A cascade rarely stays in one domain.
A cyberattack becomes a hospital failure. A drought becomes a food-price shock. A lab leak becomes a governance crisis. One AI release becomes a labour, political, and welfare rearrangement.
CREF lets you trace how that happens — what started the cascade, what amplified it, where it could have been stopped, and which systems absorbed the shock. The stakes are not theoretical: pandemics, blackouts, food shortages, financial contagion, infrastructure failures. Decisions made under cascade pressure shape lives, livelihoods, and policy.
Start where you stand.
Three ways in. Pick one.
Understand a cascade
Start with one of 336 worked cases — Fukushima, COVID supply shock, AI governance — and see how the cascade reads as a sequence of five roles.
Cascade bursts — what they look like in motion
Cross-domain risk doesn't walk politely from one system to the next. It bursts. A single trigger propagates through sci-tech, governance, economics, social dynamics, ecological systems, and health on overlapping timescales. The framework documents 35 such bursts with named causal pathways, lag distributions, and evidence anchors.
First action: open Cases, click any case card, then switch to Case → CREF grammar to see prose-to-grammar.
Build your own cascade
Pick a trigger. The tool suggests evidence-backed downstream effects. Mark intervention points. Score the cascade.
Build a chain — pick where to break it (show illustration)
Build a chain — pick where to break it
Each studio session starts from one trigger and grows outward through evidence-backed edges. You decide which branches to keep, which to mark as intervention points, and where the cascade scores high enough to merit a counter-measure.
First action: open Cascade Studio, pick a trigger from the left tabs, then add a next effect from the suggestions panel.
Browse factors
296 risk factors with full definitions, anchors, and cross-domain edges. Filter by domain, evidence tier, or role.
Audit the evidence
1,345 CEM edge records, 1,371 CEM edge records, Q-scores, source labels, year of first appearance. Filter for cross-domain or feedback edges.
Anticipate
Frontier AI, synthetic biology, pandemics, climate. Prospective cases for scenario work and foresight planning.
Cascade thinking is industry-specific.
The framework is general; the failure modes are not. CREF maps directly onto the cascade patterns that matter in your sector. The patterns below are illustrative — the full sector decompositions, with domain-weighted edge sets, mitigation libraries, and risk-register integration, ship in the commercial Cascade Logic layer.
Financial services
Liquidity → counterparty → sovereignSubprime → interbank stress → capital-rule constraint → CB liquidity → sovereign absorber. The 2008 cascade reads as four edges and finishes in fiscal capacity.
Critical infrastructure
Cyber → grid → hospital → care delayA single ransomware event in a managed-service vendor propagates through grid SCADA, hospital admissions, and elective-care backlogs within 72 hours. The dampener is redundancy; the absorber is mutual-aid.
Healthcare & biosecurity
Pathogen → supply → care → trustOutbreak → diagnostic supply shortage → ICU capacity constraint → public-trust amplifier → governance dampener. The 2020 cascade names the failure modes; CREF lets you stress-test the 2026 ones.
Five roles. Five shapes.
Every cascade reads as a sequence of five roles. Each role has a distinct physical shape here — a visual vocabulary for talking about cascade structure. Shape reuse across Cases, Glossary, and Play is an in-progress consolidation.
Crucially: a factor's role is relational. The same factor can be a Trigger in one cascade and an Absorber in another. The role label in the factor index is the modal role — the role that factor most often plays — not the only role available to it.
Every cascade reads as a chain of these five roles. A factor can play any role; role is relational, not categorical.
From the CREF paper (Figure 2 caption): Any factor from any force category can play any cascade function role; function assignment is determined by cascade context, not by category membership. The Explorer follows this rule. Where the factor index shows a single role, that is the modal role; the relational view (Find leverage) lets you ask which role a factor plays in a specific edge.
Four ways into the framework.
Read a worked case, look something up, map the whole landscape, or build your own cascade. Start wherever your question is.
Read the danger, find the levers
Read the system, then act on it. The cascades the framework documents — ranked by present danger, the candidate levers that could dampen them, and who could pull which lever. Six surfaces: cascade pressure, the danger index, documented correctives, actor-bound moves, what has worked, and a held priorities page.
Working from your own risk register rather than the evidence base? Customize the framework to your organization →
Danger Index
Every impact-resolved case, ranked by cascade pressure — a 0–100 structural reading of present danger (unmitigated severity, eased by how tractable mitigation is). This is not just the case list re-sorted: it adds the mitigation layer the plain Cases view omits — where pressure concentrates by region, and what collective action could plausibly ease. Diagnostic, not a forecast.
01 World & regional ranking
The real-world counterpart to the Game's Cascade Pressure meter. Each impact-resolved case is scored 0–100 from its unmitigated severity, eased by how tractable mitigation is — the same logic that moves the doom meter when a cascade is left unchecked versus actively broken. Diagnostic, not predictive: a structural reading of present danger, not a forecast.
02 Action meter — what collective action could ease
The real cascade-pressure score is fixed and evidence-based. For cases where collective action plausibly has a lever — ongoing or emerging crises that are at least partly tractable — pick a participation level to see a hypothetical relief overlay: a thought experiment, owner-gated, that never overwrites the real number. How is this computed?
03 All impact-resolved cases
The full ranked corpus. Each card opens its complete profile, deep-linked to its mitigation section. Filter above; page size below.
Correctives
Beneficial loops the evidence base documents — corrections already operating in the system. Grouped into the ten families of correction; open a family to see the historical loops behind it. A Corrective is a Move that history already made.
Moves
The ten correction families as intervention types, each bound to the actor classes that can carry it out. A Move is a mitigation Factor an actor could pull at a cascade location — Factor × Actor × Location. Leverage estimates are structural readings of where capability sits, not predictions that an actor will act or succeed.
Actor-class leverage is an illustrative structural estimate from the CREF mitigation typology (v0.1). It says where the lever plausibly sits — not that the actor will act, nor that action will succeed. This surface enumerates intervention types; it is not a ranked, costed set of recommendations.
What Worked
Documented interventions where a named actor acted on a cascade and the effect was recorded. Authority here comes from named provenance and editorial-board sign-off, not from a graph engine. Only board-validated cases appear; submissions stay hidden until reviewed. Why aren't these computed priorities?
Submit a documented intervention
Your submission is not published. It produces a JSON record you can send to the editorial board. A case appears in the explorer only after board sign-off (it is then re-loaded as validated JSON). Provenance and conflict-of-interest disclosure are required.
Pick a trigger.
48 factor codes act as triggers across 336 cases. Click any chip to narrow the case list below; click Clear to return to all cases.
Cases — worked events, decomposed.
★ Top reference cases curated picks — click to expand
More filters
1 · Event
2 · Causal
3 · Modal
4 · Mechanism
5 · Governance
Factors across six domains.
Click any code for the formal definition, evidence tier, manifest year and event, sub-domain, anchor citations, and the cascade neighbourhood. Total at this release: 235 factors. The CEB grows as new cases are mapped.
Build a cascade. Reason about it.
Pick a trigger. The tool suggests evidence-backed downstream effects. Mark roles. Score the cascade. Identify intervention points. Export a scenario you can use in a paper, workshop, or briefing.
- 1Trigger
- 2Propagation
- 3Roles
- 4Intervention
- 5Export
Build a cascade in five moves.
Start with a trigger. The tool will suggest downstream effects from the evidence base. Your goal is to explain how a localized disruption migrates across domains — and where it could be stopped.
How one factor pushes another.
| Edge (source factor → target factor) | Type | Year | Q-score | Audit |
|---|
Cascade bursts.
Cascade Bursts
Co-firing patterns: one source factor firing into multiple targets under a single documented trigger. Click a tile to expand. Bursts complement the binary edge table — edges remain the unit of evidence.How the framework is actually built.
Methodological provenance for reviewer-grade use. What CREF claims, what it does not, and how to read its evidence base.
What CREF is
The Cascading Risk Effects Framework (CREF) is a mid-range theory of how systemic risk propagates across domains via mechanism roles. It is not a forecast model and not a probabilistic risk register. It is a grammar — a controlled vocabulary plus a small set of compositional rules — that lets analysts annotate, compare, and audit causal claims about cross-domain risk propagation.
The companion Cascade Evidence Base (CEB) instantiates the grammar against 296 cross-domain risk factors and 1,371 evidence-anchored cascade edges drawn from 336 historical, current, and prospective cases.
Factor selection
Factors enter the CEB when they meet three criteria: (a) recurrent appearance across at least two of the case-mapping corpora; (b) traceable to a manifest event with a verifiable historical anchor (year + event description); (c) operational definition stable enough to support coding by independent reviewers.
- Factors are stratified by domain (STI, GOV, ECO, SOC, ENV, HEA) and sub-domain (37 second-level categories).
- Each factor carries an evidence tier (T0-Platinum through T4-Provisional) reflecting causal identification quality.
- Total factor count at this release is 296. This count will grow as CEB updates ingest additional cases.
Mechanism taxonomy
Each edge in the CEB is annotated with one of six mechanism codes, plus an explicit OTHER bucket and an uncoded slot for in-progress audit:
MCH.DIR— Direct propagation. Standard cause-effect transmission. (532 edges)MCH.DIS— Displacement. Risk shifts from one site to another without dissipating. (64)MCH.TRN— Transmission. Cross-network spread along physical or informational corridors. (77)MCH.CON— Convergence. Multiple upstream factors compound into a single downstream pressure. (105)MCH.ACC— Acceleration. Existing trend speeds up; not a new pathway but a velocity change. (81)MCH.FBK— Feedback. Downstream effect loops back to amplify or suppress upstream. (73)MCH.OTHER— Other / not yet normalized; carries a plain-language mechanism but does not fit the six canonical classes. (23)- Uncoded — Mechanism field blank, pending audit. (12)
Counts reconcile to 1048 edges (six canonical classes + OTHER + uncoded).
Evidence tiers & Q-score
Edge records carry a source-quality score (Q) on a 0–3 scale: 3 = peer-reviewed primary source or government statistical office; 2 = peer-reviewed secondary, major think-tank, or quality journalism; 1 = grey literature or industry report; 0 = anecdotal or unsourced placeholder pending audit.
At this release, 760 of 1048 edges carry a Q-score; 801 carry an explicit source label; 759 carry a named historical example; 604 carry a year. Missing fields are intentionally rendered blank rather than inferred.
Edge deduplication and cascade-variant grouping
The CEB carries 1,345 CEM edge records, of which 909 are canonical and 139 are corroborating evidence rows retained as dup_of for provenance. A corroborating row has identical source factor, target factor, and mechanism code as its canonical, but contributes an independent evidence source — typically created when multiple coders ingested the same cascade pathway from different sources. Duplicates are not deleted; the richest record (most filled fields, highest source-quality score) is kept canonical, and the rest carry a dup_of pointer.
Separately, 83 source-target pairs carry multiple records with different mechanism codes. These are not duplicates but legitimate cascade variants — the same pair propagating via different causal pathways (e.g. CLI→FIN with 14 documented instances across MCH.DIR via stranded assets, MCH.CON via insurance market convergence, MCH.FBK via climate finance feedback, MCH.TRN via supply-chain physical risk transmission). Each variant carries a cascade_variant_group_id so the UI can group when needed. By default the Evidence pane shows canonical edges only; the filter exposes duplicates and variants for audit work.
Known limitations
- Western/English bias. Source corpus over-represents English-language scholarship and OECD-context cases.
- Heterogeneous edge granularity. Some edges represent decades-long pathways (DEM ⇒ INE), others represent single-incident transmissions (CYB → FIN, 2017). Cascade length is annotated but not normalized.
- Q-score is judgmental. A single coder assigned most scores; inter-rater reliability work is pending.
- Prospective cases are illustrative. The fifteen prospective cases — including AI Governance, Synthetic Biology, Climate & Energy, Pandemic Preparedness, and Frontier AI — are explicitly marked illustrative, not predictive.
- The CEB is not exhaustive. 245 factors is the count at this release. The framework is open; new cases enter via the case-mapping protocol.
Advanced formalization — CCG-1
The Causation Chain Grammar (CCG-1) is the formal-methods companion to CREF, expressing cascade structure as typed sequences over the role vocabulary {[T], [×], [⊓], [↓], [∎]} with explicit modal operators for velocity and reversibility. CCG-1 enables comparison of cascade structures across cases by translating verbal causal claims into compositional algebra.
CCG-1 is intentionally separated from the main CEB UI to keep the framework accessible to non-formal users. It is documented in the CREF technical appendix and addressed by the CREF-Lang DSL parser (separate commercial thread).
Roles are relative. {{role_relativity_note}}
Composite scores
Four per-factor scores ride on top of the 296-factor / 1,371-edge corpus: Volume (raw connectivity), Severity (worst incident edge), Dominant Role (which cascade function the factor most often performs in the present corpus — not a fixed property; the same factor performs different roles in different cascades and at different stages), and Domain Rank (percentile within domain). They are diagnostic, not predictive. A high Volume tells you a factor is heavily connected in the present corpus; it does not tell you that factor will dominate a future cascade.
Volume_Composite
Total CEM edges incident on the factor: in_degree + out_degree. Range 0–194 across the current corpus. Ties broken by out_degree when needed.
What it captures. The factor's structural centrality in the cascade graph. Five factors sit at the apex: FIN (194), CLI (127), WAR (109), STS (108), HEA (103). These are the load-bearing nodes — perturbations propagate through them most often.
What it does not capture. Edge quality. A factor with 100 SR-Mined edges scores the same as one with 100 T1-Established edges. Read alongside Severity.
Volume_Z_Domain (v8.5.8): z-score of Volume_Composite within domain. Lets you ask "is this factor a hub or a periphery within its own domain?" without conflating cross-domain absolute differences. Read alongside Domain_Rank.
Severity_Composite (v8.5.8 hardened)
Three severity figures travel together. The headline number is Severity_Top3_Mean — mean of the three highest incident-edge severities. It tolerates a single irreversible edge without exploding from it. Severity_Max (the v8.5.7 figure) is preserved as an audit field. Severity_Median is the central-tendency reference for distributional reading.
Per-edge severity is Reversibility_score × max(Tier_score, 1):
- Reversibility map: IRREVERSIBLE=4 · HYSTERETIC=3 · REVERSIBLE=2 · UNKNOWN/blank=1.
- Tier map: T1-Established=3 · T2-Convergent=2 · T2-Plausible / SR-Mined / Strategic=1 · UNKNOWN/blank=0.
Severity_Provenance classifies the reliability of each factor's severity score: multi_edge_corroborated (≥5 edges, ≥2 high-severity; 75/296), sufficient (≥3 edges; 69/245), fragile_single_or_pair (1–2 edges; 96/296), or no_edges (5/296). Treat fragile and orphan scores as provisional pending hardening agenda H9/H10 backfill.
Why the change: in v8.5.7, CYB and NUC reached Severity=8 from a single edge (CEM223, CYB→NUC, IRREVERSIBLE × T2-Convergent). The figure was technically correct under a max-over-edges rule but optically misleading — it equated one well-evidenced irreversible pathway with a corroborated cascade pattern. Top3_Mean tells you whether the high severity has structural support across multiple edges. For CYB the hardened figure is 4.67, for NUC 5.00. The single CEM223 edge still surfaces in Severity_Max for audit.
Dominant_Role
Which of the five CREF cascade functions the factor most frequently performs on its outgoing edges, inferred from Mechanism_Code:
- Trigger ← MCH.SHK
- Amplifier ← MCH.ACC · MCH.AMP · MCH.FBK
- Dampener ← MCH.CBT · MCH.DIS
- Constraint ← MCH.STR · MCH.DEP · MCH.CON
- Absorber ← MCH.ENA · MCH.AUG · MCH.INT · MCH.TRN
- Role-neutral ← MCH.DIR (generic directional; 585 edges, 44% of corpus; excluded from role tally, retained in Volume)
Reported as Dominant_Role plus Dominant_Role_Share (count of dominant-role edges / total role-classifiable outgoing edges). Factors with zero role-classifiable outgoing edges receive Dominant_Role = "INSUFFICIENT" — 113 of 245 factors in this release. Most are downstream-only or low-out-degree factors; the count drops as MCH.DIR is progressively reclassified in v8.5.8+.
Distribution among classified factors (v1, v8.5.8): Amplifier 42 · Constraint 40 · Absorber 29 · Dampener 21 · Trigger 0. The zero-Trigger result reflected current MCH.SHK coding density at v8.5.8, not absence of trigger factors in the corpus.
Dominant_Role_v2 (v8.5.9): MCH.DIR heuristic reclassification pass applied. 141 of 585 MCH.DIR edges reclassified via mechanism-prose pattern matching (shock/amplifier/dampener/constraint/transmission language), with per-edge confidence flag (high / medium / low) retained for audit on the CEM. New distribution: Amplifier 53 · Constraint 47 · Dampener 37 · Absorber 22 · Trigger 5 · INSUFFICIENT 81 (down from 113). CYB shifts from Amplifier to Trigger (13 shock edges hidden under MCH.DIR; PAN→FIN, CYB→FIN, WAR→ENE patterns were the largest reclassification cluster). Dominant_Role_v1 preserved alongside v2 in the workbook; widget surfaces v2 as headline. Edges with direct-causation-only mechanism prose (71 edges) remain MCH.DIR by design — they encode pure pass-through without role function.
Domain_Rank
Percentile rank of Volume_Composite within Domain. 0–100. Ties resolved by average rank (standard percentile-rank convention). Allows like-with-like comparison without conflating cross-domain factors of different absolute connectivity.
Use Domain_Rank when comparing factors within a single domain (e.g. ranking governance factors against each other). Use Volume_Composite when comparing structural centrality across domains.
What these scores are not
- Not probabilities. A factor with Volume=194 is not "more likely" to trigger a cascade; it is more connected in the corpus.
- Not predictions. Scores describe the present corpus. Tomorrow's cascade may run through factors that today look peripheral.
- Not normalised across CEB releases. Volume rises mechanically as the corpus expands. Compare scores only within a single CEB version.
- Not a substitute for reading the edges. Severity=8 on CYB rests on one edge. The score points you to the edge; it does not replace examining it.
What is not claimed
- CREF is not a probabilistic forecast. Cases marked "prospective" are illustrative pathways, not predictions.
- The CEB is not a complete enumeration of cross-domain risks. It is an open evidence base with documented gaps.
- Q-scores are quality indicators, not certainties. A Q=3 edge is well-sourced; it is not "true with probability 1".
- Cascade length and edge weights do not commute with magnitude estimates of underlying events.
How to cite
Cascade Rehearsal 2075 — workshop game.
Mitigate cascades · Facilitator-led · 9 cities · 1–3 players.
1. Where are you working this turn?
Tap a city to travel there, then draw your event.🎲 Dice set how far you travel (auto-rolled on draw) · 🃏 Event is the crisis you mitigate.
2. Event drawn
Pick up to 3 mitigation cards whose domain affinities together cover the event profile.3. Your hand
Tap cards to select. Up to three per event. Tap again to deselect.Basic: lenient adjudication; the two NPCs cover the single biggest need. After you submit, the NPCs complete the round.
The Facilitator adjudicates each play. Submitting locks in your hand and reveals the verdict.
Cascade Logic for your firm.
CREF is the open framework. Cascade Logic is the commercial layer on top of it: custom factor sets per sector, governance dampener mapping, intervention libraries, and integration with the risk register you already run.
Three decision moments cascade thinking changes
These are the three settings where cascade mapping is most immediately actionable.
Board-level systemic-risk briefing
Your board reads a top-10 risks list. Cascade Logic reads a cascade map: which factors converge, where the absorbers are, which two-edge interactions are unmonitored. The output is a six-page briefing that names the firm's three highest-stake cascades, the dampener inventory, and the gap.
Deliverable: custom CREF map (40–60 factors, ~120 edges) + briefing pack.
Engagement: 8–10 weeks. Fixed fee.
Supplier-cascade stress test
Tier-1 suppliers are mapped. Tier-2 are sampled. Tier-3 is invisible. Cascade Logic walks the cascade backwards from the failure mode that would hurt the firm most, identifies the choke points two and three edges away, and tells you which dependencies are concentrated in regions, vendors, or chokepoints you do not currently monitor.
Deliverable: sector-specific cascade graph + chokepoint inventory + monitoring playbook.
Engagement: 6–12 weeks depending on supplier set size.
Regulatory scenario rehearsal
A regulator publishes a new disclosure rule, a capital surcharge, or an export control. The firm's compliance team produces a memo. Cascade Logic produces a scenario rehearsal: the cascade from the rule through the firm, the two non-obvious mitigations, and the three internal stakeholders who will be surprised.
Deliverable: rehearsal pack + facilitated workshop using the End of the World game on your scenario.
Engagement: 4–6 weeks per rehearsal cycle.
What Cascade Logic adds on top of CREF
CREF is open. The framework, the 296-factor base, the 1,371 edges, the 336 cases, the methodology — all of it is published. The commercial layer is what you cannot get from the open base alone.
Custom factor sets per sector
CREF's 245 factors are general. A bank does not need EVA (evapotranspiration); an insurer does. A semiconductor firm needs eight factors CREF does not yet carry. Cascade Logic ships sector packs: 80–120 factors per sector, normalised against CREF, with edges weighted for that industry's failure modes.
Governance dampener mapping
The open CREF identifies dampener-role factors. Cascade Logic maps each one to a specific governance lever in your firm: the policy, the committee, the budget line, the named owner. The output is the inventory you wish your second line of defence had.
Intervention library
The mitigation deck in the open Explorer is 15 cards, illustrative. The commercial deck is ~200 interventions, each with cost band, lead time, dependency list, and historical precedent. Filterable by domain, by factor, by mechanism.
Risk-register integration
Your ERM team already has a risk register. Cascade Logic does not replace it. It augments: each register line gets a cascade context, an unmonitored-neighbours flag, and a feedback-loop attestation. Integrations exist for ServiceNow, Archer, MetricStream, and CSV.
Live event-cascade monitoring
For clients with continuous-monitoring needs: a feed of mapped real-world cascades (the 2024 Red Sea disruption, the 2024 critical-minerals decoupling, the H5N1 dairy cluster) reaches your team within 48 hours of triggering edges, with the firm-specific cascade graph re-scored.
Workshop facilitation & training
Cascade Rehearsal 2075 (the workshop game in this Explorer) is the public version. The commercial version runs on your firm's custom cascade map. Half-day, full-day, and two-day formats. Used for board offsites, ERM team training, and tabletop exercises.
How a paid engagement starts
No procurement gauntlet for the first step. The first conversation is short and free. The first deliverable is paid but small.
-
130-minute scoping call. You describe the cascade you actually worry about. We name two existing CREF cascades that resemble it, and one that does not. End of call: you decide whether the framework speaks to your problem.
-
2Cascade mapping sprint (2 weeks, fixed fee). We build a draft cascade map for the problem you named — 25–40 factors, 50–80 edges, four reference comparators from CREF. You see how the framework holds up against your facts.
-
3Engagement scoped on the draft. Now you know what the system does on your case. The engagement (briefing pack, stress test, rehearsal cycle, full sector pack — whichever fits) gets scoped on real ground, not on a sales deck.
Start the conversation
The 30-minute scoping call is free. After that, you know enough to decide whether to spend money. If you are not sure your problem fits, send the rough shape of it and we will say so honestly.
Cascade Logic is delivered through Yegii. CREF is published independently. The framework is and remains open; the commercial layer above it is not.
The science behind CREF.
CREF is one researcher's attempt to make cross-domain cascade risk legible and audit-able. This pane shows who built the framework, what it is grounded in, and which adjacent traditions it sits alongside. For the full evidence base behind each factor and edge, see the Bibliography.
Books (8)
Selected peer-reviewed papers
- Duncan, I., Lin, J., Gill, L., Undheim, T. A. (2026). Cascading global catastrophic risks toward 2075: how optimism and expertise shape perceptions of systemic threats. Futures, 178: 103757. doi:10.1016/j.futures.2025.103757
- Malik, M. Q., Undheim, T. A. (2025). AI infrastructure for trust and learning in education: the emergence of the 'learning provenance' concept. NEAIS 2025 Proceedings, Paper 2.
- Reuel, A., Soder, L., Bucknall, B., Undheim, T. A. (2024). Technical research and talent is needed for effective AI governance. Proceedings of the 41st International Conference on Machine Learning (ICML), 235: 42543–42557.
- Undheim, T. A. (2024). An interdisciplinary review of systemic risk factors leading up to existential risks. Progress in Disaster Science, doi:10.1016/j.pdisas.2024.100326
- Undheim, T. A., Ahmad, T. (2024). Quantitative scenarios for cascading risks in AI, climate, synthetic bio, and financial markets by 2075. Frontiers in Complex Systems, 2: 1323321. doi:10.3389/fcpxs.2024.1323321
- Undheim, T. A. (2024). The whack-a-mole governance challenge for AI-enabled synthetic biology: literature review and emerging frameworks. Frontiers in Bioengineering and Biotechnology, 12: 1359768. doi:10.3389/fbioe.2024.1359768
- Undheim, T. A. (2024). Dyadic risk mechanisms — a nomenclature for 36 proto-cascading effects determining humanity's future. European Journal of Futures Research, 12: 7. doi:10.1186/s40309-024-00228-2
- Undheim, T. A. (2024). In search of better methods for the longitudinal assessment of tech-derived x-risks: how five leading scenario planning efforts can help. Technology in Society, 77: 102505. doi:10.1016/j.techsoc.2024.102505
- Altsitsiadis, E., Undheim, T., de Vries, E., et al. (2012). Health literacy, sunscreen and sunbed use: an uneasy association. British Journal of Dermatology, 167(S2): 14–24.
- Curran, P., Undheim, T. A. (2011). The Java Community Process: standardization, interoperability, transparency. 7th International Conference on Standardization and Innovation in Information Technology (SIIT), IEEE, 1–8.
- Codagnone, C., Undheim, T. A. (2008). Government efficiency and effectiveness: the theory and practice of benchmarking and measurement. European Journal of ePractice, 4: 4–18.
- Undheim, T. A. (2003). Getting connected: how sociologists can access the high-tech élite. The Qualitative Report, 8(1): 104–128.
Canonical 10
The ten works most frequently cited across the CEB's factor anchors and cascade-evidence edges. Frequency is computed across both factor source-attestation (weighted ×2) and cascade-edge evidence references; each entry's anchor count is shown in brackets. These are the framework's load-bearing references — the works a reader needs to know to evaluate CREF's empirical claims.
- IPCC (2021) — Intergovernmental Panel on Climate Change. Climate Change 2021: the physical science basis. Cambridge: Cambridge University Press; 2021. → source [cited in 41 factor/edge anchors]
- IPBES (2019) — IPBES. Global assessment report on biodiversity and ecosystem services. 2019. → source [cited in 26 factor/edge anchors]
- Vosoughi et al. (2018) — Vosoughi S, Roy D, Aral S. The spread of true and false news online. Science 2018;359(6380):1146-1151 → source [cited in 22 factor/edge anchors]
- Perrow (1984) — Perrow C. Normal Accidents: Living with High-Risk Technologies. Basic Books; 1984 [cited in 12 factor/edge anchors]
- Levitsky & Ziblatt (2018) — Levitsky S, Ziblatt D. How democracies die. New York: Crown; 2018. → source [cited in 18 factor/edge anchors]
- Tooze (2018) — Tooze A. Crashed: how a decade of financial crises changed the world. New York: Viking; 2018. → source [cited in 18 factor/edge anchors]
- Zuboff (2019) — Zuboff S. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. PublicAffairs; 2019 [cited in 18 factor/edge anchors]
- Reinhart & Rogoff (2009) — Reinhart CM, Rogoff KS (2009) This Time is Different: Eight Centuries of Financial Folly. Princeton University Press → source [cited in 16 factor/edge anchors]
- Lenton et al. (2019) — Lenton TM, Rockström J, Gaffney O, et al. Climate tipping points — too risky to bet against. Nature. 2019;575(7784):592-595. → source [cited in 10 factor/edge anchors]
- Pescaroli & Alexander (2018) — Pescaroli G, Alexander D. Understanding compound, interconnected, interacting, and cascading risks. IJDRR. 2018;30:209-220. → source [cited in 6 factor/edge anchors]
Adjacent frameworks
CREF does not replace existing risk-governance traditions. It adds one specific layer: cross-domain cascade legibility through evidence-anchored edges between named factors. The frameworks below cover different cuts of the same terrain; CREF is most useful when combined with one of them. Each row links to the canonical reference.
| Framework | Primary focus | What CREF adds |
|---|---|---|
| Renn (2008) — Risk Governance | Risk classification, framing, evaluation, management | Empirical cascade edges between named factors; per-edge mechanism codes; Q-scored citations |
| IRGC — Risk Governance Framework | Multi-stakeholder risk governance process | Domain-typed factor taxonomy with explicit role grammar (trigger, amplifier, dampener, absorber, constraint) |
| Helbing (2013) — Globally Networked Risks | Complexity-theoretic systemic risk modelling | 296 named risk factors anchored to documented historical cases; intervention-design grammar via roles |
| Goldin & Mariathasan (2014) — The Butterfly Defect | Globally interconnected crisis interdependence | Per-edge mechanism codes and time lags; reversibility tagging; full CEB audit trail |
| Lawrence et al. (2024) — Global Polycrisis | Causal entanglement of crises (common stresses, domino, feedbacks) | Operational substrate for the causal-pathway typology: 1,048 evidence-anchored edges that instantiate the three pathways at factor-level granularity |
| Pescaroli & Alexander (2018) — Compound / Cascading Risks | Compound, interconnected, interacting, cascading risk typology | Computational substrate (CREF-Lang) with role notation and an evidence base usable by both researchers and operators |
| Tonn & Stiefel (2014) — Human Extinction Risk and Conditions for Action | Ethical thresholds for catastrophic risk; conditions for action | Quantitative scenario architecture from trigger to existential-risk endpoints; explicit cascade pathways |
Acknowledgments
Co-authors on CREF-adjacent work: Isabella Duncan, Jennifer Lin, Lovejit Gill (Stanford SERI undergraduate researchers); Qaiser Malik (NEAIS 2025); Tahir Ahmad (Frontiers 2024); Anka Reuel, Lujain Soder, Ben Bucknall (ICML 2024). Institutional support from Stanford CISAC and SERI. This site runs on the CEB v8.5.36 dataset and is released under CC BY-NC 4.0 alongside the CREF manuscript.
Source register.
All 1,288 sources behind 1,371 cascade edges and 296 factors. Ranked by anchor frequency. Search, filter, or open any entry for full anchors and DOI.
Nine fields, six forces.
Nine academic fields generalised from the canonical taxonomy in Undheim (2024) "Dyadic Risk Mechanisms" (European Journal of Futures Research 12:7). Disaster studies is added as the immediate antecedent field (Pescaroli & Alexander, UNDRR, NaTech tradition) that the Dyadic paper opens with. Keywords carry the precision; field categories stay legible.
Each field contributes concepts to one or more CREF domains. Hover a field to see its primary domain links; click to filter the Bibliography by that field. The CREF cascade ontology in the center is generated by integrating these field-specific conceptual contributions.
Science & Technology Studies
Sociology of scientific knowledge, technology trajectories, public understanding of science, actor-network theory. Includes computing-and-AI literature when oriented to sociotechnical analysis. The Dyadic paper sub-cites this field as "multidisciplinary science and technology studies" [25, 56, 60, 65].
Disaster studies
Cascading-disasters lineage from Pescaroli & Alexander; NaTech research; critical infrastructure interdependency networks; UNDRR disaster-risk framework. The Dyadic Risk paper opens with this field as the immediate antecedent of CREF.
Economics
Macroeconomic crisis, financial intermediation, institutional economics, inequality, banking, sovereign debt. Includes the management-studies disruptive-innovation tradition (Christensen) when framed as economic disruption. Dyadic field [2, 24, 114].
Ethics
Ethics of emerging technologies, AI alignment, dual-use research ethics, professional codes. Dyadic field [76].
Futures research
Scenario methods, foresight, futures studies, polycrisis analysis. Dyadic field [27, 87, 91, 126].
Sociology
Risk society, social capital, trust, polarisation, sociology of risk, social-determinants tradition. Dyadic field [14, 38, 45].
Political Science
Risk governance in political science, democratic backsliding, international relations, sovereignty, governance frameworks. Dyadic field [84] plus broader political-science.
Sustainability studies
Climate science, biodiversity, planetary boundaries, sustainability, social-determinants-of-health where environmental. Dyadic field [35, 64] plus public-health environmental work.
X-risk studies
Existential risk, global catastrophic risk, cascading catastrophic scenarios. Dyadic field [8, 12, 69, 82, 83, 90, 109, 119, 121, 122, 133]. Note the Dyadic paper questions whether x-risk is qualitatively distinct from adjacent fields; we retain it as a discrete field for filter utility while acknowledging the boundary is porous.
The six disruptive forces (CREF domains)
PEST analysis [Cox 2021] extended to six forces. 6² = 36 dyads (mapped below); 6³ = 216 triads (proposal below); the combinatorial explosion is what motivates CREF's factor-level compression.
Three-letter codes (STI, GOV, ...) for CREF factors; short stems (Tech, Gov, ...) for the NaTech-style dyadic and triadic nomenclature below.
The 100 ideas the framework stands on.
A conceptual anchor is a theoretical touchstone — a work whose idea grounds one or more CREF factors. These are not the full citation list (that is the Source Register, 1,288 references). The anchors are the curated intellectual lineage: the 100 works that do load-bearing conceptual work, each tagged with the role it plays, the factors it grounds, its academic field, and its current curation status.
What is this page?
Every CREF factor is grounded in prior thought. This page indexes the 100 conceptual anchors that lineage rests on — Beck's Risk Society, Acemoglu & Robinson on institutions, the IPCC and WHO corpora, and so on. Search by author or idea; filter by field or curation status; each row lists the factors the anchor grounds. The difference from the Source Register: the register is the full bibliographic record; the anchors are the subset that carry conceptual weight, annotated for what work each one does.
36 dyads, six forces, one matrix.
The Dyadic Risk paper (Undheim 2024, EJFR 12:7) coins 36 dyadic risk names by generalising the NaTech tradition across all six force-pairs. Each cell shows the dyad name, current CEB edge count, and the number of cases whose pathway traverses that pair. Click any non-diagonal cell to list its cases.
Anchor: CEB v8.5.18 / Widget v3.9.64 · dyadic_data_v2 · Reading: Row = source force (cause); Column = target force (affected). Source→Target ordering matters. Diagonal cells (T2, E2, G2, S2, N2, H2) are intra-force dynamics, not cross-domain dyads.
Detecting evidence gaps in cross-domain risk taxonomies.
A network-topology method applied to the Cascade Evidence Base. The dominant finding: 19 of the top 25 candidate links are not absent from knowledge — they are absent from the curated taxonomy, through directional coding asymmetry (REV) or vocabulary mismatch (VOC). The method is a coverage diagnostic, not a prediction of novel risk events.
What is this page?
Cascade X scores factor-pairs the evidence base has not yet linked but whose network position suggests it should, then sorts each into one of four diagnostic categories (REV / VOC / GEN / FAL) a reviewer can act on. Sections below: the method, the gap typology, the Periodic Table of Risk, the Top 10 candidates, and the candidate-card schema. Read in order for the method; jump to the Top 10 for the intuition.
Found a gap that matters to you? The Customize pane turns these into mitigation and coverage actions.
How the discovery works
Think of the evidence base as a map of known cause-and-effect links between risks. Some links are well documented; many are not. Cascade X looks at the shape of the map — which risks sit close together, share neighbours, or cluster — and flags pairs that should be connected but aren't yet recorded. It is a way of finding the blank spots in our knowledge, not a forecast of what will happen next. Full method write-up: see the Method pane.
Why a gap isn't always real ignorance (the four types)
When the method flags a missing link, there are four reasons it might be missing — and only one is a true blind spot:
- Reverse (REV) — we recorded the link going one way but not the other. The knowledge exists; the record is one-sided.
- Vocabulary (VOC) — the link is well known in another field, just under different words, so our taxonomy missed it.
- Genuine gap (GEN) — a real blind spot worth researching. The valuable case.
- False positive (FAL) — the network shape suggested a link that doesn't actually make sense.
The headline finding: most flagged gaps are REV or VOC — the knowledge is out there, just not captured in the framework. Full typology and counts: Method pane.
How to read a candidate
Each X-Factor candidate card names the two risks it links, a confidence and priority flag, a one-line reason the link should exist, the discipline whose literature would confirm it, and the concrete coding action that would close the gap. Open the Top 10 and click any row to see all of this. The full field-by-field schema lives in the Cascading X-Factors draft.
Dyadic nomenclature (NaTech-style)
The 36 dyadic risk names — one per ordered force-pair, generalising the NaTech tradition — now live in their own interactive view, where each cell is clickable through to its CEB edge count, case count, and underlying cases. Open the Dyadic Grid →
Triadic nomenclature — proposal
The Dyadic Risk paper notes that "thoroughly describing the 216 triadic risk relationships in the manner we have done here for 36 dyads, for example, would likely be out of scope for a scientific paper" (Section: "Dyads as building blocks for causal analysis"). The cascade explorer can prototype the nomenclature for the subset with burst-level evidence — 16 of 216 triads currently covered (7.4%).
Proposed naming convention: ordered concatenation in temporal-cascade order (earliest cause first). Examples already visible on burst tiles:
- TechGovSoc — Algorithmic polarisation cascade (tech triggers governance failure that erodes social cohesion)
- NatEconHealth — Heatwave mortality cascade (climate event triggers economic disruption with health consequences)
- EconGovHealth — Pension solvency cascade (economic stress triggers governance failure that degrades health systems)
- TechHealthEcon — AMR and post-antibiotic cascade (technology trajectory affects health, with economic spillover)
Each burst tile now carries its NaTech-style name as a tag (dyadic, triadic, tetradic, or pentadic — see Bursts panel in the Apply pane).
Order of interaction (dyadic, triadic, tetradic, pentadic)
Each Cascade Burst in the framework now carries an explicit order tag showing how many distinct force-domains it spans. The Dyadic paper's argument is that dyads are the building blocks for causal analysis; triads and above are necessary compressions of the inherent combinatorial complexity. Current burst inventory by order:
6 bursts
20 bursts
7 bursts
2 bursts
No monadic (single-domain) bursts exist in the current set — by construction, a burst is multi-domain. No hexadic (all six) burst exists yet either; that would be a "polycrisis-complete" cascade.
The periodic table of risk.
Risks arranged like elements: the force that drives them (rows), the mechanism by which they spread (columns), and a magnitude — potential risk impact — that shades and ranks each one. The glowing cells are gaps: places the map is still blank. Switch the view to watch the blanks move.
Missing links worth chasing.
Connections the evidence base is missing but the network predicts should exist — ranked by prediction strength, each with the targeted search that would confirm it. Click a row to expand its full reasoning. This is the prioritised queue for hardening the framework.
The shortlist
Ranked by an ensemble link-prediction score — how strongly the network's shape implies a link should exist between two factors that currently have none. The score combines three standard topology measures, shown on every row so the ranking is fully transparent.
How the score works
For every unconnected pair of factors, three measures are computed from the existing graph: Common Neighbours (CN) — how many factors both already link to; Jaccard (J) — that overlap as a proportion; and Adamic–Adar (AA) — the same, weighting rare shared neighbours more heavily. The ensemble blends them (0.4·CN + 0.3·10J + 0.3·10PA). Higher ensemble = stronger structural case that the link is missing rather than absent. Validation: temporal-holdout on the edge record, ensemble Recall@25 = 17.0% (6.1× over chance). Each candidate also carries a class — REV, VOC, GEN, or FAL — explaining why the link is missing.
Concepts & frequently-asked framework questions.
Every named term CREF uses, every methodological question reviewers tend to ask. Sift alphabetically or by category. Use the print button for the handover packet.
Help build the cascade infrastructure.
The CREF framework, the Cascade Evidence Base, and the Atlas are open infrastructure. They get better when more people contribute cases, sources, and edges — and when more institutions adopt the grammar. This page is the entry door for everyone who is not (yet) a paying client and wants to do something other than read.
Why join
Cascade analysis is most useful when its evidence base covers more sectors, regions, and time periods than any one team can author alone. Contribution improves the work everyone uses.
What you can do
Submit a case, propose a new factor or edge, flag a citation gap, run a workshop with the Crisis-game pane, or sponsor a sector wedge. Five archetypes below.
How to start
Pick a role, read its short brief, and use the channel listed (email, GitHub, or the in-page submission form where one exists). Nothing requires institutional affiliation.
1. Roles
Fixer
Author
CREF_Coder_Materials workbook.Learner
Player
Patron
2. Contribute sources or cases
The Cascade Evidence Base currently holds 245 factors, 1,345 CEM edges, 167 reference cases, and 2,214 sources. Gaps are real: under-represented regions (Sub-Saharan Africa, South America, Oceania) and certain sector pairs (e.g., insurance–public-health, fisheries–trade) have thin evidence. New cases use a fixed template — email for the coder pack. Cite the source. Use the controlled vocabulary. Flag uncertainty.
3. Sponsor
If you or your institution would underwrite a sector wedge — say, a CREF expansion for community-bank balance-sheet cascades, or for cross-border water-energy-food triggers — write to the address below with a one-line scope. No pitch decks, no minimums. The work goes back into the open evidence base unless explicitly scoped otherwise.
4. People
5. Stay connected
Contact: tautec@gmail.com. The framework, the workbook, and the Atlas source are tracked on GitHub at github.com/trondundheim/creflang, with a citable archive on Zenodo (Zenodo record 20492776). A low-volume newsletter goes out when a new manuscript is accepted or the workbook ships a major version.
About
The CREF Atlas is the analytical companion to the Cascading Risk Effects Framework — a grammar for decomposing systemic risk into roles, mechanisms, and evidence-anchored edges across six domains.
Lineage. CREF builds on the curation model Trond began in 2013 with Yegii — a man/machine insight network for executives, strategists, and analysts disoriented by tech disruption. Yegii's transparent-criteria, 100-point credibility scoring is the antecedent of CREF's evidence-tier discipline (T0–T3); its insight-aggregation method is the antecedent of the CEB.
Citation
Versions
CREF Atlas v3.11.29 · CEB v8.5.36
245 factors · 1,345 CEM edges (CEB v8.5.35 (245 factors, 1,345 edges; Composite scores added; 909-edge canonical public subset retained for comparison) · 336 reference cases · 69 glossary entries · 100 conceptual anchors
Note: XFC in the palette is a cross-cutting/extension visual class, not a seventh core domain. CREF uses six core domains.
Modes
?mode=reviewer — pins Method as default landing and hides commercial framing.
?mode=workshop — switches Start landing to the Play pane.
Applied use in organizations
Applied organizational versions are described under Apply. The open Explorer remains the public analytical companion to CREF.
Sitemap.
Every section of the Atlas, and what each one indexes. Use site search (or press ⌘K / Ctrl-K) to jump straight to a factor, edge, case, or term.
About CREF Atlas.
CREF Atlas is an interactive companion to the Cascading Risk Effects Framework (CREF) — a grammar for decomposing how systemic risk propagates across domains.
Provenance
CREF Atlas is a product of the Yegii, Inc. R&D Lab, where the framework is developed, coded against the Cascade Evidence Base (CEB), and built into this interactive companion, available as a website. The companion GitHub repository is CREF-Lang, a domain-specific language and reference compiler for cross-domain cascade-risk notation, with a Zenodo release (see zenodo.org/records/20492776). CREF-Lang expresses how risk propagates across societal domains — science & technology, governance, economy, society, environment, health — as composable, parseable, machine-checkable pathway expressions.
CREF traces to the Stanford Existential Risks Initiative (SERI) and the Center for International Security and Cooperation (CISAC) at Stanford. It began with the Stanford Cascading Risk Study.
Author
Trond Arne Undheim — PhD, entrepreneur, researcher, speaker, and founder of Yegii, Inc. Formerly Stanford, MIT, NTNU, WPP, Hitachi, and the EU. He leads the CREF program at the Yegii R&D Lab.
How to cite
Code & archive
The framework, the Cascade Evidence Base workbook, and the Atlas source are open. Code is on GitHub at github.com/trondundheim/creflang; a citable, versioned archive is deposited on Zenodo (Zenodo record 20492776).
About Yegii R&D Lab
The Yegii, Inc. R&D Lab builds risk and innovation insight tools and platforms for the business professional disoriented by technological disruption. We focus on three kinds of professionals — executives, strategists, and analysts — and work especially with the financial sector and other regulated industries, while also addressing the distinct challenges of the technology sector.