🔄 Cross-Run Diff Template

📌 Template Instructions: Copy to analysis/daily/{date}/{article-type}-run{N}/intelligence/cross-run-diff.md. Bayesian delta vs. the previous run of the same article type: what changed in data and assessment. See methodologies/per-artifact-methodologies.md §cross-run-diff.

🎯 Purpose: Track evolution of political assessment across runs. For each prior-run finding, apply Bayesian update based on new evidence to produce posterior assessment.

---

📋 Document Metadata

Field	Value
Report ID	[REQUIRED: XRD-YYYY-MM-DD-runNN]
Current Run	[REQUIRED: {type}-run{N}, YYYY-MM-DD]
Prior Run	[REQUIRED: {type}-run{N-1}, YYYY-MM-DD]
Days Between Runs	[REQUIRED: # days]
Confidence	[REQUIRED: 🟢/🟡/🔴]

---

1️⃣ Prior-Run Header

Prior run: [REQUIRED: path to analysis/daily/{date}/{type}-run{N-1}/]
Prior run date: [REQUIRED: YYYY-MM-DD]
Prior run number: [REQUIRED: run{N-1}]

Key findings from prior run:

1. [REQUIRED: finding 1 from prior synthesis-summary]
2. [REQUIRED: finding 2]
3. [REQUIRED: finding 3]
4. [REQUIRED: finding 4]
5. [REQUIRED: finding 5]

Key scores from prior run:

Metric	Prior Value
Overall risk score	[#]
Coalition stability	[%]
Pipeline health	[#/10]
Top threat severity	[#]

---

2️⃣ Data Delta

New documents since prior run: [REQUIRED: count]
New votes since prior run: [REQUIRED: count]
New events since prior run: [REQUIRED: count]
New procedures since prior run: [REQUIRED: count]

Significant new data:

Data Type	ID	Title/Topic	Significance
[REQUIRED: Document/Vote/Event/Procedure]	[ID]	[REQUIRED]	[REQUIRED: one-line explanation]
[REQUIRED]	[ID]	[REQUIRED]	[REQUIRED]
[REQUIRED]	[ID]	[REQUIRED]	[REQUIRED]

MCP endpoint refresh status:

Endpoint	Records Prior Run	Records This Run	Delta
get_plenary_sessions	[#]	[#]	[±#]
get_voting_records	[#]	[#]	[±#]
get_adopted_texts	[#]	[#]	[±#]
get_procedures	[#]	[#]	[±#]

---

3️⃣ Assessment Delta — Bayesian Update

%%{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
    PRIOR1[Prior Finding 1:<br/>[SUMMARY]] -->|new evidence| EV1{Evidence<br/>Class}
    EV1 -->|supporting| POST1A[Posterior:<br/>Confirmed/Upgraded]
    EV1 -->|contradicting| POST1B[Posterior:<br/>Downgraded/Reversed]
    EV1 -->|orthogonal| POST1C[Posterior:<br/>Unchanged]
    
    style PRIOR1 fill:#1565C0,color:#ffffff
    style POST1A fill:#2E7D32,color:#ffffff
    style POST1B fill:#D32F2F,color:#ffffff
    style POST1C fill:#FFC107,color:#000000

Per-finding Bayesian updates:

Finding 1: [REQUIRED: prior finding text]

Prior confidence: [REQUIRED: 🟢/🟡/🔴]
Status: [REQUIRED: ✅ Confirmed / ⬆️ Upgraded / ⬇️ Downgraded / ❌ Reversed]
Posterior confidence: [REQUIRED: 🟢/🟡/🔴]

New evidence:

Evidence Source	Evidence Class	Description
[REQUIRED: document/vote/event ID]	[Supporting / Contradicting / Orthogonal]	[REQUIRED: one-line]
[REQUIRED]	[...]	[REQUIRED]

Bayesian posterior:

[REQUIRED: ≥80 words explaining how new evidence updates the prior assessment. If supporting → why confidence rises or claim is confirmed. If contradicting → what changed and why. If orthogonal → why finding remains valid despite new data. Cite political-risk-methodology.md §Bayesian Update.]

---

Finding 2: [REQUIRED: prior finding text]

Prior confidence: [🟢/🟡/🔴]
Status: [✅ Confirmed / ⬆️ Upgraded / ⬇️ Downgraded / ❌ Reversed]
Posterior confidence: [🟢/🟡/🔴]

New evidence: [REQUIRED: table as above]

Bayesian posterior: [REQUIRED: ≥80 words]

---

Finding 3: [REQUIRED]

(repeat structure for all top-5 findings from prior run)

---

4️⃣ Confidence Migration

Claims that moved from 🔴 LOW to 🟡 MEDIUM or 🟢 HIGH:

1. [REQUIRED: claim + explanation for upgrade]
2. [REQUIRED]

Claims that moved from 🟢 HIGH to 🟡 MEDIUM or 🔴 LOW:

1. [REQUIRED: claim + explanation for downgrade]
2. [REQUIRED]

Claims that remain at same confidence level but with strengthened evidence:

[REQUIRED: list or note "none"]

---

5️⃣ Open Questions for Next Run

What the next same-type run should investigate:

1. [REQUIRED: specific question or data gap to address in next run]
2. [REQUIRED]
3. [REQUIRED]
4. [REQUIRED]

Recommended follow-up MCP queries:

[REQUIRED: list specific EP MCP tools to call with parameters, or note monitoring triggers]

---

6️⃣ Data Sources

Prior run artifacts read: [REQUIRED: list manifest.json, synthesis-summary.md, risk-matrix.md paths]
EP MCP tools re-queried: [REQUIRED: list which endpoints were called fresh vs. which used cached prior-run data]

---

7️⃣ Confidence Assessment

Overall confidence: [REQUIRED: 🟢 HIGH / 🟡 MEDIUM / 🔴 LOW]

Confidence rationale: [REQUIRED: ≥60 words explaining where Bayesian updates are data-backed vs. expert judgment. Note any gaps in roll-call data, document availability, or MCP reliability.]

---

🛠️ Worked example — diff between two breaking-news runs

Run baseline (2026-04-22 run 01) → current (2026-04-25 run 03)

Element	Prior	Current	Delta	Reason
Lead procedure	2026/0142(COD) AI Act IA	unchanged	—	—
Coalition margin	408 vs 360 (48-seat cushion)	408 vs 360	0	Stable
EPP cohesion	91%	88%	-3 pp	47-MEP defection on industry-pressure amendment
Trilogue ETA	Q3 2026	Q3 2026 (early)	-2 weeks	Council adopted position faster than expected
Risk score (industry softening)	4/10	6/10	+2	New trade-association letters surfaced
Cross-coalition issue clusters	2 (rule-of-law, fiscal rules)	3 (added energy security)	+1	Council winter-package response

Bayesian update narrative: prior-run forecast assigned 65% to "adoption Q3 in original scope"; current run reduces this to 50% on strength of new industry-letter evidence and EPP defection trend, with the residual 50% redistributed: 35% softened scope, 12% delayed, 3% failure.

What changed in confidence: 🟡 → 🟡 (unchanged band), but rationale narrows: more evidence for softening pathway, less weight on adoption within-original-scope.

🚫 Anti-patterns — cross-run-diff failures

Anti-pattern	Why it fails	Correct approach
Diff without prior-run version stamp	Cannot reconcile	Cite prior-run path explicitly
Numerical change without unit	Ambiguity	"+3 pp" / "+2 percentage points" / "-2 weeks"
Bayesian update without prior	Math doesn't compute	State prior probability + posterior
New evidence not tagged	Evidence chain unclear	Mark "NEW since prior run"
Diff that just restates current	Provides no comparison	Each row must show prior + current
Confidence band re-stated without rationale	Stale Pass-1 leftover	Update rationale even if band unchanged
Risk delta without driver	Cannot be acted on	Each delta has named driver
Skipping unchanged elements	Loses audit trail	Stable items show "0 / unchanged"

🎯 EP MCP tool inputs

Tool	Used for
Re-query analyze_coalition_dynamics	Cohesion delta vs prior
Re-query get_voting_records	New RCVs since prior run
Re-query track_legislation	Stage / timeline shift
Read prior-run manifest.json	Anchor for diff
Read prior-run risk-matrix.md	Risk-score baseline
Read prior-run synthesis-summary.md	Narrative baseline

🔗 Controlling methodology cross-references

• ../methodologies/per-artifact-methodologies.md §cross-run-diff
• cross-session-intelligence.md — companion (longer-window momentum)

✅ Stage-C completeness signals

• Line floor: 100 lines
• Prior-run version-stamped (path + run number)
• ≥ 5 element-level diffs (procedure / coalition / risk / etc.)
• Bayesian update narrative present where forecasts revised
• Confidence band + rationale present

---

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