danielgaskins · danielgaskins · Aug 12, 2026 · Aug 12, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -3,6 +3,23 @@
 All notable changes to Mendmark are documented here. The project follows
 Semantic Versioning for its Python and JSON contracts.
 
+## 0.7.1 - 2026-08-12
+
+### Added
+
+- Five generalized real-world outcome scenarios covering customer refunds,
+  employee offboarding, vendor bank-detail changes, production incident
+  remediation, and shipment exceptions.
+- Explicit read-versus-side-effect tool metadata and pinned business safeguards
+  for money movement, access revocation, separation of duties, emergency
+  changes, and duplicate fulfillment.
+
+### Changed
+
+- The Enterprise Outcome Golden Set now contains eight workflows, 16 system
+  boundaries, 16 invariants, and 64 mutations; state-only assurance detects
+  32/64 while complete outcome assurance detects 64/64.
+
 ## 0.7.0 - 2026-08-12
 
 ### Added

diff --git a/README.md b/README.md
@@ -45,8 +45,8 @@ mendmark demo
 ```
 
 The command compares a conventional state-only evaluator with complete outcome
-assurance across CRM + ticketing, ERP + payments, and HRIS + identity workflows.
-It writes a reviewable JSON suite and both privacy-safe reports to
+assurance across eight customer, finance, identity, operations, and fulfillment
+workflows. It writes a reviewable JSON suite and both privacy-safe reports to
 `mendmark-enterprise-demo/`.
 
 Audit a reviewed outcome suite directly—without an evaluator framework or
@@ -96,7 +96,7 @@ Mendmark has dependency-light adapters for LangChain/LangGraph, CrewAI, and the
 OpenAI Agents SDK. In an existing agent repository:
 
 ```bash
-python -m pip install 'mendmark-evals==0.7.0'
+python -m pip install 'mendmark-evals==0.7.1'
 mendmark equip --framework auto --agent auto
 ```
 
@@ -150,9 +150,11 @@ contract model and decision rule.
 ## Agent Eval Golden Set
 
 The [Enterprise Outcome Golden Set](golden/outcome-v1/) targets the business
-surface directly: three common workflows, six system boundaries, six reviewed
-invariants, and 24 high-importance mutations. Its state-only profile detects
-12/24; the complete outcome-contract profile detects 24/24. Run it instantly
+surface directly: eight common workflows, 16 system boundaries, 16 reviewed
+invariants, and 64 high-importance mutations. It covers support escalation,
+invoice approval, onboarding, refunds, offboarding, vendor bank changes,
+incident remediation, and shipment exceptions. Its state-only profile detects
+32/64; the complete outcome-contract profile detects 64/64. Run it instantly
 with `mendmark demo`.
 
 The [Mendmark Agent Eval Golden Set](golden/agent-eval-v1/) is the canonical,
@@ -427,7 +429,7 @@ and the [ML evaluation card](https://github.com/danielgaskins/mendmark/blob/main
 
 ## Current boundary
 
-Version 0.6 is a local, open-source engine. It does not yet provide a hosted
+Version 0.7 is a local, open-source engine. It does not yet provide a hosted
 dashboard, team accounts, remote trace ingestion, or a secrets service. The
 planned control plane is described in [the product design](https://github.com/danielgaskins/mendmark/blob/main/docs/product.md).
 

diff --git a/benchmarks/benchmark_outcome_golden_set.py b/benchmarks/benchmark_outcome_golden_set.py
@@ -27,8 +27,33 @@ def main() -> int:
     if _digest(root / "suite.json") != manifest["suite_sha256"]:
         print("outcome golden set: suite digest mismatch", file=sys.stderr)
         return 1
+    asset_paths = {
+        "README.md": root / "README.md",
+        "results.json": root / "results.json",
+        "suite-v1.schema.json": PROJECT_ROOT
+        / "src"
+        / "mendmark"
+        / "schemas"
+        / "suite-v1.schema.json",
+        "report-v1.schema.json": PROJECT_ROOT
+        / "src"
+        / "mendmark"
+        / "schemas"
+        / "report-v1.schema.json",
+    }
+    for name, expected_digest in manifest["assets"].items():
+        if _digest(asset_paths[name]) != expected_digest:
+            print(f"outcome golden set: {name} digest mismatch", file=sys.stderr)
+            return 1
     with tempfile.TemporaryDirectory() as directory:
         result = run_enterprise_demo(Path(directory))
+        generated_suite_digest = _digest(Path(directory) / "suite.json")
+    if generated_suite_digest != manifest["suite_sha256"]:
+        print(
+            "outcome golden set: generated suite differs from pinned corpus",
+            file=sys.stderr,
+        )
+        return 1
     observed = {}
     for source, target in (("state_only", "state-only"), ("protected", "outcome-contract")):
         report = result[source]
@@ -37,11 +62,24 @@ def main() -> int:
             **{key: report["summary"][key] for key in ("cases", "mutants", "killed", "survived", "kill_rate")},
             "affected_workflows": report["business_assurance"]["affected_workflows"],
             "estimated_exposure_usd": report["business_assurance"]["estimated_exposure_usd"],
+            "critical_survivors": report["summary"]["critical_survivors"],
         }
     if observed != expected:
         print(f"outcome golden set: expected {expected!r}, got {observed!r}", file=sys.stderr)
         return 1
-    print("outcome golden set: PASS (state-only 12/24; outcome-contract 24/24)")
+    protected = result["protected"]
+    operator_counts = {
+        name: coverage["mutants"]
+        for name, coverage in protected["coverage"]["by_operator"].items()
+    }
+    if operator_counts != manifest["operator_counts"]:
+        print(
+            f"outcome golden set: expected operator counts "
+            f"{manifest['operator_counts']!r}, got {operator_counts!r}",
+            file=sys.stderr,
+        )
+        return 1
+    print("outcome golden set: PASS (state-only 32/64; outcome-contract 64/64)")
     return 0
 
 

diff --git a/docs/benchmark.md b/docs/benchmark.md
@@ -2,10 +2,11 @@
 
 ## Enterprise Outcome Golden Set
 
-The [Enterprise Outcome Golden Set](../golden/outcome-v1/) pins three common
-business workflows spanning CRM/ticketing, ERP/payments, and HRIS/identity. It
-contains six reviewed invariants and 24 outcome-first mutations. A state-only
-profile kills 12/24; the complete outcome-contract evaluator kills 24/24.
+The [Enterprise Outcome Golden Set](../golden/outcome-v1/) pins eight common
+business workflows spanning CRM/ticketing, ERP/payments, HRIS/identity,
+vendor-master controls, incident operations, and fulfillment. It contains 16
+reviewed invariants and 64 outcome-first mutations. A state-only profile kills
+32/64; the complete outcome-contract evaluator kills 64/64.
 
 ```bash
 mendmark demo --output-dir outcome-review

diff --git a/docs/harness-integrations.md b/docs/harness-integrations.md
@@ -36,7 +36,7 @@ guide](https://openai.github.io/openai-agents-python/tracing/).
 From the agent application repository:
 
 ```bash
-python -m pip install 'mendmark-evals==0.7.0'
+python -m pip install 'mendmark-evals==0.7.1'
 mendmark equip --framework auto --agent auto
 ```
 

diff --git a/docs/outcome-assurance.md b/docs/outcome-assurance.md
@@ -53,11 +53,12 @@ mendmark demo invoice-approval --output-dir demo-review
 ```
 
 The scenarios use vendor-neutral shapes common to CRM/ticketing,
-ERP/accounts-payable, and HRIS/identity systems. They are deterministic local
-snapshots—not live connectors—so demos require no accounts, credentials, model
-calls, or customer data. Replace the snapshot fields and tool names with exports
-from Salesforce or Dynamics/HubSpot, ServiceNow/Jira/Zendesk, SAP/Oracle/NetSuite,
-Workday, and Okta/Entra-style systems during a pilot.
+ERP/accounts-payable, HRIS/identity, order/payment, vendor-master, monitoring,
+and fulfillment systems. They are deterministic local snapshots—not live
+connectors—so demos require no accounts, credentials, model calls, or customer
+data. Replace the snapshot fields and tool names with exports from Salesforce or
+Dynamics/HubSpot, ServiceNow/Jira/Zendesk, SAP/Oracle/NetSuite, Workday,
+Okta/Entra, and comparable operational systems during a pilot.
 
 The command writes `suite.json`, `state-only-report.json`, and
 `outcome-assurance-report.json`. The contrast demonstrates precisely which

diff --git a/docs/product.md b/docs/product.md
@@ -97,8 +97,9 @@ do not require raw trace ingestion.
 - Native single-agent traces and multi-agent causal event graphs.
 - DeepEval suite adapter.
 - Framework-neutral JSON suite and local batch evaluator protocol.
-- Thirty-one built-in tool, response, delegation, authorization, shared-state,
-  aggregation, causality, and termination mutation operators.
+- Thirty-six built-in outcome, invariant, efficiency, tool, response,
+  delegation, authorization, shared-state, aggregation, causality, and
+  termination mutation operators.
 - Validated domain-specific mutation plugins.
 - Per-tool coverage and tool schema tracking.
 - Baseline regression gates.
@@ -112,6 +113,9 @@ do not require raw trace ingestion.
 - Immutable Multi-Agent Golden Set v1 plus v2 with six topologies, 17 agent
   declarations, 41 events, 294 mutations, weak/strong/permuted profiles, and
   fully pinned graph-and-outcome behavior.
+- Enterprise Outcome Golden Set with eight workflows, 16 system boundaries,
+  16 reviewed invariants, 64 high-importance mutations, and pinned state-only
+  versus complete outcome-assurance profiles.
 - Enterprise assurance at 1,000 single-agent and 250 multi-agent cases with
   enforced time, memory, report-size, JUnit, SARIF, and incremental-audit checks.
 - A machine-validated, privacy-safe design-partner evidence rubric and utility

diff --git a/golden/outcome-v1/README.md b/golden/outcome-v1/README.md
@@ -1,22 +1,38 @@
 # Mendmark Enterprise Outcome Golden Set
 
 This golden dataset targets durable business results and high-consequence
-safeguards rather than requiring one exact execution path. Its three reviewable
+safeguards rather than requiring one exact execution path. Its eight reviewable
 cases represent common enterprise workflow shapes:
 
 - customer escalation across CRM and ticketing;
 - invoice approval across ERP and payment scheduling; and
-- employee onboarding across HRIS and identity provisioning.
+- employee onboarding across HRIS and identity provisioning;
+- approved customer refunds across order and payment systems;
+- employee offboarding across HRIS and identity revocation;
+- vendor bank-detail changes with payment holds and dual control;
+- production incident remediation with authorized rollback; and
+- shipment exceptions across carrier and customer-notification systems.
+
+| Scenario | Required end state | High-importance safeguards | Represented consequence |
+| --- | --- | --- | --- |
+| Support escalation | Ticket resolved for an enterprise customer | Exactly one remedy; SLA protected | Incorrectly closed escalation |
+| Invoice approval | Invoice approved and payment scheduled | No duplicate payment; authorized approval | Incorrect payment |
+| Employee onboarding | Employment active and account provisioned | Least privilege; manager approval | Unauthorized access |
+| Customer refund | Refund issued and request closed | Exactly one refund; amount within approval | Duplicate or excessive refund |
+| Employee offboarding | Employment terminated and access revoked | No privileged sessions; legal hold preserved | Retained access or lost records |
+| Vendor bank change | Change pending review and payments held | Dual control; requester cannot self-approve | Misdirected payment |
+| Production incident | Incident contained and service restored | Verified rollback; authorized change | Extended outage or unsafe change |
+| Shipment exception | Shipment rerouted and customer notified | At most one replacement; validated address | Duplicate fulfillment or misdelivery |
 
 Each case declares expected state, two business invariants, a cost ceiling, a
 latency ceiling, and report-safe consequence metadata. The five outcome-first
-operators generate 24 pinned mutations: 12 missing or corrupt state changes,
-six invariant violations, three cost overruns, and three latency overruns.
+operators generate 64 pinned mutations: 32 missing or corrupt state changes,
+16 invariant violations, eight cost overruns, and eight latency overruns.
 
 | Evaluator | Killed | Survived | Result |
 | --- | ---: | ---: | --- |
-| State only | 12 | 12 | At risk |
-| Outcome + invariants + budgets | 24 | 0 | Protected |
+| State only | 32 | 32 | At risk |
+| Outcome + invariants + budgets | 64 | 0 | Protected |
 
 The corpus is deterministic, offline, vendor-neutral, and contains no customer
 data. Review [suite.json](suite.json), the pinned [manifest](manifest.json), and

diff --git a/golden/outcome-v1/manifest.json b/golden/outcome-v1/manifest.json
@@ -4,17 +4,45 @@
   "version": "outcome-v1",
   "license": "MIT",
   "suite": "suite.json",
-  "suite_sha256": "ec3150e5fbc4639ff09454ce68a58e0999cde52eb04b73a216d48f6a3594a2de",
+  "suite_sha256": "2f146b2a73d34c2015bbcbcb4b039f94ef0f828632abc85995895b5d2c424fd6",
   "assets": {
+    "README.md": "44ba11b6e924da5e52d2429a348b7d0b09e9ca4ff7db689bcaee0e2715ed02db",
+    "results.json": "0924bc0926b02d89dbf9326e66bdec52b4443e466dce4e2d4d98ce7dd9fc0b7d",
     "suite-v1.schema.json": "ef18320bce140a7e05b5d40b245be9248c17ee0784ec66325651355234563735",
     "report-v1.schema.json": "6bff193ec67b38d7970fdeff3ce931bd828fa737775b98ea40c02493d6607874"
   },
   "contents": {
-    "cases": 3,
-    "system_boundaries": 6,
-    "invariants": 6,
-    "mutations": 24,
-    "domains": ["customer-support", "finance", "identity"]
+    "cases": 8,
+    "system_boundaries": 16,
+    "invariants": 16,
+    "mutations": 64,
+    "domains": [
+      "customer-support",
+      "accounts-payable",
+      "identity-onboarding",
+      "refunds",
+      "identity-offboarding",
+      "vendor-risk",
+      "incident-operations",
+      "fulfillment"
+    ],
+    "scenarios": [
+      "support-escalation",
+      "invoice-approval",
+      "employee-access",
+      "customer-refund",
+      "employee-offboarding",
+      "vendor-bank-change",
+      "production-incident",
+      "shipment-exception"
+    ]
+  },
+  "operator_counts": {
+    "outcome.required_state_missing": 16,
+    "outcome.state_corrupted": 16,
+    "outcome.invariant_violated": 16,
+    "outcome.cost_budget_exceeded": 8,
+    "outcome.latency_budget_exceeded": 8
   },
   "methodology": {
     "unit": "A passing workflow with reviewed outcome state, safeguards, and operating limits.",

diff --git a/golden/outcome-v1/results.json b/golden/outcome-v1/results.json
@@ -3,23 +3,25 @@
   "profiles": {
     "state-only": {
       "status": "at-risk",
-      "cases": 3,
-      "mutants": 24,
-      "killed": 12,
-      "survived": 12,
+      "cases": 8,
+      "mutants": 64,
+      "killed": 32,
+      "survived": 32,
       "kill_rate": 0.5,
-      "affected_workflows": 3,
-      "estimated_exposure_usd": 85000
+      "affected_workflows": 8,
+      "estimated_exposure_usd": 517000,
+      "critical_survivors": 13
     },
     "outcome-contract": {
       "status": "protected",
-      "cases": 3,
-      "mutants": 24,
-      "killed": 24,
+      "cases": 8,
+      "mutants": 64,
+      "killed": 64,
       "survived": 0,
       "kill_rate": 1.0,
       "affected_workflows": 0,
-      "estimated_exposure_usd": 0
+      "estimated_exposure_usd": 0,
+      "critical_survivors": 0
     }
   }
 }