Add compound tool reliability scoring

ROADMAP.md

@@ -191,7 +191,7 @@ Derived data pipelines, performance optimization, durable background jobs, enter
 - [x] [P1] Outcome attribution for customizations: which MCPs, skills, commands, and subagents improve workflow, quality, cost, and friction outcomes
 - [x] [P1] Cross-team tooling landscape: overlap, reuse, and local best-of-breed tools
 - [x] [P1] High-performer agent stack analysis: which combinations of MCPs, skills, commands, and subagents differentiate top performers
-- [ ] [P0] Per-tool success rate tracking with compound reliability computation (10 steps at 99% = 90.4% end-to-end)
+- [x] [P0] Per-tool success rate tracking with compound reliability computation (10 steps at 99% = 90.4% end-to-end)
 - [ ] [P0] Harness configuration fingerprinting from session telemetry (tools, context files, permissions, customizations)
 - [ ] [P1] Context quality scoring: AGENTS.md freshness, token efficiency, guide/sensor coverage
 - [ ] [P1] Harness evolution timeline: before/after correlation of configuration changes with outcome changes

frontend/src/components/maturity/__tests__/toolchain-reliability-table.test.tsx

@@ -20,12 +20,15 @@ describe("ToolchainReliabilityTable", () => {
             source_classification: "marketplace",
             session_count: 4,
             engineer_count: 2,
+            total_call_count: 12,
+            avg_calls_per_session: 3,
             friction_session_count: 2,
             friction_session_rate: 0.5,
             failure_session_count: 1,
             failure_session_rate: 0.25,
             recovery_rate: 0.5,
             success_rate: 0.75,
+            compound_reliability_rate: 0.904,
             abandonment_rate: 0.25,
             avg_recovery_steps: 2.5,
             top_friction_types: ["tool_error", "timeout"],
@@ -37,7 +40,11 @@ describe("ToolchainReliabilityTable", () => {
     expect(screen.getByText("github")).toBeInTheDocument()
     expect(screen.getByText("Mcp")).toBeInTheDocument()
     expect(screen.getByText("Marketplace")).toBeInTheDocument()
+    expect(screen.getByText("12")).toBeInTheDocument()
+    expect(screen.getByText("3.0 / session")).toBeInTheDocument()
     expect(screen.getAllByText("50%")).toHaveLength(2)
+    expect(screen.getByText("90%")).toBeInTheDocument()
+    expect(screen.getByText("10-step chain")).toBeInTheDocument()
     expect(screen.getByText("25%")).toBeInTheDocument()
     expect(screen.getByText("25% abandoned")).toBeInTheDocument()
     expect(screen.getByText("2.5 steps")).toBeInTheDocument()

frontend/src/components/maturity/toolchain-reliability-table.tsx

@@ -35,10 +35,12 @@ export function ToolchainReliabilityTable({ rows }: ToolchainReliabilityTablePro
               <tr className="border-b border-border text-left text-muted-foreground">
                 <th className="pb-2 font-medium">Surface</th>
                 <th className="pb-2 text-right font-medium">Sessions</th>
+                <th className="pb-2 text-right font-medium">Calls</th>
                 <th className="pb-2 text-right font-medium">Friction</th>
                 <th className="pb-2 text-right font-medium">Failures</th>
                 <th className="pb-2 text-right font-medium">Recovery</th>
                 <th className="pb-2 text-right font-medium">Success</th>
+                <th className="pb-2 text-right font-medium">Compound</th>
                 <th className="pb-2 font-medium">Top Frictions</th>
               </tr>
             </thead>
@@ -70,6 +72,14 @@ export function ToolchainReliabilityTable({ rows }: ToolchainReliabilityTablePro
                       </p>
                     </div>
                   </td>
+                  <td className="py-2 text-right">
+                    <div>
+                      <p>{row.total_call_count}</p>
+                      <p className="text-xs text-muted-foreground">
+                        {row.avg_calls_per_session.toFixed(1)} / session
+                      </p>
+                    </div>
+                  </td>
                   <td className="py-2 text-right">
                     <div>
                       <p>{formatPercent(row.friction_session_rate)}</p>
@@ -104,6 +114,12 @@ export function ToolchainReliabilityTable({ rows }: ToolchainReliabilityTablePro
                       </p>
                     </div>
                   </td>
+                  <td className="py-2 text-right">
+                    <div>
+                      <p>{formatPercent(row.compound_reliability_rate)}</p>
+                      <p className="text-xs text-muted-foreground">10-step chain</p>
+                    </div>
+                  </td>
                   <td className="py-2">
                     <div className="flex flex-wrap gap-2">
                       {row.top_friction_types.length === 0 ? (

frontend/src/types/api.ts

@@ -1956,12 +1956,15 @@ export interface ToolchainReliabilityEntry {
   source_classification: CustomizationSourceClassification | null
   session_count: number
   engineer_count: number
+  total_call_count: number
+  avg_calls_per_session: number
   friction_session_count: number
   friction_session_rate: number | null
   failure_session_count: number
   failure_session_rate: number | null
   recovery_rate: number | null
   success_rate: number | null
+  compound_reliability_rate: number | null
   abandonment_rate: number | null
   avg_recovery_steps: number | null
   top_friction_types: string[]

src/primer/common/schemas.py

@@ -2375,12 +2375,15 @@ class ToolchainReliabilityEntry(BaseModel):
     source_classification: CustomizationSourceClassification | None = None
     session_count: int
     engineer_count: int
+    total_call_count: int = 0
+    avg_calls_per_session: float = 0.0
     friction_session_count: int
     friction_session_rate: float | None = None
     failure_session_count: int
     failure_session_rate: float | None = None
     recovery_rate: float | None = None
     success_rate: float | None = None
+    compound_reliability_rate: float | None = None
     abandonment_rate: float | None = None
     avg_recovery_steps: float | None = None
     top_friction_types: list[str] = []

src/primer/server/services/maturity_service.py

@@ -57,6 +57,7 @@
 from primer.server.services.effectiveness_service import build_effectiveness_score
 
 _RELIABILITY_FAILURE_FRICTION_TYPES = frozenset({"tool_error", "exec_error", "timeout"})
+_COMPOUND_RELIABILITY_CHAIN_LENGTH = 10
 
 
 def get_maturity_analytics(
@@ -707,6 +708,7 @@ def _ensure_reliability_bucket(
                 "recovery_steps": [],
                 "success_sessions": set(),
                 "friction_type_counts": Counter(),
+                "total_call_count": 0,
             },
         )
 
@@ -741,6 +743,7 @@ def _apply_session_reliability_metrics(
                 continue
             bucket = _ensure_reliability_bucket("built_in_tool", tool_name, "built_in", "built_in")
             bucket["sessions"].add(sid)
+            bucket["total_call_count"] += call_count
             if sid in session_metrics:
                 session_metric = session_metrics[sid]
                 engineer_id_for_session = session_metric["engineer_id"]
@@ -850,12 +853,23 @@ def _apply_session_reliability_metrics(
         )
         reliability_bucket["sessions"].add(session_id)
         reliability_bucket["engineers"].add(engineer_id)
+        reliability_bucket["total_call_count"] += invocation_count or 0
         _apply_session_reliability_metrics(
             reliability_bucket,
             session_id,
             session_metrics.get(session_id),
         )
 
+    def _compound_reliability_rate(bucket: dict) -> float | None:
+        if not bucket["sessions"]:
+            return None
+        success_rate = len(bucket["success_sessions"]) / len(bucket["sessions"])
+        avg_calls = bucket["total_call_count"] / len(bucket["sessions"])
+        if avg_calls <= 0:
+            return None
+        per_call_rate = success_rate ** (1.0 / avg_calls)
+        return round(per_call_rate**_COMPOUND_RELIABILITY_CHAIN_LENGTH, 3)
+
     customization_breakdown = [
         CustomizationUsage(
             identifier=bucket["identifier"],
@@ -933,6 +947,12 @@ def _apply_session_reliability_metrics(
             source_classification=bucket["source_classification"],
             session_count=len(bucket["sessions"]),
             engineer_count=len(bucket["engineers"]),
+            total_call_count=bucket["total_call_count"],
+            avg_calls_per_session=(
+                round(bucket["total_call_count"] / len(bucket["sessions"]), 1)
+                if bucket["sessions"]
+                else 0.0
+            ),
             friction_session_count=len(bucket["friction_sessions"]),
             friction_session_rate=(
                 round(len(bucket["friction_sessions"]) / len(bucket["sessions"]), 3)
@@ -958,6 +978,7 @@ def _apply_session_reliability_metrics(
                 if bucket["sessions"]
                 else None
             ),
+            compound_reliability_rate=_compound_reliability_rate(bucket),
             abandonment_rate=(
                 round(len(bucket["abandoned_sessions"]) / len(bucket["sessions"]), 3)
                 if bucket["sessions"]

tests/test_maturity.py

@@ -476,6 +476,12 @@ def test_maturity_builds_toolchain_reliability_view(
     s1 = seeded_maturity_data["s1"]
     s2 = seeded_maturity_data["s2"]
 
+    db_session.add_all(
+        [
+            ToolUsage(session_id=s1.id, tool_name="Bash", call_count=2),
+            ToolUsage(session_id=s2.id, tool_name="Bash", call_count=2),
+        ]
+    )
     db_session.add_all(
         [
             SessionFacets(
@@ -530,18 +536,29 @@ def test_maturity_builds_toolchain_reliability_view(
     assert github["failure_session_rate"] == 1.0
     assert github["recovery_rate"] == 1.0
     assert github["success_rate"] == 1.0
+    assert github["total_call_count"] == 3
+    assert github["avg_calls_per_session"] == 3.0
+    assert github["compound_reliability_rate"] == 1.0
     assert github["abandonment_rate"] == 0.0
     assert github["top_friction_types"] == ["tool_error", "timeout"]
 
     read_tool = reliability_rows[("built_in_tool", "Read")]
     assert read_tool["session_count"] == 2
     assert read_tool["engineer_count"] == 2
+    assert read_tool["total_call_count"] == 30
+    assert read_tool["avg_calls_per_session"] == 15.0
     assert read_tool["friction_session_rate"] == 1.0
     assert read_tool["failure_session_rate"] == 0.5
     assert read_tool["recovery_rate"] == 0.5
     assert read_tool["success_rate"] == 0.5
+    assert read_tool["compound_reliability_rate"] == 0.63
     assert read_tool["abandonment_rate"] == 0.5
 
+    bash_tool = reliability_rows[("built_in_tool", "Bash")]
+    assert bash_tool["success_rate"] == 0.5
+    assert bash_tool["avg_calls_per_session"] == 2.0
+    assert bash_tool["compound_reliability_rate"] == 0.031
+
 
 def test_maturity_builds_delegation_patterns(
     client, admin_headers, seeded_maturity_data, db_session
@@ -654,6 +671,8 @@ def test_maturity_toolchain_reliability_deduplicates_duplicate_tool_rows(
         if row["surface_type"] == "built_in_tool" and row["identifier"] == "Read"
     )
     assert read_tool["session_count"] == 2
+    assert read_tool["total_call_count"] == 37
+    assert read_tool["avg_calls_per_session"] == 18.5
     assert read_tool["friction_session_count"] == 2
     assert read_tool["avg_recovery_steps"] == 2.0