Iran Transition Project Status: Continuously developed and refined — critique welcome
This document describes the analytical methodology used by the Iran Transition Project. It is written for readers familiar with traditional analytical methods (intelligence community structured analysis, academic political science, think tank policy analysis) who want to understand what this project does differently and why.
This methodology is not finished. It is being developed and refined through application to a live, high-stakes analytical problem. Feedback and critique are actively sought — see How to Critique This Methodology at the end of this document.
Traditional analytical frameworks for Iran exhibit recurring failure modes:
-
Wrong interlocutor. Negotiations focus on pragmatist factions while actors with veto power (eschatological hardliners, IRGC economic networks) are neither at the table nor analytically modeled as decision-relevant.
-
Surface legibility bias. The regime's constitutional architecture is treated as decorative, its ideological claims dismissed as rhetoric, and its institutional depth underestimated. This produces analysis that consistently overestimates regime flexibility and underestimates institutional inertia.
-
English-language source monoculture. The vast majority of Western analysis relies on English-language sources, missing the regime's own signaling (KHAMENEI.IR, seminary networks, IRGC-affiliated media) and the analytical signal embedded in Farsi-language institutional discourse.
-
Snapshot analysis. Think tank products and intelligence assessments tend to capture point-in-time snapshots rather than tracking structural dynamics across time. Variables change; the analysis does not update.
This methodology is designed to address these specific failure modes through structured frameworks, explicit epistemic discipline, continuous variable tracking, and multilingual source integration.
The ITB maps the regime's institutional architecture across eight pillars:
| Pillar | Coverage |
|---|---|
| A | Constitutional architecture, IRGC, ideology, parallel society, eschatological faction, coercive doctrine |
| B | Security and military architecture |
| C | Economic structures and sanctions |
| D | International relations and alignment |
| E | Domestic society and demographics |
| F | Transition dynamics and historical cases |
| G | Nuclear program and proliferation |
| H | Information environment and media |
Each pillar contains one or more analytical modules — structured documents with explicit section numbering, cross-references, and epistemic tags on every claim. The ITB is not a static reference; it is updated as new information changes the analytical picture.
The ISA identifies structural vulnerabilities and analytical hazards:
-
Traps: Circular logic structures that catch policymakers. Example: a nuclear deal requires trust → trust requires verification → verification requires access → access requires a deal. Each trap documents its mechanism, circular structure, resolution path, and historical parallels.
-
Observations: "So-what" findings that emerge from cross-referencing ITB modules. Each observation states what is true (diagnosis) and what it means for planning (strategic implication).
-
Scenarios: Modeled transition and conflict pathways with probability ranges, leading indicators, and cross-referenced variables.
The project tracks 86 variables across five tables: stock (slow-moving structural conditions), flow (dynamic indicators), threshold (trigger points), positive optionality (opportunities), and normalization quality (governance readiness). Variables are updated with explicit trend indicators and confidence bands.
Open questions are registered, prioritized (1-4), and tracked. When a gap is filled, the session and method of resolution are recorded. This creates an auditable trail of what the project knows, what it does not know, and what it is actively trying to learn.
Every analytical claim carries an epistemic tag:
| Tag | Meaning |
|---|---|
[Fact] |
Directly verifiable, multiple independent sources |
[Inference] |
Reasoned from established facts, reasoning chain stated |
[Uncertain] |
Single source, contested, or extrapolated |
[Speculation] |
Acknowledged hypothesis, forward projection |
Tags are paired with confidence bands (High, Medium, Low) that indicate the overall strength of evidence supporting a conclusion.
Why this matters: In traditional analysis, a reader cannot distinguish between a conclusion the analyst is confident about and one that rests on thin evidence without reading the full sourcing appendix. Inline tagging makes evidence quality visible at the point of consumption.
The project uses a five-tier source taxonomy:
- Regime primary sources (KHAMENEI.IR, official institutional output)
- Human rights monitoring organizations (HRANA, Amnesty, CHRI)
- Academic Iran studies (peer-reviewed research)
- Diaspora investigative outlets (with transparent sourcing methodology)
- Unverified / single-source
Sources at tiers 1-3 receive priority weighting. Wikipedia is excluded as a primary or corroborating source for Iran content due to documented state-affiliated manipulation.
Regime-controlled media is not treated as factual reporting. It is treated as a signaling channel — what the regime chooses to say, to whom, and when, is itself analytical data. The content is claims-only; the decision to publish is the signal.
The project explicitly models taqiyyah (religiously sanctioned dissimulation) as an institutional capability, not a cultural stereotype. Four specific failure modes are documented where Western analysts misread regime behavior because they do not account for this institutionalized deception doctrine. This is framed with strict anti-Islamophobic discipline — the analytical point is institutional, not civilizational.
This project uses Claude (Anthropic) as a research assistant. The AI's role is documented publicly through two instruction files:
- CLAUDE_CHAT_INSTRUCTIONS.md governs analytical sessions: epistemic discipline, source standards, module activation, stakeholder analysis
- CLAUDE_CODE_INSTRUCTIONS.md governs repository maintenance: YAML operations, schema validation, build pipeline
- Accelerates multilingual source research (Farsi, Arabic, English)
- Maintains structured data consistency across 86 variables, 57 gaps, and 22 content modules
- Drafts analytical content under explicit epistemic constraints
- Validates cross-references and schema compliance
- Builds publication-ready output from structured data
- AI output is not treated as a source. All claims require independent sourcing per the epistemic framework above.
- AI does not make analytical judgments autonomously. The human analyst sets the analytical direction, evaluates findings, and decides what conclusions the evidence supports.
- AI-generated summaries of other AI systems (e.g., Gemini) are treated as potentially contaminated and require independent verification.
Analytical transparency requires disclosing methods. The instruction files document exactly what constraints the AI operates under, what it is told to prioritize, and how its output is validated. If the AI introduces bias or methodological weakness, the instructions that produced that bias are available for anyone to examine.
| Dimension | Traditional IC / Think Tank | This Project |
|---|---|---|
| Update cycle | Point-in-time assessments | Continuous variable tracking with session-based updates |
| Evidence transparency | Sourcing appendix or footnotes | Inline epistemic tags on every claim |
| Source language | Predominantly English | Multilingual mandate (Farsi, Arabic, English) |
| Structural modeling | Narrative-driven | Structured data (YAML) with schema validation |
| Cross-reference integrity | Manual, error-prone | Automated build pipeline with validation |
| Accessibility | Classified or paywalled | Open source (CC BY-SA 4.0) |
| Methodology disclosure | Rarely published | Fully public, open to critique |
| Factional position | Often aligned with policy preference | Explicitly neutral — the test is structural, not preferential |
What traditional methods do better: Classified intelligence access, human source networks, satellite imagery analysis, signals intelligence. This project cannot replace those capabilities. It can structure and validate the analytical framework that interprets their output.
-
No classified sources. The project relies entirely on open sources. This creates blind spots on operational military details, internal regime communications, and real-time intelligence.
-
Single analyst. The current framework reflects one analyst's judgment. Peer review and methodological critique are essential correctives — this is an active invitation, not a disclaimer.
-
AI contamination risk. Despite safeguards, AI assistance introduces the risk of plausible-sounding but unsourced claims. The epistemic tagging system is designed to catch this, but no system is foolproof.
-
Regime deception. The project explicitly models taqiyyah, but modeling deception does not guarantee detecting it in every case. The framework can identify structural conditions where deception is likely; it cannot always identify the specific deception.
-
English-language output. Despite multilingual source integration, all output is in English. This limits accessibility to Farsi-speaking audiences who might provide the most valuable feedback.
This methodology is being developed in the open specifically to invite critique. The most valuable feedback addresses:
-
Structural blind spots. Where does the framework systematically miss something? Not individual factual errors, but categories of information or analysis that the structure itself excludes.
-
Epistemic overconfidence. Where do confidence bands seem too high for the evidence available? Where does an
[Inference]tag mask what should be[Uncertain]? -
Source hierarchy problems. Is the five-tier taxonomy appropriate? Are there source categories that deserve higher or lower weighting?
-
Framework transferability. Could this methodology be applied to other opaque regimes (North Korea, Myanmar, Eritrea)? What would need to change?
-
AI methodology risks. Does the AI-assisted workflow introduce biases that the current safeguards do not catch?
How to submit critique:
- Public: GitHub Discussions (Feedback & Critique category)
- Private: irantransitionproject.org/submit
- Email: admin@irantransitionproject.org
All methodological critiques that identify genuine weaknesses will be acknowledged and addressed in the framework, with attribution if the submitter requests it.