⚙️ Workflow Audit Template

📌 Template Instructions: Copy to analysis/daily/{date}/{article-type}-run{N}/intelligence/workflow-audit.md. The AI agent produces this at the end of the run as a structured self-audit of how the 10-step protocol was executed. See methodologies/per-artifact-methodologies.md §workflow-audit.

🎯 Purpose: Transparent record of workflow execution — which phases ran, which MCP tools were called, which rules of ai-driven-analysis-guide.md were satisfied, and where the run fell short. A downstream reviewer (and the next same-type run) can use this to understand the run's internal quality.

---

📋 Document Metadata

articleType: [REQUIRED]
runId: [REQUIRED]
date: [REQUIRED: YYYY-MM-DD]
analysisPhase: workflow-audit
confidenceLevel: [REQUIRED: HIGH / MEDIUM / LOW]
rulesAudited: [REQUIRED: integer — 11 Core Principles, see §3 below]
complianceRate: [REQUIRED: 0–100]

---

1️⃣ Workflow Execution Summary

%%{init: {"theme":"dark","themeVariables":{"primaryColor":"#1565C0","primaryTextColor":"#ffffff","primaryBorderColor":"#0A3F7F","lineColor":"#90CAF9","secondaryColor":"#2E7D32","secondaryTextColor":"#ffffff","tertiaryColor":"#FF9800","tertiaryTextColor":"#000000","mainBkg":"#1565C0","secondBkg":"#2E7D32","tertiaryBkg":"#FF9800","noteBkgColor":"#FFC107","noteTextColor":"#000000","errorBkgColor":"#D32F2F","fontFamily":"Inter, Helvetica, Arial, sans-serif"}}}%%
flowchart LR
    P0["Phase 0<br/>Health gate"] --> P1["Phase 1<br/>Data collection"]
    P1 --> P2["Phase 2<br/>Editorial context"]
    P2 --> P3["Phase 3<br/>Analysis"]
    P3 --> P4["Phase 4<br/>Significance gate"]
    P4 --> P5["Phase 5<br/>Validation"]
    P5 --> P6["Phase 6<br/>PR creation"]

    style P0 fill:#1565C0,color:#ffffff
    style P1 fill:#1565C0,color:#ffffff
    style P2 fill:#0288D1,color:#ffffff
    style P3 fill:#7B1FA2,color:#ffffff
    style P4 fill:#FF9800,color:#000000
    style P5 fill:#2E7D32,color:#ffffff
    style P6 fill:#2E7D32,color:#ffffff

Phase	Status	Notes
Phase 0 — Health gate	[✅ / ⚠️ / ❌]	[REQUIRED: what gate returned]
Phase 1 — Data collection	[✅ / ⚠️ / ❌]	[REQUIRED: feeds attempted, feeds succeeded]
Phase 2 — Editorial context	[✅ / ⚠️ / ❌]	[REQUIRED]
Phase 3 — Analysis	[✅ / ⚠️ / ❌]	[REQUIRED: artifacts produced]
Phase 4 — Significance gate	[✅ / ⚠️ / ❌]	[REQUIRED: composite score vs. threshold]
Phase 5 — Validation	[✅ / ⚠️ / ❌]	[REQUIRED: validator exit code]
Phase 6 — PR creation	[✅ / ⚠️ / ❌ / ⏳]	[REQUIRED]

---

2️⃣ MCP Tool Call Log

#	Tool	Args (truncated)	Result	Records	Latency
1	[REQUIRED]	[REQUIRED]	[✅/⚠️/❌]	[#]	[ms]
2	[REQUIRED]	[REQUIRED]	[...]	[#]	[ms]
3	[REQUIRED]	[REQUIRED]	[...]	[#]	[ms]

Total MCP calls: [REQUIRED] · Successful: [#] · Degraded: [#] · Failed: [#]

---

3️⃣ Core Principles Compliance (11-principle protocol)

See ai-driven-analysis-guide.md §Core Principles.

#	Principle	Status	Evidence in run
1	Folder isolation	[✅/❌]	[REQUIRED: run dir path]
2	AI does analysis, scripts do HTML	[✅/❌]	[REQUIRED]
3	Methodologies before analysis	[✅/❌]	[REQUIRED: METHODOLOGIES_READ log line]
4	Multi-framework depth	[✅/⚠️/❌]	[REQUIRED: frameworks applied]
5	Always commit analysis	[✅/❌]	[REQUIRED]
6	Article type everywhere	[✅/❌]	[REQUIRED]
7	Two passes, full time budget	[✅/⚠️/❌]	[REQUIRED: Pass 1 min + Pass 2 min]
8	AI-authored headlines + descriptions	[✅/❌/ N/A]	[REQUIRED]
9	Complete data + historical baseline	[✅/⚠️/❌]	[REQUIRED]
10	Read-before-article + footer + ratio + floors	[✅/⚠️/❌]	[REQUIRED: validator output]
11	OSINT / INTOP tradecraft discipline (WEP bands, Admiralty grades, ≥10 SATs in methodology-reflection)	[✅/⚠️/❌]	[REQUIRED: link to methodology-reflection.md §12 SATs applied + Admiralty-graded source list]

Compliance rate: [# principles ✅] / 11 = [%]

---

4️⃣ Artifact Production

Folder	Planned	Produced	Short (< floor)	Notes
intelligence/	[#]	[#]	[#]	[REQUIRED]
classification/	[#]	[#]	[#]	[REQUIRED]
risk-scoring/	[#]	[#]	[#]	[REQUIRED]
threat-assessment/	[#]	[#]	[#]	[REQUIRED]
documents/	[#]	[#]	[#]	[REQUIRED]

---

5️⃣ Time Budget

Step	Target	Actual	Delta
Step 2 — Read methodologies	4–6 min	[REQUIRED]	[± min]
Step 3 — Collect EP data	5–10 min	[REQUIRED]	[± min]
Steps 4–7 — Analysis Pass 1	≥12 min	[REQUIRED]	[± min]
Step 9 — Pass 2	≥8 min	[REQUIRED]	[± min]
Step 10 — Validate + commit	2–5 min	[REQUIRED]	[± min]
Total run time	≥20 min active	[REQUIRED]	[± min]

---

6️⃣ Issues & Deviations

For every deviation from the standard protocol:

Issue 1 — [REQUIRED: short name]

• Symptom: [REQUIRED]
• Root cause: [REQUIRED]
• Workaround in this run: [REQUIRED]
• Next-run recommendation: [REQUIRED]

(repeat per issue; no issues → state "No deviations from protocol observed.")

---

7️⃣ Recommendations for the Next Same-Type Run

1. [REQUIRED: concrete, actionable — e.g. "raise historical-baseline depth to ≥200 lines by fetching 3 additional prior-run comparables"]
2. [REQUIRED]
3. [REQUIRED]

---

8️⃣ EP MCP Tool Inputs (auditor's evidence basis)

EP MCP tool	Used for which audit principle	Notes
get_voting_records	P3 Evidence — RCV citations in artifacts	Auditor cross-checks every RCV ID claimed.
analyze_coalition_dynamics	P4 Coalition arithmetic veracity	Two-window verification.
get_adopted_texts	P3 Evidence — text-citation veracity	Confirms quoted recitals exist.
get_committee_documents	P5 Source diversity	≥1 committee-doc per major artifact.
track_legislation / get_procedures	P6 Legislative-pipeline accuracy	Procedure-status drift between artifacts.
get_speeches	P3 Evidence — quote attribution	Confirms speakers + verbatim.
get_meeting_decisions	P7 Procedural veracity	Bureau / CoP rulings cited correctly.
correlate_intelligence	P11 Cross-tool correlation	Confirms intelligence alerts feed into artifacts.

(Auditor uses these tools directly — does not rely on prior-stage agent outputs.)

---

9️⃣ Worked Pass-1 → Pass-2 Example (60-min breaking-news run audit)

❌ Pass-1 (thin, 17 words):

"Run completed in 58 minutes. All artifacts present. Quality acceptable. No major issues observed during execution."

✅ Pass-2 (compliant, 95 words):

Run 184 (breaking, 2026-04-18) executed in 58:42 (target 60:00). Stage budget actuals: Data 11:30 / Analysis 21:50 / Gate 3:20 / Article 17:40 / PR 4:22. Two principle deviations: P3 Evidence — voting-patterns.md cited RCV 2026/0089(COD) but no get_voting_records tool call recorded in tool-call log (LOW confidence flag missing — Partial); P9 Reflection — methodology-reflection.md documented only 8 of required 10 SATs (Partial). All other 9 principles Pass. Stage-C gate caught both deviations and triggered Pass-2 rewrites of the affected artifacts. Net verdict: CONDITIONAL PASS — recommendations §7 cover both deviations.

---

🔟 11-Core-Principles Compliance Scorecard

#	Principle	Compliance level	Worked observation
P1	BLUF discipline	✅ Pass	All intelligence-tier artifacts open with ≤2-sentence BLUF.
P2	WEP / Admiralty tradecraft	✅ Pass	All forward judgements carry WEP band + horizon.
P3	Evidence sourcing (cite-or-don't-claim)	⚠️ Partial	1 RCV cited without get_voting_records tool-call evidence.
P4	Coalition arithmetic discipline	✅ Pass	Seat shares + cohesion deltas all from analyze_coalition_dynamics.
P5	Source diversity	✅ Pass	9 distinct EP MCP tools used across 25 artifacts.
P6	Legislative-pipeline accuracy	✅ Pass	Procedure-status verified against track_legislation.
P7	Procedural veracity	✅ Pass	Rule numbers cited match published Rules of Procedure.
P8	Read-Before-Write (article cites artifacts)	✅ Pass	All article §§ cite per-section artifacts per 04-article-generation.md §7.1.
P9	Methodology reflection (≥10 SATs)	⚠️ Partial	8 of 10 SATs documented.
P10	Per-artifact depth floors	✅ Pass	All artifacts meet reference-quality-thresholds.json.
P11	Time-budget discipline (≥45 min active work on 60-min budget)	✅ Pass	47 min active per timestamps.

Composite: 9 Pass / 2 Partial / 0 Fail = CONDITIONAL PASS, requires Pass-2 fixes on P3 and P9 before PR.

---

1️⃣1️⃣ Worked 60-Minute Breaking-News Time-Budget Breakdown

Phase	Target	Actual (Run 184)	Delta	Notes
Stage A — Data collection (MCP feeds)	12:00	11:30	-0:30	Within tolerance.
Stage B Pass 1 — Analysis artifacts (initial draft)	14:00	13:20	-0:40	OK.
Stage B Pass 2 — Analysis rewrite + depth	09:00	08:30	-0:30	Below 9 min — auditor flag: rushed.
Stage C — Completeness gate	03:00	03:20	+0:20	Includes the two Partial-principle re-runs.
Stage D Pass 1 — Article draft	11:00	10:50	-0:10	OK.
Stage D Pass 2 — Article rewrite	07:00	06:50	-0:10	OK.
PR creation + final checks	04:00	04:22	+0:22	OK.
Total	60:00	58:42	-1:18	Active work 47:00 (≥45-min floor).

---

1️⃣2️⃣ Anti-patterns — REJECT on Pass-2 Review

#	Banned pattern	Why it fails
1	"All principles Pass" with no per-principle worked observation	Audit must show its working; reviewer cannot verify.
2	Time-budget table with target column only (no actual)	Drift cannot be measured.
3	"No deviations from protocol" claim while >1 artifact below depth floor	Internally inconsistent.
4	Recommendation "tighten Pass-2 review" without specifying artifact + principle	Non-actionable.
5	Audit assigning Pass to P9 with <10 SATs documented	Hard contract; SAT count is checked.
6	Citing tool-calls from get_voting_records while the EP roll-call publication delay >4 weeks (claim impossible)	Auditor must verify upstream feed availability.

---

1️⃣3️⃣ Cross-References — Controlling Methodology

• ../methodologies/ai-driven-analysis-guide.md — § Rules 1–22, the 11 Core Principles being audited.
• ../methodologies/per-artifact-methodologies.md#workflow-audit — construction rules.
• ../methodologies/osint-tradecraft-standards.md — § SAT enumeration (≥10 required for P9 Pass).
• ./methodology-reflection.md — sister artifact; auditor cross-reads.
• scripts/validate-analysis-completeness.js — depth-floor check feeds P10 verdict.
• scripts/lint-prompts.js — banned-pattern check for prompt drift.

---

1️⃣4️⃣ Stage-C Completeness Signals

scripts/validate-analysis-completeness.js checks for this artifact:

Check	Threshold	Source
Line floor	≥100 lines (default)	reference-quality-thresholds.json
Required H2 substrings	"Compliance", "Time-Budget", "Recommendations"	structural contract
Mermaid block	optional (Gantt of stage timings if drift >5 %)	visual contract
Tradecraft markers	Per-principle Pass/Partial/Fail explicit; SAT count documented	ai-driven-analysis-guide.md
Source diversity	≥5 EP MCP tools cross-checked	per-artifact rule
Time-budget arithmetic	Targets sum to 60/120 min; actuals + deltas explicit	template logic

---

Document Control: /analysis/daily/{date}/{type}-run{N}/intelligence/workflow-audit.md · Template v1.2 · Depth floor: 100 lines.