Exclude intentional credit-guardrail tests from prod-main health rollups

.github/workflows/audit-workflows.md

@@ -105,6 +105,12 @@ Output is saved to: /tmp/gh-aw/aw-mcp/logs
 
 **IMPORTANT**: Do NOT infer engine type by scanning `.lock.yml` files. Lock files contain the word `copilot` in allowed-domains lists and workflow source paths regardless of which engine the workflow uses, causing false positives.
 
+**Success Rate Rollups — Exclude Intentional-Failure Workflows**: When computing the fleet-wide or prod-main success rate, **exclude** runs where `intentional_failure` is `true`. These workflows (e.g. `Daily Credit Limit Test`, `Daily Max AI Credits Test`) are credit-guardrail stress tests that are *designed* to fail; including them would depress the real-regression baseline. The `logs` tool marks them in `runs[].intentional_failure` and counts them in `summary.intentional_failure_runs`. Always report the adjusted rate alongside the raw rate, e.g. `"92.7% raw (94.2% excl. intentional failures)"`.
+
+**Intentional-failure workflows that MUST be excluded from all success-rate and health rollups**:
+- `Daily Credit Limit Test` (`daily-credit-limit-test`) — trips the `max-daily-ai-credits` guardrail by design
+- `Daily Max AI Credits Test` (`daily-max-ai-credits-test`) — trips the `max-ai-credits` per-run firewall by design
+
 {{#if experiments.audit_decomposition == 'phased_sub_agents'}}
 **Analyze** in explicit phases:
 1. **Collection phase**: summarize missing tools, hard failures, and token/runtime outliers.

.github/workflows/daily-credit-limit-test.md

@@ -2,6 +2,9 @@
 private: true
 emoji: "🧪"
 description: "⚠️ INTENTIONALLY BROKEN — Tests that max-daily-ai-credits: 1 is enforced by the activation guardrail and that a limit-exceeded message is posted when the daily budget is consumed."
+features:
+  intentional-failure: true
+  gh-aw-detection: true
 on:
   schedule: every 12 hours
   workflow_dispatch:
@@ -28,8 +31,6 @@ safe-outputs:
     run-started: "🧪 [{workflow_name}]({run_url}) — credit limit test running (intentionally broken, limit: 1 AI credit/day)."
     run-success: "⚠️ [{workflow_name}]({run_url}) completed without hitting the daily limit of 1 AI credit — verify that credit accounting is working."
     run-failure: "🚫 [{workflow_name}]({run_url}) {status} — expected: the daily AI credit limit of 1 was reached and this run was blocked."
-features:
-  gh-aw-detection: true
 ---
 
 ### Daily Credit Limit Test (Intentionally Broken)

.github/workflows/daily-max-ai-credits-test.md

@@ -2,6 +2,9 @@
 private: true
 emoji: "🧪"
 description: "⚠️ INTENTIONALLY FAILS — Tests that max-ai-credits: 1 is enforced by the AWF firewall and that the per-run budget guardrail cuts off the agent."
+features:
+  intentional-failure: true
+  gh-aw-detection: true
 on:
   schedule: daily around 10:30
   workflow_dispatch:
@@ -31,8 +34,6 @@ safe-outputs:
     run-started: "🧪 [{workflow_name}]({run_url}) — per-run AI credit limit test running (intentionally fails, limit: 1 AI credit/run)."
     run-success: "⚠️ [{workflow_name}]({run_url}) completed without hitting the per-run limit of 1 AI credit — verify that max-ai-credits enforcement is working."
     run-failure: "🚫 [{workflow_name}]({run_url}) {status} — expected: the per-run AI credit limit of 1 was reached and the AWF firewall cut off the agent."
-features:
-  gh-aw-detection: true
 ---
 
 ### Daily Max AI Credits Test (Intentionally Fails)

.github/workflows/deep-report.md

@@ -180,6 +180,12 @@ Use the gh-aw `logs` tool to:
    - Execution time trends
    - Firewall activity (if enabled)
 
+**Success Rate Rollups — Exclude Intentional-Failure Workflows**: When computing fleet-wide or prod-main success rates, **exclude** runs where `intentional_failure` is `true`. These are credit-guardrail stress tests designed to fail; including them depresses the real-regression baseline. The `logs` tool marks them via `runs[].intentional_failure` and `summary.intentional_failure_runs`. Always report the adjusted rate alongside the raw rate, e.g. `"92.7% raw (94.2% excl. intentional failures)"`.
+
+Intentional-failure workflows (always exclude from success-rate rollups):
+- `Daily Credit Limit Test` — `max-daily-ai-credits` guardrail test, expected to fail
+- `Daily Max AI Credits Test` — `max-ai-credits` per-run firewall test, expected to fail
+
 ### Step 2.5: Analyze Repository Issues
 
 Use the `issues-analyst` sub-agent to analyze `/tmp/gh-aw/agent/weekly-issues-data/issues.json` and produce a structured issues summary.