@@ -35,15 +35,23 @@
             "level": "high",
             "reasons": [
               "tool exposure is observational rather than randomized",
-              "actors who use a tool may differ from the control cohort before exposure"
+              "actors who use a tool may differ from the control cohort before exposure",
+              "invoked-at gaps are descriptive timing only, not elapsed-time causal windows"
             ]
           },
           "interpretation": "linked_cohort_association_only",
           "returnWindow": {
+            "aggregate": {
+              "immediateNextSessionReturns": 6,
+              "medianInvokedAtGapToReturn": 1135,
+              "returningActorCount": 6,
+              "sustainedMultiSessionReturns": 0
+            },
+            "claimBoundary": "not_incremental_effect_not_causal",
             "kind": "next_observed_session",
             "label": "second_observed_session",
-            "timingMode": "session_count_only",
-            "explanation": "Return currently means the actor had any later observed session. Pathloom does not yet model elapsed time, censoring, or intervention timing."
+            "timingMode": "session_order_plus_invoked_at_gap",
+            "explanation": "Return means the actor had a later observed session. Pathloom also reports immediate (exactly two sessions) vs sustained (three or more) return shapes and median invoked-at gap between first and second sessions when timestamps exist. This is still observational — not incremental lift or causal timing."
           },
           "signalTaxonomy": {
             "candidate": "candidate_return_correlate",
@@ -58,6 +66,23 @@
       "id": "activation_tool_report",
       "items": [
         {
+          "cohortContext": {
+            "cohortSemantics": "first_session_exposure_vs_linked_control_without_exposure",
+            "comparisonId": "activation:shadow_ops_context:first_session_exposure",
+            "controlActorCount": 6,
+            "controlCohortLabel": "linked actors without shadow_ops_context in their first observed session",
+            "controlReturnRate": 0,
+            "exposedActorCount": 6,
+            "exposureCohortLabel": "linked actors who invoked shadow_ops_context in their first observed session",
+            "exposedReturnRate": 1,
+            "returnWindow": {
+              "immediateNextSessionReturns": 6,
+              "kind": "next_observed_session",
+              "label": "second_observed_session",
+              "medianInvokedAtGapToReturn": 1135,
+              "sustainedMultiSessionReturns": 0
+            }
+          },
           "confidenceBand": {
             "controlLower": 0,
             "controlUpper": 0.39,
@@ -201,7 +226,16 @@
           "sequenceLabel": "shadow_search_path -> shadow_inspect_file",
           "sessionCount": 6,
           "terminalToolBaselineErrorRate": 0,
-          "terminalToolErrorLift": 1
+          "terminalToolErrorLift": 1,
+          "trajectoryContext": {
+            "cohortSemantics": "ordered_paths_sharing_terminal_tool_and_suffix_peer_bucket_excludes_focus_path",
+            "familyDistinctPathCount": 2,
+            "familyId": "terminal:shadow_inspect_file:suffix_tail:shadow_inspect_file:len:2:risky_sequence",
+            "familyLabel": "other routes into shadow_inspect_file",
+            "peerObservationCount": 6,
+            "suffixBucketObservationCount": 12,
+            "windowLength": 2
+          }
         },
         {
           "baselineKind": "matched_suffix_peers",
@@ -252,7 +286,16 @@
           "successLift": 1,
           "successRate": 1,
           "terminalToolBaselineSuccessRate": 0,
-          "terminalToolSuccessLift": 1
+          "terminalToolSuccessLift": 1,
+          "trajectoryContext": {
+            "cohortSemantics": "ordered_paths_sharing_terminal_tool_and_suffix_peer_bucket_excludes_focus_path",
+            "familyDistinctPathCount": 2,
+            "familyId": "terminal:shadow_inspect_file:suffix_tail:shadow_inspect_file:len:2:golden_path",
+            "familyLabel": "other routes into shadow_inspect_file",
+            "peerObservationCount": 6,
+            "suffixBucketObservationCount": 12,
+            "windowLength": 2
+          }
         }
       ],
       "recommendation": "Focus on the highest-lift failure endpoint first. Tighten the handoff into that terminal tool or add a safer bridge before the failing step.",
@@ -428,7 +471,7 @@
       }
     }
   ],
-  "generatedAt": "2026-05-27T08:45:52.841Z",
+  "generatedAt": "2026-05-29T07:40:46.792Z",
   "reportVersion": "1.0",
   "sourceKey": "logfile:liquid-shadow-dogfood",
   "summary": {
 
@@ -1,7 +1,7 @@
 # Pathloom Report
 
 Source: `logfile:liquid-shadow-dogfood`
-Generated: 2026-05-27T08:45:52.841Z
+Generated: 2026-05-29T07:40:46.792Z
 Sessions: 18
 Events: 42
 Catalog tools: 6
@@ -16,12 +16,12 @@ Evidence strength: **Credible finding** (full claim)
 The dataset supported the full claim, and the observed pattern cleared the emission threshold.
 
 Interpretation limits:
-- Linked-cohort association only: Pathloom is not estimating incremental effect or causal lift here.
-- Return window: second_observed_session (Return currently means the actor had any later observed session. Pathloom does not yet model elapsed time, censoring, or intervention timing.)
-- Confounding risk: high; tool exposure is observational rather than randomized; actors who use a tool may differ from the control cohort before exposure.
+- Linked-cohort association only: Pathloom is not estimating incremental effect or causal lift here (not_incremental_effect_not_causal).
+- Return window: second_observed_session (session_order_plus_invoked_at_gap). Return means the actor had a later observed session. Pathloom also reports immediate (exactly two sessions) vs sustained (three or more) return shapes and median invoked-at gap between first and second sessions when timestamps exist. This is still observational — not incremental lift or causal timing. Aggregate timing: 6 returning actors (6 immediate next-session, 0 sustained multi-session; median invoked-at gap 1135).
+- Confounding risk: high; tool exposure is observational rather than randomized; actors who use a tool may differ from the control cohort before exposure; invoked-at gaps are descriptive timing only, not elapsed-time causal windows.
 - Actor linkage gate: 12 linked actors satisfied the privacy-safe cohort requirement.
 
-- `shadow_ops_context` is a credible cohort signal: 6/6 linked actors who used it in session 1 returned for a second observed session (100%) versus 0% for the linked control cohort with no control-group returns observed. 95% interval 61-100% vs control 0-39%
+- `shadow_ops_context` is a credible cohort signal: 6/6 linked actors who used it in session 1 returned for a second observed session (100%) versus 0% for the linked control cohort with no control-group returns observed. 95% interval 61-100% vs control 0-39% Cohort `activation:shadow_ops_context:first_session_exposure`: linked actors who invoked shadow_ops_context in their first observed session vs linked actors without shadow_ops_context in their first observed session (6 immediate next-session returns, 0 sustained multi-session returns; median invoked-at gap to return 1135).
 
 Recommendation: Surface this tool earlier in onboarding and make its value legible, because it is the strongest confidence-backed cohort return signal in this dataset.
 
@@ -46,8 +46,8 @@ Evidence strength: **Credible finding** (full claim)
 
 The dataset supported the full claim, and the observed pattern cleared the emission threshold.
 
-- `shadow_search_path -> shadow_inspect_file` reliably precedes the `shadow_inspect_file` failure endpoint: 100% errors across 6 repeated paths. other routes into shadow_inspect_file fail 0% across 6 peer paths, and other routes ending at shadow_inspect_file fail 0%; seen in cursor only.
-- `shadow_search_concept -> shadow_inspect_file` is a reusable success path into `shadow_inspect_file`: 100% success across 6 repeated paths. other routes into shadow_inspect_file succeed 0% across 6 peer paths, and other routes ending at shadow_inspect_file succeed 0%; seen in claude-desktop only.
+- `shadow_search_path -> shadow_inspect_file` reliably precedes the `shadow_inspect_file` failure endpoint: 100% errors across 6 repeated paths. other routes into shadow_inspect_file fail 0% across 6 peer paths, and other routes ending at shadow_inspect_file fail 0%; seen in cursor only. Trajectory family `terminal:shadow_inspect_file:suffix_tail:shadow_inspect_file:len:2:risky_sequence` (2-hop): other routes into shadow_inspect_file; compared vs 6 suffix-peer observations (ordered paths sharing terminal tool and suffix peer bucket excludes focus path; 2 distinct paths in the peer bucket).
+- `shadow_search_concept -> shadow_inspect_file` is a reusable success path into `shadow_inspect_file`: 100% success across 6 repeated paths. other routes into shadow_inspect_file succeed 0% across 6 peer paths, and other routes ending at shadow_inspect_file succeed 0%; seen in claude-desktop only. Trajectory family `terminal:shadow_inspect_file:suffix_tail:shadow_inspect_file:len:2:golden_path` (2-hop): other routes into shadow_inspect_file; compared vs 6 suffix-peer observations (ordered paths sharing terminal tool and suffix peer bucket excludes focus path; 2 distinct paths in the peer bucket).
 
 Recommendation: Focus on the highest-lift failure endpoint first. Tighten the handoff into that terminal tool or add a safer bridge before the failing step.
 
 
@@ -3,7 +3,7 @@
 6 ready findings, 0 clear findings, 0 suppressed findings, for logfile:liquid-shadow-dogfood across 18 sessions.
 
 Source: `logfile:liquid-shadow-dogfood`
-Generated: 2026-05-27T08:45:52.841Z
+Generated: 2026-05-29T07:40:46.792Z
 Telemetry: sessions explicit_session_id; trace none; span lineage none; first-class fields none
 
 ## Top findings
 
@@ -1,10 +1,10 @@
 # Pathloom Feedback Review
 
-0 of 11 feedback targets reviewed for snapshot logfile:liquid-shadow-dogfood@20260527T084552878Z.
+0 of 11 feedback targets reviewed for snapshot logfile:liquid-shadow-dogfood@20260529T074046834Z.
 
 Source: `logfile:liquid-shadow-dogfood`
-Snapshot: `logfile:liquid-shadow-dogfood@20260527T084552878Z`
-Captured: 2026-05-27T08:45:52.878Z
+Snapshot: `logfile:liquid-shadow-dogfood@20260529T074046834Z`
+Captured: 2026-05-29T07:40:46.834Z
 
 ## Summary