fix(compliance): skip observations when controlled forcing is unavailable

.changeset/unsupported-force-scenario-cascade.md

@@ -0,0 +1,5 @@
+---
+"adcontextprotocol": patch
+---
+
+Clarify that unsupported stateful test-controller forcing skips its dependent observations as unavailable coverage, including when forcing is the only stateful step in its phase. Preserve independent coverage and genuine failures without dispatching observations against state the controller never established.

static/compliance/source/universal/runner-output-contract.yaml

@@ -1218,6 +1218,14 @@ skip_result:
                               # canonical_detail_sub_reasons below.
   canonical_detail_sub_reasons:
     not_applicable:
+      force_scenario_unsupported:
+        description: |
+          A comply_test_controller call to a force_* scenario returned
+          success: false with error: UNKNOWN_SCENARIO. Grade the call
+          not_applicable before its authored validations, with an empty
+          validations array. Detail MUST name force_scenario_unsupported and
+          the requested scenario. Apply forced_scenario_cascade below when
+          the skipped call is stateful.
       rate_limit_not_triggered:
         spec_source: test-kits/rate-limit-trip-runner.yaml
         description: |
@@ -1700,13 +1708,32 @@ cascade_rules:
     still applies because the structural condition — no peer available to
     establish substitute state — is identical. A sole-stateful-step
     `not_applicable` that cascades to all downstream phases would collapse
-    useful coverage; the exemption prevents this regardless of the specific
-    skip detail. Dashboard consumers should read both `reason` and `detail`
+    useful coverage; the exemption prevents this except for unavailable
+    controlled forcing, as defined in forced_scenario_cascade below.
+    Dashboard consumers should read both `reason` and `detail`
     rather than inferring the cause from whether downstream phases ran.
 
     References: adcp-client#1146 (original `not_applicable` exemption);
     adcp-client#1545 (extension to `missing_tool` / `missing_test_controller`);
     adcp-client-python#550 (adopter pattern that surfaced the cross-runner gap).
+  forced_scenario_cascade: |
+    A stateful comply_test_controller force_* call that grades not_applicable
+    with detail force_scenario_unsupported did not establish the controlled
+    state required by its dependent observations. The runner MUST NOT apply
+    sole_stateful_step_exemption to this skip. It MUST skip subsequent
+    stateful steps in the same phase and stateful steps in phases depending
+    on it, including through intermediate skipped phases, before dispatch or
+    authored validation. These dependent skips MUST grade not_applicable,
+    cite the unsupported force step, and receive no executed-step or
+    validation credit. They MUST NOT grade prerequisite_failed solely
+    because optional forcing is unavailable.
+
+    Explicitly independent phases and any_of branch-set alternatives retain
+    their existing dependency semantics. Genuine failures take precedence
+    over this neutral coverage gap, and successful forcing does not suppress
+    observation failures. Other not_applicable, missing_tool and
+    missing_test_controller skips retain the ordinary exemption and
+    substitution rules above.
   branch_set_cascade_exemption: |
     When two or more phases share the same `branch_set.id` with `any_of`
     semantics, the runner MUST NOT cascade `prerequisite_failed` from one peer