source-triangulation.md

May 15, 2026 · View on GitHub

🔍 Source Triangulation Methodology

📊 Formalised 4-Step Fallback Ladder with Confidence-Label Rules
🎯 Primary Sources · Cross-Source Checks · Structural Inference · KB Integration · Confidence Labelling

📋 Document Owner: CEO | 📄 Version: 1.0 | 📅 Last Updated: 2026-05-14 (UTC) 🔄 Review Cycle: Quarterly | ⏰ Next Review: 2026-08-14 🏢 Owner: Hack23 AB (Org.nr 5595347807) | 🏷️ Classification: Public

🎯 Purpose

This document codifies the 4-step source-triangulation fallback ladder that every analytic artifact follows when primary European Parliament MCP data is unavailable, incomplete, or degraded. It defines:

When each fallback step activates.
How to label confidence at each step using Admiralty grades and 🟢/🟡/🔴 markers.
What disclosure to include in the artifact and in mcp-reliability-audit.md.

This methodology is non-optional: every artifact that reaches Step 2 or beyond must include an explicit [TRIANGULATION-STEP: N] marker on every affected claim, naming the fallback method used and its Admiralty grade.

Relationship to other documents: This file documents the what and when of fallbacks. Admiralty grading rules live in osint-tradecraft-standards.md §2. Confidence-marker semantics live in confidence-calibration.md. IMF-specific fallback rules live in imf/cross-source-triangulation.md.

🛡️ ISMS Policy Alignment

Policy	Relevance
Hack23 AI_Policy §3	Evidence and provenance — every claim must carry a source grade regardless of fallback step
Information Security Policy §6	Evidence handling — degraded-mode evidence must be disclosed, not silently upgraded
Secure Development Policy	Analytic fallbacks are equivalent to SDLC graceful-degradation paths: they must be documented and tested

📐 Ladder Overview

%%{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 TD
    S0["🎯 Analytical claim<br/>to be supported"] --> Q0{EP MCP primary<br/>data available?}

    Q0 -- "✅ Yes — A1/A2" --> STEP1["Step 1 · Direct EP MCP<br/>🟢 HIGH confidence<br/>Admiralty A1–A2"]
    Q0 -- "⚠️ Partial / degraded" --> Q1{Cross-source<br/>triangulation<br/>possible?}

    Q1 -- "✅ Yes" --> STEP2["Step 2 · Cross-Source Triangulation<br/>🟢/🟡 HIGH-to-MEDIUM<br/>Admiralty A3–B2"]
    Q1 -- "❌ No" --> Q2{Historical analogy<br/>or structural<br/>inference available?}

    Q2 -- "✅ Yes" --> STEP3["Step 3 · Historical Analogy<br/>/ Structural Inference<br/>🟡 MEDIUM confidence<br/>Admiralty B3–C3"]
    Q2 -- "❌ No" --> STEP4["Step 4 · KB Integration<br/>/ Institutional-Framework Analysis<br/>🔴 LOW confidence<br/>Admiralty C4–F6"]

    STEP1 --> DISCLOSE["Disclose step<br/>in artifact header"]
    STEP2 --> DISCLOSE
    STEP3 --> DISCLOSE
    STEP4 --> DISCLOSE

    style STEP1 fill:#2E7D32,color:#fff
    style STEP2 fill:#2E7D32,color:#fff
    style STEP3 fill:#FF9800,color:#000
    style STEP4 fill:#D32F2F,color:#fff
    style DISCLOSE fill:#1565C0,color:#fff

1️⃣ Step 1 — Direct EP MCP Data (Preferred)

Activates when: The European Parliament MCP Server returns a complete, current-year, directly-queryable response for the artifact's required data fields.

Admiralty grade: A1 (primary EP record, confirmed by multiple tool calls) or A2 (single EP primary record, uncontested).

Confidence marker: 🟢 HIGH.

Disclosure in artifact: None required beyond the standard mcp-reliability-audit.md endpoint ledger entry.

Qualifying EP MCP Tools for Step 1

Tool	Source type	Grade
`get_adopted_texts`, `get_procedures` (with document ID)	Direct EP primary record	A1
`get_voting_records` (with plenary session ID)	Direct EP plenary record	A1
`get_mep_details`, `get_committee_info`	Direct EP institutional record	A1
`analyze_coalition_dynamics`, `assess_mep_influence`	EP-derived aggregate	A2
`get_adopted_texts_feed` with `freshness === "augmented"`	EP augmented feed	A2

When Step 1 is Incomplete

Step 1 is considered incomplete when:

The MCP tool returns 4xx/5xx/timeout with no fallback.
The tool returns 0 results and the mcp-reliability-audit.md cannot confirm this is a genuine null-result.
dataQualityWarnings contains FRESHNESS_FALLBACK_FAILED.
The data age is > 21 days for a breaking/weekly run.

Proceed to Step 2 when any of the above apply.

2️⃣ Step 2 — Cross-Source Triangulation

Activates when: Step 1 is incomplete and at least one of the following authoritative secondary sources can corroborate or supplement the claim.

Admiralty grade: A3 (multiple B-graded sources + partial A-grade corroboration) or B2 (single well-established secondary source).

Confidence marker: 🟢 HIGH if two independent A/B-grade sources agree; 🟡 MEDIUM if only one secondary source available.

Mandatory disclosure in artifact:

[TRIANGULATION-STEP: 2 — Cross-source triangulation]
Primary MCP data: [describe what was unavailable/degraded]
Secondary source(s) used: [EUR-Lex / Eurostat / ECB / IMF / official MS publication / ...]
Admiralty grade of claim: [A3 / B2 / ...]
Confidence: [🟢 HIGH / 🟡 MEDIUM]

Approved Step-2 Cross-Sources

Source	Scope	Trust tier	Admiralty grade
EUR-Lex	Legislative texts, procedure status	Official EU publication	A3
Eurostat	Statistical indicators, demographic data	Official EU statistics	A2
ECB SDW	Monetary, inflation (HICP), banking data	Official EU/EA monetary	A2
IMF SDMX REST	Economic context (GDP, fiscal, trade)	Global primary (sole authority for economic claims)	A1 for economic claims
World Bank	Non-economic context (health, education, demographics)	International primary	A2
Official national-government publications	Member-state positions, national statistics	Official MS publication	A3
EP press releases, plenary records (web)	Committee decisions, group positions	Official EP secondary	B2
EU Council web portal	Council positions, presidency texts	Official Council secondary	B2

IMF Cross-Source Discipline

When the EP MCP economic context is unavailable, IMF remains the sole authoritative economic source. Follow the IMF fallback ladder documented in imf/cross-source-triangulation.md:

Try IMF SDMX REST → imf-fetch-data tool.
If unavailable, use IMF WEO April [current year] figures from training data — must be labelled as data-vintage="IMF WEO April YYYY (training-data fallback)".
Never substitute non-IMF economic projections for IMF WEO even when Step 2 fails — proceed to Step 3 for economic claims only if IMF training-data figures are also unusable.

3️⃣ Step 3 — Historical Analogy / Structural Inference

Activates when: Step 2 is unavailable or returns no usable data, and the analyst can derive the claim from:

Documented EP historical precedents (prior terms, analogous votes, precedent coalition configurations).
Structural inference from known institutional rules (Rules of Procedure, Treaty articles, committee mandates).
Established political-science literature on EP dynamics.

Admiralty grade: B3 (pattern consistent with historical evidence, no primary contradiction) or C3 (academic / policy-institute source, not directly confirmed by EP data).

Confidence marker: 🟡 MEDIUM. Never 🟢 unless the structural inference is a formal rule (e.g. "absolute majority = 361 seats" is C1, not B3).

Mandatory disclosure in artifact:

[TRIANGULATION-STEP: 3 — Historical analogy / structural inference]
Primary/secondary MCP data: [describe what was unavailable]
Inference basis: [historical precedent / Rules of Procedure / political-science literature]
Source(s): [procedure ID / RoP article / academic citation]
Admiralty grade of claim: [B3 / C3 / ...]
Confidence: 🟡 MEDIUM

Valid Step-3 Bases

Basis	Example	Admiralty
EP10 seat distribution (fixed institutional fact)	"EPP holds 188 of 720 seats"	A1 (official EP record)
EP Rules of Procedure (formal rule)	"Absolute majority = 361"	A1
Prior-term RCV pattern (same issue type)	"EP10 follows EP9 pattern on CAP reform"	B3
CJEU jurisprudence (established)	"Treaty Article 7 requires 4/5 majority"	A1
Academic political-science consensus	"Hooghe/Marks trilateral EP coalition model"	C2
Policy-institute analysis (Bruegel, CEPS, ECFR)	"ECFR: ECR likely to obstruct Rule-of-Law"	C3

What Step 3 Cannot Support

Vote-count predictions for specific, unnamed votes not yet taken.
MEP-level position claims not backed by any prior public vote or public statement.
Economic projections beyond the IMF training-data vintage.

4️⃣ Step 4 — KB Integration / Institutional-Framework Analysis

Activates when: Steps 1–3 are all unavailable or insufficient, and the analyst must rely exclusively on:

General knowledge-base understanding of EU institutional mechanics.
Established institutional-framework patterns (e.g. "the Commission typically publishes…", "the Council presidency rotates…").
Widely-known EP procedural conventions that do not require current data.

Admiralty grade: C4 (structural claim, limited corroboration) or F6 (institutional convention, no primary source this run).

Confidence marker: 🔴 LOW. Mandatory in all Step-4 claims.

Mandatory disclosure in artifact:

[TRIANGULATION-STEP: 4 — KB integration / institutional-framework analysis]
All upstream steps: [describe why Steps 1–3 were unavailable]
Basis: [general EU institutional mechanics / KB understanding]
Admiralty grade: [C4 / F6]
Confidence: 🔴 LOW
CAUTION: This claim is based on institutional-framework inference only.
It must NOT be cited in headline judgements or executive-brief conclusions.

Step-4 Scope Limits

Step-4 claims are admissible only in:

"What we do not know" caveat sections.
Background framing paragraphs with explicit LOW-confidence labelling.
Monitoring plans ("watch for X to confirm or refute this claim").

Step-4 claims are inadmissible in:

Executive finding / BLUF paragraphs.
Top-5 findings tables in synthesis-summary.md.
Any section in scenario-forecast.md labelled as the "Base" or "Most-likely" scenario.
economic-context.md quantitative claims.

🔖 Per-Artifact Triangulation Trigger Table

The table below records the minimum step reached before a claim may be included in each artifact, and the maximum confidence label allowed.

Artifact	Max confidence without Step 1	Step-4 claims allowed?	Notes
`intelligence/synthesis-summary.md`	🟢 HIGH (Step 2, two A/B sources) / 🟡 MEDIUM (single source)	❌ Never	Top-5 findings require Step 1 or 2
`intelligence/scenario-forecast.md`	🟢 HIGH (Step 2, two A/B sources) / 🟡 MEDIUM (Step 2 single / Step 3)	❌ Scenarios only	Background only for Step 3/4
`intelligence/economic-context.md`	🟢 HIGH (Step 2 IMF corroborated) / 🟡 MEDIUM (single IMF source)	❌ Never	IMF fallback ladder mandatory
`intelligence/coalition-dynamics.md`	🟡 MEDIUM (Step 3)	❌ Never	Group composition from Step 1; cohesion from Step 3
`intelligence/threat-model.md`	🟡 MEDIUM (Step 3)	⚠️ Background	Label LOW; no headline threat from Step 4
`intelligence/stakeholder-map.md`	🟢 HIGH (Step 2, two A/B sources) / 🟡 MEDIUM (Step 2 single / Step 3)	⚠️ Background	Never cite Step 4 for power/alignment scores
`intelligence/pestle-analysis.md`	🟡 MEDIUM (Step 3)	⚠️ Background	Economic dimension requires Step 1/2
`risk-scoring/risk-matrix.md`	🟢 HIGH (Step 2, two A/B sources) / 🟡 MEDIUM (Step 2 single / Step 3)	❌ Never	Risk likelihood/impact requires ≥Step 2
`extended/voter-segmentation.md`	🟡 MEDIUM (Step 3)	⚠️ Background	Eurobarometer first; see `voter-segmentation-methodology.md`
`executive-brief.md`	🟢 HIGH (Step 2, two A/B sources) / 🟡 MEDIUM (single source)	❌ Never	Decision-brief quality floor

🔗 Disclosure Chain

Every artifact that uses Step 2, 3, or 4 must:

Include inline [TRIANGULATION-STEP: N] markers on each affected claim.
Log every fallback endpoint attempt in intelligence/mcp-reliability-audit.md with the step number reached and the resulting Admiralty grade.
Record the aggregate step distribution in intelligence/methodology-reflection.md §Data sources and provenance table, using a new "Max triangulation step reached" column.

📋 Quick-Reference Checklist

Run this checklist before submitting any artifact to Stage C:

Step 1 exhausted first: Every MCP tool relevant to this claim was attempted.
Step markers present: Every Step-2/3/4 claim carries a [TRIANGULATION-STEP: N] marker.
Admiralty grades correct: Grades match the step table above, not inflated.
Confidence markers correct: 🟢 for Step 1, or Step 2 when two independent A/B-grade sources agree; 🟡 for Step 2 (single source) or Step 3; 🔴 for Step 4.
Step-4 placement: Step-4 claims appear only in "what we do not know" / background / monitoring sections.
mcp-reliability-audit.md updated: Step reached and grade recorded for every degraded endpoint.
methodology-reflection.md updated: "Max triangulation step reached" logged.

confidence-calibration.md — unified 🟢/🟡/🔴 + WEP + Admiralty mapping
osint-tradecraft-standards.md §2 — Admiralty grading rules
voter-segmentation-methodology.md — Eurobarometer fallback rules for voter-segmentation artifacts
imf/cross-source-triangulation.md — IMF-specific triangulation matrix
ai-driven-analysis-guide.md §Step 6 — where triangulation is applied in the 10-step protocol
per-artifact-methodologies.md — per-artifact required SAT list

Document Control:

Path: analysis/methodologies/source-triangulation.md
Classification: Public
Version: 1.0 — Initial codification of the 4-step fallback ladder derived from methodology-reflection.md patterns observed in 2026-05-14 run set.
ISMS Reference: Hack23/ISMS-PUBLIC AI Policy §3 (Evidence and provenance)