feat(reasoning): ReasoningEvent.controlFlow + control-flow eventType vocabulary

CHANGELOG.md

@@ -9,6 +9,7 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/). Thi
 ## [Unreleased]
 
 ### Added
+- `ReasoningEvent.controlFlow` (optional) + a reserved control-flow `eventType` vocabulary (`reasoning.tool.called` / `reasoning.decision.branched` / `reasoning.subrun.spawned` / `reasoning.subrun.joined` / `reasoning.run.completed`) so emitters can carry a run's operational control flow for downstream narration-fidelity verification (SP-TRACE-CFR). Backward-compatible; operational structure only (no raw reasoning). See `docs/reasoning-control-flow-events.md` and `examples/reasoning_event_control_flow.json`.
 - SourceOS interaction substrate top-level index and README discovery links for `SourceOSInteractionEvent`, generated TypeScript/Python artifacts, and the Noetica → Superconscious → AgentPlane → AgentTerm reference flow.
 - Runtime observability and capability governance contracts: `CapabilityLedger`, `BrowserAutomationReceipt`, `GitWorkspaceState`, `OrphanEventReceipt`, and `RuntimeInstallReceipt` with canonical examples, validation wiring (`tools/validate_runtime_observability_examples.py`), a contract catalog, and ADR-0012.
 - Reasoning run contracts: `ReasoningRun`, `ReasoningEvent`, `ReasoningReceipt`, `ReasoningReplayPlan`, and `ReasoningBenchmark` with canonical examples and a contract-additions note for the Superconscious reference loop.

docs/reasoning-control-flow-events.md

@@ -0,0 +1,45 @@
+# Reasoning Control-Flow Events
+
+**Status:** additive extension to `ReasoningEvent` (v2). Backward-compatible: `eventType`
+stays a free string and `controlFlow` is optional.
+
+## Purpose
+
+`ReasoningRun` records run lifecycle; it did not carry the *control-flow* of a run
+(the sequence of tool calls, branches, and delegated sub-runs). Downstream
+narration-fidelity verification (SP-TRACE-CFR) recovers orchestration structure from
+this control flow to check that an agent's stated narration matches what it actually
+did. This extension lets emitters (Noetica → Superconscious → AgentPlane → AgentTerm,
+TurtleTerm, BearBrowser) produce that structure.
+
+**Privacy posture (intentional):** control-flow events carry *operational structure
+only* — a site id, a branch label, a guard position, a sub-run id — never raw model
+reasoning. `traceLevel` and `trustLevel` continue to scope disclosure and provenance.
+This is the "verify what the agent did, not what it thought" guarantee.
+
+## Reserved control-flow eventTypes
+
+| `eventType` | `controlFlow` fields | Recovered structure |
+|---|---|---|
+| `reasoning.tool.called` | `site` | a tool-call node |
+| `reasoning.decision.branched` | `site`, `branchTaken`, `guardPosition?` | branch / loop guard (pre=WHILE, post=DO_WHILE) |
+| `reasoning.subrun.spawned` | `site`, `sidechainId` | delegated sub-run entry (SESE) |
+| `reasoning.subrun.joined` | `site`, `sidechainId` | delegated sub-run return |
+| `reasoning.run.completed` | `site` | terminal |
+
+`site` is a stable orchestration-site identity (the decision-fold key — not a payload
+hash). Events are consumed in the run's causal order.
+
+## Argument-level provenance
+
+A `reasoning.tool.called` event MAY set `controlFlow.arg` to name the argument whose
+provenance carries this event's `trustLevel`. Consumers map `trustLevel` onto an
+integrity lattice for argument-level information-flow control (e.g. an argument derived
+from `untrusted-observation` is denied at a trusted sink).
+
+## Consumer
+
+AgentPlane's `sp-run narration-gate` / `sp-run attest-run` project a stream of these
+events into a control-flow segment, recover it with two engines, compare it against the
+agent's narration claims, and emit a signed run attestation. See
+`agentplane/tools/trace_cfr_reasoning_bridge.py`.

examples/reasoning_event_control_flow.json

@@ -0,0 +1,16 @@
+{
+  "id": "urn:srcos:reasoning-event:demo-dowhile-guard",
+  "type": "ReasoningEvent",
+  "specVersion": "2.0.0",
+  "runRef": "urn:srcos:reasoning-run:superconscious-demo",
+  "eventType": "reasoning.decision.branched",
+  "summary": "Loop guard evaluated after the body executed (post-checked).",
+  "traceLevel": "workspace-safe",
+  "trustLevel": "trusted-control-input",
+  "capturedAt": "2026-07-04T00:00:00Z",
+  "controlFlow": {
+    "site": "retry-loop-guard",
+    "branchTaken": "false",
+    "guardPosition": "post"
+  }
+}

schemas/ReasoningEvent.json

@@ -15,6 +15,19 @@
     "summary": { "type": "string" },
     "traceLevel": { "enum": ["public-safe", "workspace-safe", "operator-private", "restricted"] },
     "trustLevel": { "enum": ["trusted-control-input", "trusted-workspace-source", "semi-trusted-project-source", "untrusted-observation", "restricted-material"] },
-    "capturedAt": { "type": "string", "format": "date-time" }
+    "capturedAt": { "type": "string", "format": "date-time" },
+    "controlFlow": {
+      "type": "object",
+      "description": "Optional operational control-flow annotation. Present on the reserved control-flow eventTypes so downstream consumers can recover orchestration structure (e.g. SP-TRACE-CFR narration-fidelity verification) WITHOUT collecting raw reasoning — operational structure only, honoring traceLevel/trustLevel. Reserved eventTypes: 'reasoning.tool.called' (site), 'reasoning.decision.branched' (site, branchTaken, guardPosition), 'reasoning.subrun.spawned' / 'reasoning.subrun.joined' (site, sidechainId), 'reasoning.run.completed' (site).",
+      "properties": {
+        "site": { "type": "string", "description": "Stable orchestration-site identity; the decision-fold key (not a payload hash)." },
+        "branchTaken": { "type": ["string", "null"], "description": "Edge label on a decision successor (e.g. true/false/case:k)." },
+        "guardPosition": { "enum": ["pre", "post", null], "description": "Loop guard relative to first body execution: pre = pre-checked (WHILE), post = post-checked (DO_WHILE)." },
+        "sidechainId": { "type": ["string", "null"], "description": "Delegated sub-run id for spawn/join (SESE sidechain)." },
+        "arg": { "type": ["string", "null"], "description": "Argument name whose provenance carries this event's trustLevel (argument-level taint / IFC)." }
+      },
+      "required": ["site"],
+      "additionalProperties": true
+    }
   }
 }