QURI (Quantified Uncertainty Research Institute)

📋Page Status

Page Type:ContentStyle Guide →Standard knowledge base article

Quality:48 (Adequate)⚠️

Importance:28.5 (Peripheral)

Last edited:2026-01-29 (3 days ago)

Words:4.4k

Structure:

📊 21📈 1🔗 1📚 70•25%Score: 13/15

LLM Summary:QURI develops Squiggle (probabilistic programming language with native distribution types), SquiggleAI (Claude-powered model generation producing 100-500 line models), Metaforecast (aggregating 2,100+ forecasts from 10+ platforms), and related epistemic tools. Founded 2019 by Ozzie Gooen with $850K+ funding (SFF $650K, Future Fund $200K), primarily serving EA/rationalist community for cost-effectiveness analysis and Fermi estimation.

Issues (2):

QualityRated 48 but structure suggests 87 (underrated by 39 points)
Links4 links could use <R> components

Quick Assessment

Dimension	Assessment	Evidence
Innovation	High	Squiggle is unique probabilistic language with native distribution algebra; SquiggleAI generates 100-500 line models
Practical Impact	Growing	GiveWell CEA quantification projects (≈300 hours of work), AI timeline models, EA cause prioritization
Open Source	Fully	All projects MIT licensed on GitHub; monorepo includes Squiggle, Hub, and Metaforecast
Funding	Stable	$850K+ total: SFF ($650K through 2022), Future Fund ($200K, 2022), LTFF (ongoing)
Team Size	Small	≈3-5 core contributors including Ozzie Gooen (founder), with fiscal sponsorship from Rethink Priorities
AI Integration	Active	SquiggleAI uses Claude Sonnet 4.5 (20K token context cached); RoastMyPost uses Claude + Perplexity
Community	Niche but Engaged	EA Forum presence, Forecasting & Epistemics Slack (#squiggle-dev), $300 Fermi competitions
Data Scale	Significant	Metaforecast indexes 2,100+ forecasts; Guesstimate models: 17,000+ public

Organization Details

Attribute	Details
Full Name	Quantified Uncertainty Research Institute
Founded	2019 (evolved from Guesstimate, founded 2016)
Founder & Executive Director	Ozzie Gooen
Location	Berkeley, California (primarily remote)
Status	501(c)(3) nonprofit; fiscally sponsored by Rethink Priorities
EIN	84-3847921
Website	quantifieduncertainty.org
GitHub	github.com/quantified-uncertainty (monorepo structure)
Substack	quri.substack.com
Primary Funders	Survival and Flourishing Fund (SFF), Long-Term Future Fund (LTFF), Future Fund (pre-FTX)
Total Funding	$850,000+ through 2022

Overview

The Quantified Uncertainty Research Institute (QURI) is a nonprofit research organization focused on developing tools and methodologies for probabilistic reasoning and forecasting. Founded in 2019 by Ozzie Gooen, QURI aims to make uncertainty quantification more accessible and rigorous, particularly for decisions affecting the long-term future of humanity. The organization evolved from Gooen’s earlier work on Guesstimate, a spreadsheet tool for Monte Carlo simulations that demonstrated strong demand for accessible uncertainty tools in the effective altruism community.

QURI’s flagship project is Squiggle, a domain-specific programming language designed specifically for probabilistic estimation. Unlike general-purpose languages, Squiggle provides first-class support for probability distributions, enabling analysts to express complex uncertainties naturally. The language powers Fermi estimates, cost-effectiveness analyses, and forecasting models used throughout the effective altruism and rationalist communities. QURI’s experience building tools like Squiggle and Guesstimate has revealed a significant challenge: even highly skilled domain experts frequently struggle with basic programming requirements and often make errors in their probabilistic models.

The organization has expanded beyond Squiggle to encompass a suite of epistemic tools: Squiggle Hub for collaborative model sharing (hosting 17,000+ Guesstimate models), Metaforecast for aggregating predictions across 10+ platforms, SquiggleAI for LLM-assisted model generation, and RoastMyPost for AI-powered blog post evaluation. Together, these tools form an ecosystem aimed at improving quantitative reasoning in high-stakes domains.

Organizations in the effective altruism and rationalist communities regularly rely on cost-effectiveness analyses and Fermi estimates to guide their decisions. QURI’s mission is to make these probabilistic tools more accessible and reliable for altruistic causes, bridging the gap between sophisticated quantitative methods and practical decision-making.

Founder: Ozzie Gooen

Ozzie Gooen is the founder and Executive Director of QURI, with a background spanning engineering, effective altruism, and forecasting research.

Background and Career

Period	Role	Focus
2008-2012	Harvey Mudd College	B.S. General Engineering (Economics concentration)
≈2014	Founding Engineer, ZenTrust	Estate planning web service
2015-2016	.impact co-founder	EA community building infrastructure
2016	Guesstimate founder	Monte Carlo spreadsheet tool
2017-2019	Research Scholar, Future of Humanity Institute	Forecasting infrastructure research
2019-present	Executive Director, QURI	Epistemic tools development

Gooen discovered effective altruism in college after learning he “sounded like a utilitarian.” His path from engineering to forecasting research reflects a consistent interest in applying quantitative methods to important decisions. At FHI, he focused on forecasting infrastructure research, exploring how prediction markets and aggregation methods could improve institutional decision-making.

Beyond QURI, Gooen serves on the boards of Rethink Charity and Rethink Priorities. He has contributed to discussions on AI governance, forecasting methodology, and quantitative cause prioritization through extensive writing on the EA Forum and LessWrong.

Notable Contributions

Guesstimate: Created the first widely-used Monte Carlo spreadsheet, which runs 5,000 simulations per model and has accumulated 17,000+ public models
Forecasting Research: Developed proposals for prediction market improvements and forecast aggregation methods
QURI Tools Ecosystem: Oversaw development of Squiggle, Metaforecast, SquiggleAI, and RoastMyPost
EA Community: Helped build early EA infrastructure through .impact and ongoing community engagement

Core Projects

Loading diagram...

Squiggle Language

Squiggle is QURI’s primary project—a domain-specific programming language for probabilistic estimation that runs in the browser via JavaScript. It can be useful to think of Squiggle as similar to SQL, Excel, or probabilistic programming languages like WebPPL: there are simple ways to declare variables and write functions, but don’t expect classes, inheritance, or monads.

Design Philosophy

Squiggle is meant for intuitively-driven quantitative estimation rather than data analysis or data-driven statistical techniques. It’s designed for situations where there is very little data available, and most variables will be intuitively estimated by domain experts. The syntax is forked from Guesstimate and Foretold, optimized for readable probabilistic expressions.

Core Features

Feature	Description	Example
Distribution Types	Native support for normal, lognormal, uniform, beta, cauchy, gamma, logistic, exponential, bernoulli, triangular	`normal(10, 2)`, `beta(2, 8)`
The “to” Syntax	Intuitive lognormal creation using 5th/95th percentiles	`1 to 10` equals `lognormal({p5: 1, p95: 10})`
Distribution Algebra	Mathematical operations propagate through distributions	`normal(10, 2) * uniform(0.8, 1.2)`
Multiple Parameterizations	Same distribution, different inputs	`normal({p5: 5, p95: 15})`, `normal({mean: 10, stdev: 3})`
Functions with Domains	Type-checked parameter ranges	`f(x: [1,10]) = x * 2` fails if called with `f(15)`
Three Representations	Sample Set (default, supports correlations), Point Set, Symbolic	`Sym.normal(10, 2)` for symbolic
Monte Carlo Sampling	Automatic sampling and propagation	Built into runtime
Visualization	Native plotting of distributions	Integrated in playground

Distribution Creation Syntax

Squiggle offers flexible distribution creation with multiple parameterization options:

// Normal distribution - multiple ways to specify
normal(10, 2)                    // mean, standard deviation
normal({mean: 10, stdev: 2})     // explicit parameters
normal({p5: 5, p95: 15})         // percentile-based
normal({p10: 6, p90: 14})        // different percentiles
normal({p25: 8, p75: 12})        // quartile-based

// Lognormal distribution - the workhorse for positive quantities
lognormal(2, 0.5)                // mu, sigma (log-scale parameters)
lognormal({mean: 10, stdev: 5})  // arithmetic mean/stdev
5 to 50                          // shorthand for lognormal({p5: 5, p95: 50})

// Beta distribution - for probabilities between 0 and 1
beta(2, 8)                       // alpha, beta parameters (~20% mean)
beta({mean: 0.3, stdev: 0.1})    // mean/stdev parameterization

// Other distributions
uniform(5, 15)                   // uniform between bounds
triangular(5, 10, 15)            // min, mode, max
exponential(0.1)                 // rate parameter

Squiggle Code Examples

Basic Cost-Effectiveness Model:

// Cost-effectiveness model for AI safety intervention
interventionCost = lognormal(1e6, 1.5) // \$1M median, high uncertainty
probabilityOfSuccess = beta(2, 8) // ~20% base rate
valueIfSuccessful = lognormal(1e12, 2) // High but uncertain value

expectedValue = probabilityOfSuccess * valueIfSuccessful
costEffectiveness = expectedValue / interventionCost

Function with Domain Constraints:

// Calculate expected value with bounded probability
calculateEV(probability: [0, 1], value) = {
  adjustedProb = probability * normal(1, 0.1) // Add estimation uncertainty
  truncate(adjustedProb, 0, 1) * value
}

// Use the function
projectValue = calculateEV(0.3, lognormal(1e9, 2))

Multi-Variable Fermi Estimate:

// Estimate: Number of piano tuners in Chicago
chicagoPopulation = 2.7M to 2.9M
householdsPerPerson = 0.35 to 0.45
pianoOwnershipRate = 0.02 to 0.05
tuningsPerYear = 0.5 to 2
hoursPerTuning = 1.5 to 2.5
workingHoursPerYear = 1800 to 2200

totalTunings = chicagoPopulation * householdsPerPerson *
               pianoOwnershipRate * tuningsPerYear
tunerCapacity = workingHoursPerYear / hoursPerTuning
numberOfTuners = totalTunings / tunerCapacity

Version History

Version	Date	Key Changes
0.10.0	January 2025	SqProject rewrite, Web Workers by default, compile-time type inference, unit type annotations, UI overhaul
0.9.4-0.9.5	2024	Experimental Web Worker runner, version selection in playground
0.8.6	2024	Import/export support, multi-model projects
0.8.x	2023	Performance improvements, Squiggle Hub integration
0.7.0	2023	SquiggleAI integration foundations
Early Access	2020	Initial public release

Squiggle 0.10.0 Deep Dive

The January 2025 release of Squiggle 0.10.0 represented six months of development with significant architectural changes:

Web Workers: All Squiggle code now runs in a separate Web Worker thread by default, with results marshaled back asynchronously. This prevents UI freezes during complex calculations.

Type System: New compile-time type inference transforms AST to typed AST, enabling earlier error detection. The pipeline now includes semantic analysis for type checks.

Unit Type Annotations (experimental, contributed by Michael Dickens): Variables can be annotated with physical units like kilograms, dollars, or compound units like m/s^2.

UI Changes: The output viewer now defaults to collapsed variables. Use the @startOpen decorator to expand variables by default.

Strengths and Limitations

Strengths:

Simple, readable syntax optimized for probabilistic math
Fast for small-to-medium models
Strong for rapid prototyping
Optimized for numeric and symbolic approaches, not just Monte Carlo
Embeddable in JavaScript applications
Free and open-source (MIT license)

Limitations:

Does not support Bayesian inference (cannot do backwards inference from data)
Much slower than languages like Stan or PyMC on large models
Limited scientific computing ecosystem
Beta distributions display poorly when alpha or beta are below 1.0

SquiggleAI

SquiggleAI integrates large language models to assist with probabilistic model creation, addressing a key challenge QURI identified: even highly skilled domain experts frequently struggle with basic programming requirements for probabilistic modeling.

How It Works

SquiggleAI uses prompt caching to cache approximately 20,000 tokens of information about the Squiggle language, ensuring the LLM has deep knowledge of syntax, distributions, and best practices. Most workflows complete within 20 seconds to 3 minutes, producing models typically 100-500 lines long depending on the model used.

Model Support

Model	Status	Output Size	Speed	Best For
Claude Sonnet 4.5	Primary	≈500 lines	Medium	Complex multi-variable models
Claude Haiku 4.5	Available	≈150 lines	Fast	Quick prototypes
Grok Code Fast 1	Available	≈200 lines	Fast	Alternative provider
Claude Sonnet 3.5	Legacy	≈200 lines	Medium	Stable fallback

Capabilities

Capability	Description	Example Use Case
Model Generation	Describe a problem in natural language, receive executable Squiggle code	”Estimate the cost-effectiveness of hiring an AI safety researcher”
Iterative Refinement	Conversation-based model improvement	”Add more uncertainty to the timeline assumptions”
Fermi Estimation	Generate complete uncertainty models from vague questions	”How many piano tuners are in Chicago?”
Code Debugging	Identify and fix syntax and logic errors	Fix distribution domain mismatches
Model Explanation	Explain what existing Squiggle code does	Document inherited models

SquiggleAI Outputs

SquiggleAI outputs on Squiggle Hub are private by default. Users who want to share models or make them public can explicitly do so by creating new public models. This privacy-first approach encourages experimentation without concern about incomplete drafts being visible.

Integration with Squiggle Hub

SquiggleAI is directly accessible from within Squiggle Hub, allowing users to generate models and immediately save, version, and share them through the platform. The tight integration reduces friction between ideation and publication.

Squiggle Hub

Squiggle Hub is a platform for creating, sharing, and collaborating on Squiggle models. It was announced in 2024 as a central repository for probabilistic models in the EA and rationalist communities.

Core Features

Feature	Description	Details
Model Hosting	Publish and share Squiggle models	Public or private visibility options
Versioning	Track model history and changes	Git-like version control for models
Imports/Exports	Multi-model projects	Import functions and distributions between models
Collaboration	Multiple contributors per project	Group-based permissions
Embedding	Embed models in external sites	React components for documentation
Version Selection	Pick Squiggle version per model	Test on different versions, avoid breaking changes
SquiggleAI Integration	AI-assisted model creation	Direct access within the platform

Groups and Collections

Squiggle Hub organizes models into groups for easier discovery:

Meta-Squiggle: Models about Squiggle itself and QURI operations
Innovation: Models exploring novel estimation techniques
User Collections: Individual user portfolios

Notable Models and Use Cases

The platform hosts various types of probabilistic models:

Category	Example Models	Typical Complexity
Cost-Effectiveness	GiveWell charity evaluations, AI safety interventions	200-500 lines
Fermi Estimates	Population estimates, market sizing	50-150 lines
Forecasting	Election probabilities, technology timelines	100-300 lines
Decision Models	Career choice analysis, funding allocation	150-400 lines

Guesstimate Legacy

Squiggle Hub also provides access to 17,000+ public models from Guesstimate, Gooen’s earlier Monte Carlo spreadsheet tool. This archive represents years of probabilistic modeling by the EA community and serves as a reference library for common estimation patterns.

Metaforecast

Metaforecast aggregates forecasts from multiple prediction platforms into a single searchable interface. The initial version was created by Nuño Sempere, with help from Ozzie Gooen, at QURI.

Purpose and Design

Forecasting is a public good, but platform fragmentation reduces its utility. Metaforecast addresses this by combining data from 10+ platforms into a unified search interface. Data is fetched daily, showing immediate forecasts without historical data—optimized for quick lookups rather than trend analysis.

Platforms Aggregated

Platform	Type	Questions	Notes
Metaculus	Reputation-based	≈1,200 (55% of total)	Largest source; research-focused
Manifold	Play money market	Varies	Research-focused, no profit motive
Polymarket	Real money (crypto)	Varies	Rose to prominence in 2024 US election
Good Judgment Open	Superforecaster platform	≈100-200	Connected to Good Judgment Project
PredictIt	Political market	≈50-100	US political focus
Kalshi	CFTC-regulated market	≈100-200	Regulatory compliance focus
Insight Prediction	Prediction market	Varies	Additional source
INFER	Policy forecasting	Varies	Government-adjacent

Scale: Currently indexes approximately 2,100 active forecasting questions across all platforms, plus access to 17,000+ public Guesstimate models.

Features

Feature	Description	Technical Details
Unified Search	Find forecasts across all platforms	Elasticsearch-powered
Platform Comparison	See estimates from different sources	Side-by-side probability views
GraphQL API	Programmatic access	Available for integrations
Daily Updates	Fresh data fetched automatically	Scheduled scraping pipeline
Open Source	GitHub repository	Part of Squiggle monorepo

Integrations

Metaforecast has been integrated with several external services:

Twitter: Bot posting notable forecasts
Fletcher: Discord bot integration
GlobalGuessing: Forecasting community platform
Elicit (previously): AI research assistant

The open-source nature and GraphQL API enable custom integrations for research and analysis workflows.

2024 US Election Performance

During the 2024 US presidential election, Polymarket (one of Metaforecast’s sources) demonstrated the value of prediction market aggregation. Polymarket strongly favored Trump’s victory even as traditional polls showed a closer race. The market’s prediction proved correct, with reports of approximately $30 million in bets from individual traders helping shape the odds.

Additional Tools

RoastMyPost

RoastMyPost is a QURI application that uses LLMs and code to evaluate blog posts and research documents. It was announced as a tool for improving writing quality in the EA and rationalist communities.

Evaluators Available:

Evaluator	Focus	Technical Implementation
Fact Check	Verify claims against sources	Perplexity API via OpenRouter
Spell Check	Identify typos and errors	Programmatic checks
Fallacy Check	Detect logical fallacies	Claude Sonnet 4.5
Math Check	Verify calculations	Claude + symbolic verification
Link Check	Validate URLs	HTTP requests
Forecast Check	Evaluate probabilistic claims	Squiggle integration
Custom Evaluators	User-submitted checks	Community contributions

How It Works:

Import document via markdown text or URL (optimized for EA Forum and LessWrong)
Select evaluators (system-recommended or custom)
Processing takes 1-5 minutes
Review flagged issues for human judgment

Use Cases:

Draft polishing: Catch errors before publication
Public trust signaling: Link to evaluations in blog posts (like GitHub badges)
LLM-assisted workflows: Automated quality checks on AI-generated content

Limitations: The false positive rate for error detection is significant. RoastMyPost is best for flagging issues for human review, not for treating results as authoritative. Fact Check and Fallacy Check work best on fact-dense, rigorous articles.

Fermi Competition

QURI runs periodic $100 Fermi Model Competitions to encourage creative Fermi estimation using AI tools. The February 2025 competition used Claude 3.5 Sonnet and the QURI team as judges.

Evaluation Criteria:

Criterion	Weight	Description
Surprise	40%	How unexpected/novel are the findings?
Topic Relevance	20%	Relevance to rationalist/EA communities
Robustness	20%	Reliability of methodology and assumptions
Other	20%	Presentation, creativity, clarity

The competition explicitly encourages novel approaches over exhaustively researched calculations, promoting experimentation with AI-assisted quantitative reasoning.

Squigglepy (Python Implementation)

Squigglepy is a Python package developed by Rethink Priorities that implements Squiggle-like functionalities for those who prefer working in the Python ecosystem. While Squiggle is implemented in JavaScript, Squigglepy allows integration with Python’s scientific computing stack (NumPy, SciPy, pandas).

Funding and Organization

Funding History

Source	Amount	Period	Notes
Survival and Flourishing Fund	$650,000+	2019-2022	Primary early funder; Jaan Tallinn’s philanthropic vehicle
Future Fund	$200,000	2022	Pre-FTX collapse; lost future commitments
Long-Term Future Fund	Ongoing	2023-present	Current primary funder
Individual Donors	Various	Ongoing	Via Every.org and direct giving
Total Historical	$850,000+	Through 2022	Documented funding

Funding Context

The Survival and Flourishing Fund (SFF) is financed primarily by Jaan Tallinn (Skype co-founder, AI safety advocate). In 2025, SFF allocated $34.33 million to organizations working on existential risk, with ~86% ($29M) going to AI-related projects. QURI represents one of several epistemic infrastructure investments in SFF’s portfolio.

The Future Fund’s 2022 grant was affected by the FTX collapse, which eliminated potential future funding from that source. QURI has since diversified funding through LTFF and individual donors.

Organizational Structure

Aspect	Details
Legal Status	501(c)(3) nonprofit (EIN: 84-3847921)
Fiscal Sponsor	Rethink Priorities Special Projects Program
Sponsorship Model	QURI maintains autonomy; RP provides operational support, fiduciary oversight, tax-exempt status
Team Size	≈3-5 core contributors
Key Contributors	Ozzie Gooen (founder), Sam Nolan, Nuño Sempere (Metaforecast), Michael Dickens (unit types)
Communication	EA Forecasting & Epistemics Slack (#squiggle-dev), QURI Substack
Code Repository	Monorepo at github.com/quantified-uncertainty/squiggle

Rethink Priorities Fiscal Sponsorship

QURI works with Rethink Priorities under their Special Projects Program, which provides:

Fiscal sponsorship and tax-exempt status
Financial administration and payroll
Recruitment support
Operational infrastructure

This model allows QURI to focus on tool development while RP handles administrative overhead. The arrangement is common in the EA ecosystem for small research organizations.

Research Collaborations

Arb Research Collaboration

QURI collaborated with Arb Research on their 2025 technical AI safety review, building the interactive shallowreview.ai website. The Shallow Review is “a shallow-by-design review of technical AI safety research in 2025: 800+ papers and posts across 80+ research agendas.”

The website was created by QURI (Ozzie and Tomáš) to make the research more navigable. The review covers alignment, control, capability restraint, and risk awareness work, processing every arXiv paper on alignment and all Alignment Forum posts. The project was funded by Coefficient Giving.

GiveWell Cost-Effectiveness Quantification

Multiple projects have used Squiggle to add uncertainty quantification to GiveWell’s cost-effectiveness analyses:

Project	Authors	Effort	Key Finding
GiveDirectly CEA	Sam Nolan	≈40 hours	Mean cost to double consumption: $469 (95% CI: $131-$1,185)
Against Malaria Foundation	Various	Research project	GiveWell point estimate $7,759 vs. Squiggle mean $6,980
GiveWell Change Our Mind Submission	Sam Nolan, Hannah Rokebrand, Tanae Rao	≈300 hours	Full CEA uncertainty quantification

These projects demonstrate Squiggle’s utility for making cost-effectiveness analyses more transparent and explorable.

EA/Rationalist Community Integration

QURI tools are widely used across the effective altruism and rationalist communities:

Use Case	Tools Used	Example Organizations
Cost-effectiveness analyses	Squiggle, Squiggle Hub	GiveWell evaluators, EA Funds
Fermi estimates for cause prioritization	SquiggleAI, Squiggle	Coefficient Giving, 80,000 Hours
AI timeline modeling	Squiggle	AI safety researchers
Forecast aggregation	Metaforecast	Forecasting community
Writing quality checks	RoastMyPost	EA Forum authors

Comparison with Similar Tools

Tool	Focus	Strengths	Limitations	Learning Curve
Squiggle	Probabilistic estimation	Native distributions, web-based, readable syntax	No Bayesian inference, smaller ecosystem	Low-Medium
Guesstimate	Spreadsheet Monte Carlo	Familiar spreadsheet UI, 5,000 simulations	Less programmable, limited functions	Low
Stan	Bayesian inference	Powerful MCMC, HMC sampling	Steep learning curve, slower iteration	High
PyMC	Bayesian Python	Full Python ecosystem, Theano/JAX backend	Requires Python expertise	Medium-High
WebPPL	Probabilistic programming	Inference, conditioning	Academic focus, limited tooling	Medium
Excel	General spreadsheets	Ubiquitous, familiar	Poor uncertainty support, no distributions	Low
Causal	Business modeling	Scenario planning, team collaboration	Less probabilistic focus	Low

When to Use Squiggle vs. Alternatives

Use Squiggle when:

You need intuition-driven estimation without much data
Rapid prototyping of uncertainty models is priority
Web-based sharing and collaboration is important
You want readable, auditable probabilistic code

Use Stan/PyMC when:

You have data and need Bayesian inference
Model complexity requires advanced MCMC methods
Performance on large models is critical

Use Guesstimate when:

Spreadsheet interface is preferred
Quick one-off calculations
Non-programmers need to contribute

Use Cases

Cost-Effectiveness Analysis

Organizations use Squiggle for intervention comparisons with explicit uncertainty:

Use Case	Example	Typical Model Size
Charity evaluation	GiveDirectly, AMF cost per life saved	200-500 lines
AI safety interventions	Research funding, field-building ROI	150-400 lines
Policy cost-benefit	Regulation impacts, safety standards	200-600 lines
Career decisions	Expected value of different paths	100-300 lines

The GiveWell CEA quantification project demonstrated that Squiggle can make charity evaluations more transparent by showing full probability distributions rather than point estimates.

Forecasting Models

Squiggle enables structured forecasts with explicit uncertainty:

AI timeline models: When will specific capabilities emerge?
Technology adoption curves: S-curves with uncertainty bounds
Risk scenario analysis: Probability-weighted outcome trees
Market sizing: Fermi estimates for business planning

Research Communication

Squiggle models embedded in publications enable:

Transparent assumptions: Every input visible and adjustable
Reproducible calculations: Same code produces same outputs
Interactive exploration: Readers can modify parameters
Sensitivity analysis: Identify which inputs matter most

Academic and Policy Applications

While primarily used in EA/rationalist communities, Squiggle has potential applications in:

Academic research requiring uncertainty quantification
Policy analysis requiring transparent assumptions
Risk assessment in regulated industries
Teaching probabilistic reasoning

QURI Timeline

Year	Milestone	Significance
2016	Guesstimate launched	First Monte Carlo spreadsheet for EA community
2017-2019	Gooen at FHI	Forecasting infrastructure research
2019	QURI founded	501(c)(3) nonprofit established
2020	Squiggle Early Access	First public release of Squiggle language
2021	Metaforecast launched	Forecast aggregation across 10+ platforms
2022	$650K+ SFF funding	Major funding milestone
2022	Future Fund grant ($200K)	Expansion funding (pre-FTX)
2023	Squiggle 0.8.x	Performance improvements, Hub integration
2024	Squiggle Hub launch	Collaborative model platform
2024	SquiggleAI released	LLM-powered model generation
2024	RoastMyPost launch	LLM blog evaluation tool
2025	Squiggle 0.10.0	Major release with type inference, Web Workers
2025	shallowreview.ai collaboration	Arb Research AI safety review website
2025	Fermi Competition	$300 prizes for creative Fermi estimates

Strengths and Limitations

Organizational Strengths

Purpose-built tools: Each product addresses specific epistemic needs
Accessible: Browser-based, no installation needed
Open Source: All code MIT licensed, community contributions welcome
AI Integration: SquiggleAI and RoastMyPost leverage frontier LLMs
Community: Active EA/rationalist user base, responsive to feedback
Ecosystem approach: Tools work together (Squiggle + Hub + AI + Metaforecast)

Organizational Limitations

Niche adoption: Limited use outside EA/rationalist communities
Small team: ~3-5 core contributors limits development velocity
Funding dependency: Reliant on EA-adjacent funders (SFF, LTFF)
Learning curve: New syntax requires investment to learn
Documentation gaps: Some features under-documented
Ecosystem size: Fewer libraries than general-purpose languages

Strategic Challenges

Challenge	Description	Mitigation
Adoption ceiling	EA community is finite	Explore academic/policy applications
LLM competition	General LLMs can do Fermi estimates	Integrate LLMs (SquiggleAI) rather than compete
Sustainability	Small team, concentrated funding	Fiscal sponsorship, diverse funders
Feature scope	Pressure to add features vs. maintain simplicity	Clear design philosophy prioritizing estimation

Future Directions

Based on QURI’s development trajectory and stated priorities:

Near-Term (2025-2026)

Squiggle language improvements: Continued type system development, performance optimization
SquiggleAI expansion: Support for more LLM providers, longer context windows
Squiggle Hub features: Better collaboration tools, API improvements
Community building: More Fermi competitions, educational content

Medium-Term

Academic adoption: Partnerships with universities for teaching probabilistic reasoning
Policy applications: Tools for government cost-benefit analysis
Integration depth: Tighter connections between Squiggle, Metaforecast, and RoastMyPost

QURI (Quantified Uncertainty Research Institute)

Quick Assessment

Organization Details

Overview

Founder: Ozzie Gooen

Background and Career

Notable Contributions

Core Projects

Squiggle Language

Design Philosophy

Core Features

Distribution Creation Syntax

Squiggle Code Examples

Version History

Squiggle 0.10.0 Deep Dive

Strengths and Limitations

SquiggleAI

How It Works

Model Support

Capabilities

SquiggleAI Outputs

Integration with Squiggle Hub

Squiggle Hub

Core Features

Groups and Collections

Notable Models and Use Cases

Guesstimate Legacy

Metaforecast

Purpose and Design

Platforms Aggregated

Features

Integrations

2024 US Election Performance

Additional Tools

RoastMyPost

Fermi Competition

Squigglepy (Python Implementation)

Funding and Organization

Funding History

Funding Context

Organizational Structure

Rethink Priorities Fiscal Sponsorship

Research Collaborations

Arb Research Collaboration

GiveWell Cost-Effectiveness Quantification

EA/Rationalist Community Integration

Comparison with Similar Tools

When to Use Squiggle vs. Alternatives

Use Cases

Cost-Effectiveness Analysis

Forecasting Models

Research Communication

Academic and Policy Applications

QURI Timeline

Strengths and Limitations

Organizational Strengths

Organizational Limitations

Strategic Challenges

Future Directions

Near-Term (2025-2026)

Medium-Term

Strategic Questions

Sources and Citations

Primary Sources

Squiggle Documentation

EA Forum Posts

Funding and Organization

Collaborations

External Links