUNNER_TEMP/llama-settings-tests.list"
+            python3 -c 'import pathlib, sys; prefix = sys.argv[2] + "::"; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith(prefix) and line.endswith(": test")]; print(prefix, "tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/llama-settings-tests.list" "$test_filter"
+            cargo test -p inference --features backend-llamacpp --lib "$test_filter"
+          done
+
+      - name: Run contract-only node lifecycle tests
+        run: |
+          cargo test -p node-engine --features inference-nodes --lib rejects_contract_only_with_lifecycle -- --list > "$RUNNER_TEMP/node-validation-lifecycle-tests.list"
+          python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith("core_executor::tests::inference_tests::") and "rejects_contract_only_with_lifecycle" in line and line.endswith(": test")]; print("Contract-only lifecycle tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/node-validation-lifecycle-tests.list"
+          cargo test -p node-engine --features inference-nodes --lib rejects_contract_only_with_lifecycle
+
       - name: Run node-engine tests
         run: cargo test -p node-engine --lib
 
       - name: Run workflow-nodes tests
         run: cargo test -p workflow-nodes --lib
 
+      - name: Run runtime registry contract tests
+        run: cargo test -p pantograph-runtime-registry
+      - name: Run task summary contract tests
+        run: |
+          cargo test -p pantograph-workflow-service --lib workflow::task_run_summary::tests:: -- --list > "$RUNNER_TEMP/task-summary-tests.list"
+          python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith("workflow::task_run_summary::tests::") and line.endswith(": test")]; print("Task summary tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/task-summary-tests.list"
+          cargo test -p pantograph-workflow-service --lib workflow::task_run_summary::tests::
+      - name: Run complete session execution regression suite
+        run: |
+          cargo test -p pantograph-workflow-service --lib workflow::tests::session_execution:: -- --list > "$RUNNER_TEMP/session-execution-tests.list"
+          python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith("workflow::tests::session_execution::") and line.endswith(": test")]; print("Session execution tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/session-execution-tests.list"
+          cargo test -p pantograph-workflow-service --lib workflow::tests::session_execution::
+      - name: Run terminal diagnostic regression suite
+        run: |
+          cargo test -p pantograph-workflow-service --lib workflow::runtime_branch_run_finalization::tests:: -- --list > "$RUNNER_TEMP/terminal-diagnostic-tests.list"
+          python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith("workflow::runtime_branch_run_finalization::tests::") and line.endswith(": test")]; print("Terminal diagnostic tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/terminal-diagnostic-tests.list"
+          cargo test -p pantograph-workflow-service --lib workflow::runtime_branch_run_finalization::tests::
+      - name: Run workflow inference projection regression suites
+        run: |
+          for module in workflow::tests::task_graph:: workflow::tests::task_binding_resolution:: workflow::executable_validation_snapshot::tests::; do
+            cargo test -p pantograph-workflow-service --lib "$module" -- --list > "$RUNNER_TEMP/projection-tests.list"
+            python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith(sys.argv[2]) and line.endswith(": test")]; print(sys.argv[2], "tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/projection-tests.list" "$module"
+            cargo test -p pantograph-workflow-service --lib "$module"
+          done
+      - name: Run private workflow owner regression suites
+        run: |
+          for module in graph::inference_validation_state::tests:: workflow::task_execution_worker::tests::; do
+            cargo test -p pantograph-workflow-service --lib "$module" -- --list > "$RUNNER_TEMP/private-owner-tests.list"
+            python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith(sys.argv[2]) and line.endswith(": test")]; print(sys.argv[2], "tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/private-owner-tests.list" "$module"
+            cargo test -p pantograph-workflow-service --lib "$module"
+          done
+      - name: Run workflow style contract regressions
+        run: |
+          cargo test -p pantograph-workflow-service --lib lint_style_ -- --list > "$RUNNER_TEMP/workflow-style-tests.list"
+          python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if "lint_style_regressions::lint_style_" in line and line.endswith(": test")]; print("Workflow style tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/workflow-style-tests.list"
+          cargo test -p pantograph-workflow-service --lib lint_style_
+      - name: Run workflow scheduler orchestrator regression tests
+        run: |
+          cargo test -p pantograph-workflow-service --lib scheduler::task_orchestrator::tests:: -- --list > "$RUNNER_TEMP/orchestrator-tests.list"
+          python3 -c 'import pathlib, sys; names = [line for line in pathlib.Path(sys.argv[1]).read_text().splitlines() if line.startswith("scheduler::task_orchestrator::tests::") and line.endswith(": test")]; print("Orchestrator tests discovered:", len(names)); sys.exit(0 if names else 1)' "$RUNNER_TEMP/orchestrator-tests.list"
+          cargo test -p pantograph-workflow-service --lib scheduler::task_orchestrator::tests::
       - name: Run workflow-service contract tests
         run: cargo test -p pantograph-workflow-service --test contract
 

CHANGELOG.md

@@ -14,6 +14,29 @@ The format is based on Keep a Changelog.
   canonicalization records.
 
 ### Changed
+- Rust source API: `LlamaServer::start_sidecar_inference` and
+  `matches_inference_runtime` now borrow the existing `LlamaCppRuntimeSettings`
+  for effective device/context/thread/batch values. Model/mmproj paths, process
+  spawner and port override retain their existing roles and ownership.
+- Rust source API: PyTorch `generate_with_top_k` and `generate_stream_with_top_k`
+  now accept `PyTorchTextGenerationRequest` with the same seven named inputs.
+  Legacy `generate`/`generate_stream` signatures and the worker wire contract
+  remain unchanged.
+- Rust source API: `resolve_managed_binary_command` and
+  `resolve_task_registry_entry_from_evidence` now return boxed concrete errors.
+  Error variants, diagnostic JSON and Display text remain unchanged; callers
+  matching an owned error must dereference the box.
+- Rust source API: `ManagedRuntimeCommandResolutionError::MissingRuntimeVariant.diagnostic`
+  and `ImageGenerationPlanningOutcome::Planned.plan` now hold boxed payloads.
+  Direct constructors must box their values and consuming plan callers must unbox;
+  tagged JSON, diagnostic fields, and error Display text are unchanged.
+- Rust source API: `MediaConversionResult::try_new` now accepts one
+  `MediaConversionResultInput` with the same eight named fields instead of eight
+  positional arguments. Result fields and validation behavior are unchanged.
+- Rust source API: `SchedulerTaskExecutionIntent::Runtime.task_intent` is now boxed.
+  Construct runtime variants with `SchedulerTaskExecutionIntent::runtime(intent)`;
+  direct struct-variant callers must pass `Box::new(intent)`. Serialized JSON and
+  borrowed `runtime_task_intent()` access remain unchanged.
 - Root project `README.md` reorganized around install, usage, development, and contribution workflows.
 - Documentation consolidated around current guides, accepted decisions, audits,
   and active plans; superseded narration remains available in Git history.
@@ -24,3 +47,19 @@ The format is based on Keep a Changelog.
 
 ### Security
 - Canonical path validation enforced at file-boundary entry points to block traversal and symlink escape paths.
+
+### Runtime-host borrowed accessor compatibility
+
+The seven validated runtime-host contract wrappers now implement standard `AsRef<Raw>` in place of inherent `as_ref` methods. Borrowed values and lifetimes, wrapper types, validation and `into_inner` are unchanged. Ordinary method syntax, qualified calls and function pointers continue to resolve through the standard prelude. Rust callers using `no_implicit_prelude` must explicitly import `std::convert::AsRef`. The workspace crate remains unpublished (`publish = false`).
+
+### Diagnostic event inference payload construction
+
+`DiagnosticEventPayload::InferenceExecutionDiagnosticObserved` now stores `Box<InferenceExecutionDiagnosticObservedPayload>`. Rust callers constructing this public variant must wrap the existing raw payload in `Box::new`; owned pattern bindings now contain a box. Tagged serialized JSON, raw payload fields and validation remain unchanged. This adds one allocation for this event variant; no measured performance improvement is claimed. The crate remains `publish = false`.
+
+### Diagnostic event artifact payload construction
+
+`DiagnosticEventPayload::IoArtifactObserved` now stores `Box<IoArtifactObservedPayload>`. Rust callers constructing the public variant use `Box::new`, and owned pattern bindings contain a box. Raw DTO fields, validation and tagged JSON stay unchanged; this adds one allocation for this variant without a measured performance claim. The crate remains `publish = false`.
+
+### Workflow ready inference projection construction
+
+`WorkflowSchedulerInferenceTaskProjection::Ready` now holds `Box<WorkflowSchedulerReadyInferenceTaskProjection>`. Rust constructors use `Box::new`, and owned pattern bindings contain a box; the raw record, equality, borrowed readers and downstream scheduler intent serialization remain unchanged. The projection enum itself has no Serde implementation. This adds one allocation for ready projections without a measured performance claim; the crate remains `publish = false`.

crates/inference/src/backend/compatibility.rs

@@ -181,21 +181,25 @@ impl BackendCapabilities {
             &mut issues,
         );
         let option_diagnostics = self.option_compatibility_diagnostics(backend_key, &request);
-        issues.extend(option_diagnostics.iter().filter_map(|diagnostic| {
-            matches!(
-                &diagnostic.state,
-                OptionSupportState::Unsupported | OptionSupportState::Rejected
-            )
-            .then(|| BackendCompatibilityIssue {
-                kind: BackendCompatibilityIssueKind::UnsupportedOption,
-                phase: InferenceLifecyclePhase::TaskValidation,
-                message: diagnostic.message.clone().unwrap_or_else(|| {
-                    format!("option {} is not supported", diagnostic.option_path)
+        issues.extend(
+            option_diagnostics
+                .iter()
+                .filter(|diagnostic| {
+                    matches!(
+                        &diagnostic.state,
+                        OptionSupportState::Unsupported | OptionSupportState::Rejected
+                    )
+                })
+                .map(|diagnostic| BackendCompatibilityIssue {
+                    kind: BackendCompatibilityIssueKind::UnsupportedOption,
+                    phase: InferenceLifecyclePhase::TaskValidation,
+                    message: diagnostic.message.clone().unwrap_or_else(|| {
+                        format!("option {} is not supported", diagnostic.option_path)
+                    }),
+                    model_id: Some(request.package_facts.model_ref.model_id.clone()),
+                    path: None,
                 }),
-                model_id: Some(request.package_facts.model_ref.model_id.clone()),
-                path: None,
-            })
-        }));
+        );
         let compatible = [task, model_source, preprocessing, postprocessing]
             .into_iter()
             .all(|status| status == BackendCompatibilityStatus::Supported)
@@ -452,10 +456,7 @@ fn compatibility_report_status_label(report: &BackendCompatibilityReport) -> &'s
         report.preprocessing,
         report.postprocessing,
     ];
-    if statuses
-        .iter()
-        .any(|status| *status == BackendCompatibilityStatus::Unsupported)
-    {
+    if statuses.contains(&BackendCompatibilityStatus::Unsupported) {
         "rejected"
     } else {
         "degraded"
@@ -909,6 +910,7 @@ mod tests {
             Some("llama_cpp"),
             BackendCompatibilityRequest::new(&task, &package).with_options(
                 BackendCompatibilityOptions {
+                    streaming: true,
                     cache: CacheGenerationOptions {
                         use_cache: Some(true),
                         kv_cache_checkpoint_requested: Some(true),
@@ -919,6 +921,17 @@ mod tests {
         );
 
         assert!(!report.compatible);
+        assert_eq!(
+            report
+                .issues
+                .iter()
+                .map(|issue| issue.message.as_str())
+                .collect::<Vec<_>>(),
+            vec![
+                "option cache.use_cache is not supported by this backend",
+                "option cache.kv_cache_checkpoint_requested is not supported by this backend",
+            ],
+        );
         assert!(report.option_diagnostics.iter().any(|diagnostic| {
             diagnostic.option_path == "cache.use_cache"
                 && matches!(&diagnostic.state, OptionSupportState::Unsupported)

crates/inference/src/backend/llamacpp.rs

@@ -212,7 +212,6 @@ impl InferenceBackend for LlamaCppBackend {
 
         let runtime_settings = LlamaCppRuntimeSettings::try_from_backend_config(config)?;
         let device_config = runtime_settings.device_config();
-        let context_size = runtime_settings.context_size;
 
         if config.embedding_mode {
             // Start in embedding mode
@@ -289,11 +288,7 @@ impl InferenceBackend for LlamaCppBackend {
             if self.server.matches_inference_runtime(
                 &model_path.to_string_lossy(),
                 mmproj_path.as_deref(),
-                &device_config,
-                context_size,
-                runtime_settings.cpu_threads,
-                runtime_settings.batch_size,
-                runtime_settings.ubatch_size,
+                &runtime_settings,
                 config.port_override,
             ) {
                 return Ok(BackendStartOutcome {
@@ -307,11 +302,7 @@ impl InferenceBackend for LlamaCppBackend {
                     spawner,
                     &model_path.to_string_lossy(),
                     mmproj_path.as_deref(),
-                    &device_config,
-                    context_size,
-                    runtime_settings.cpu_threads,
-                    runtime_settings.batch_size,
-                    runtime_settings.ubatch_size,
+                    &runtime_settings,
                     config.port_override,
                 )
                 .await

crates/inference/src/backend/mod.rs

@@ -63,7 +63,7 @@ pub use llamacpp::LlamaCppBackend;
 pub use candle::CandleBackend;
 
 #[cfg(feature = "backend-pytorch")]
-pub use pytorch::PyTorchBackend;
+pub use pytorch::{PyTorchBackend, PyTorchTextGenerationRequest};
 
 pub use compatibility::{
     BackendCompatibilityIssue, BackendCompatibilityIssueKind, BackendCompatibilityOptions,