KAAI Workshop 2026 — Foundation Models & Agentic Methods for Astrophysical Simulation Wean Hall 8330, Carnegie Mellon University, Pittsburgh · 31 August – 3 September 2026
M. D. Schwartz, “Resummation of the C-Parameter Sudakov Shoulder Using Effective Field Theory”, arXiv:2601.02484 (2026)
Two “co-scientist” systems published in Nature — May 2026
Off-the-shelf tools — Claude Code, Codex — are multiplying researcher productivity.
“Although LLMs cannot yet do original theoretical physics research autonomously, they can vastly accelerate the research done by experts.” — Matthew Schwartz (Harvard), Vibe Physics, Anthropic 2026
2025 was also the year of dedicated co-scientist systems, each generating and refining its own hypotheses in a closed loop. Google’s AI Co-Scientist proposed new therapeutic targets for liver fibrosis; Robin (FutureHouse) identified a drug candidate for dry AMD.
NeurIPS 2025
Open competition on weak-gravitational-lensing measurement — one of the hardest inference problems in cosmology.
I’ve worked on this problem for 7 years. You could say I’m somewhat of an expert…
Winning teams — NeurIPS 2025
METR, Task-Completion Time Horizons (metr.org/time-horizons, May 2026 snapshot, CC-BY) — Claude Mythos Preview at ≥16 h.
AI task horizons are doubling every ~89 days (~17×/year, METR TH 1.1, post-2024 window). Today’s “noisy” or “mildly useful” outputs won’t stay that way. Build for where models will be in a year, not where they are today.
Denario, Kosmos, Sakana — all tightly coupled to yesterday’s models. Denario is already obsolete 6 months later. As models improve, these AI co-scientist systems are replaced wholesale.
clawRxiv — AI agents are already publishing papers around the clock
ICML 2026 — 24,371 full submissions, more than double 2025 (12,107)
arXiv — now rejects unreviewed review & position papers (AI slop)
Peer review — hundreds of reviewers broke the rules on LLM use
Lanusse & Parker · May 2026
In this position paper, published earlier this year, we picture a future of accelerated research, where every scientist works with AI assistance.
Our outlook is resolutely optimistic: this future can unleash curiosity, open up far more ambitious questions, and multiply the rate at which results come into being.
But that same speed makes every result harder to check: a gap is opening between what gets produced and what can be verified.
And so none of those benefits will materialize if we can no longer tell what is trustworthy, what we can build on, from what is not.
… and that need for trust shows up at every stage in the life of a result:
When the scientist reviews what their agent has just produced.
When a reviewer has to judge an analysis they did not run themselves.
When a third party later sets out to reproduce or extend the work.
The question
It all starts with the form a scientific result takes.
Code
Executable but opaque.
Buried assumptions, no intent.
Missing link
Decisions, assumptions,
evidence, provenance
Paper
The traditional format.
Legible but lossy — the analysis cannot be regenerated.
A result can be vetted efficiently — by a human or by a machine — if three conditions are met:
Every plot, number, and claim ties back to the data, code, and decisions that produced it — eliminating fabricated results without requiring re-execution.
Code and artifacts, but also every consequential decision — estimator, prior, cutoff, dataset — and the reasoning behind it are inspectable.
Organized around the claims, decisions, and insights that matter — with direct paths down into the evidence and code behind any point.
Every result becomes a solid foundation for the next — science that compounds at scale, robust to the acceleration of scientific output.
Introducing
An open-source initiative to build tooling
for robust scientific research in the age of AI.
An international, open-source initiative — with scientists, for scientists — based at UC Berkeley (BIDS) and the CNRS (AISSAI).
Team
François
Lanusse
CNRS
Liam
Parker
UC Berkeley
Alexandre
Boucaud
CNRS
Cail
Daley
CEA
Kangning
Diao
UC Berkeley
Nolan
Koblischke
U. Toronto
Kirstie
Whitaker
UC Berkeley
Advisors
Debbie Bard
Science Engagement &
Workflows Dept. Head
NERSC
Kyle Cranmer
Director,
Data Science Institute
UW–Madison
Fernando Pérez
Faculty Director,
Berkeley Inst. for Data Science
UC Berkeley
Uroš Seljak
Co‑Director, Berkeley Center
for Cosmological Physics
UC Berkeley
Associated centers
Philanthropic support
Eric & Wendy Schmidt
Milestones
Mid‑Jan 2026
Project inception
May 2026
Project launch
July 28–31, 2026
Agentic AI for Science Developer Summit · Berkeley
September 2026
· this month
First stable version of the stack
December 2026
Agentic AI for Science Developer Summit · Paris
Anthropic, Economic Index (2026) — what AI can do in principle (blue) vs. what it is actually used for (red). The gap is still enormous.
Tools that strengthen scientists rather than replace them.
Tuned to computation and data analysis, across many fields.
Edison Scientific, “Kosmos” (arXiv:2511.02824) — the world model: the exploration trace of a research project.
An open standard, not a closed proprietary platform.
Governed by the research community — universities, labs, and national facilities at the table.
Others are already moving — an open solution has to get there first.
Across six participant-led discussions, one tension kept resurfacing: results are becoming easier to generate than to select, verify, and attribute.
35
participants
4
days together
12
collaborative projects
Unconference sessions
Publication, verification, provenance, access, interfaces, and HPC.
Collaborative hacking
Prototypes for decision graphs, multiverses, figures, notebooks, and HPC.
Open specification work
Projects and discussions became RFCs that will inform ASTRA v1.
Final-day demonstrations
Teams shared working interfaces, benchmarks, and provenance tools.
An agent reports evidence for evolving dark energy after reviewing its own work for weeks. Should we believe it?
Stronger grounds for trust
Established pipeline
Preregistered choices
Blinded analysis
Applied to new data
Requires deeper scrutiny
New pipeline developed on the data
Choices changed as results appeared
Discarded paths shaped the outcome
Agent self-review is not enough
Many defensible choices—and the paths discarded along the way—can drift an analysis toward a preferred result.
Verification must ask both: Is the conclusion correct? and Was the scientific process sound?
An event log can be technically comprehensive and still be impossible for a human to review.
linked causal
view
Inspectable decision graph · linked to the trace
Consequential choices
What changed the analysis or conclusion?
Evidence & alternatives
What supported each decision—and what was rejected?
Human accountability
Who authorized or redirected the work?
Complete provenance underneath. Inspectable causal views on top.
Provenance can make scientific judgment visible—so credit and accountability follow the people who guide the work.
Agents can absorb more implementation work; scientists remain responsible for deciding what matters and what is credible.
Human contribution
concentrates in judgment
Implementation
becomes abundant
Frame worthwhile questions · validate claims · take responsibility
Steer the exploration · interpret anomalies · choose consequential paths
Implement the work · code · run · test · document
INCREASINGLY AGENT-ASSISTEDKeep the complete machine-readable record; let human-facing publications concentrate context and significance.
“Write less, so we can read more.”
FUTURE
Platform — Hosting & sharing infrastructure
COMING SOON
UI Layer — Visual interface for analyses
ALPHA — TECH PREVIEW
Agent Layer — portable agent skills for AI-assisted research
ALPHA — TECH PREVIEW
CLI & Tooling — Validation, execution, workflows, HPC
ALPHA — CORE
ASTRA — Agentic Schema for Transparent Research Analysis — Core specification format
Everything builds on ASTRA — the declarative spec that captures the scientific intent of an analysis. The layers above read from and write to this single source of truth.
Open source
BSD 3-Clause · co-developed in the open with the scientific community.
Technical deep dive
Agentic Schema for Transparent Research Analysis
Our open specification for structuring computational research — making analyses inspectable, reproducible, and legible to humans and agents alike.
A quick walkthrough of the schema.
Every spec declares what it needs, what it produces, and what choices it makes.
# astra.yaml — Iris classification, trimmed
version: "1.0"
name: "Iris Classification Study"
inputs:
- id: iris_data
type: data
source: "sklearn.datasets.load_iris"
outputs:
- id: accuracy
type: metric
recipe:
command: python src/evaluate.py
- id: confusion_matrix
type: figure
decisions:
scaling:
label: "Feature Scaling"
rationale: Scaling changes how features weigh in.
default: standard
options:
none: { label: "No Scaling" }
standard: { label: "StandardScaler" }
minmax: { label: "MinMaxScaler" }
model:
label: "Classification Model"
default: random_forest
options:
svm: { label: "Support Vector Machine" }
random_forest: { label: "Random Forest" }
logistic: { label: "Logistic Regression" }Inputs
Data sources or references to other ASTRA analyses (type: analysis). How projects compose into chains.
Outputs
Five kinds — metric, figure, table, data, report. Each carries an optional recipe (build rule).
Decisions
Named choice points with options, a default, and a rationale. Each option can carry its own description and links to supporting evidence.
Picking one option per decision gives you a universe — a single, executable configuration.
Every claim backed by evidence — either a quote from the literature, or an artifact produced by the analysis itself.
# Claims in, claims out — same shape, different direction
prior_insights:
scaling_svm:
claim: >-
Standard scaling consistently outperforms min-max
normalization for SVMs on tabular data.
created_at: "2026-03-12T09:00:00Z"
evidence:
- id: ev_paper
doi: "10.48550/arXiv.1706.03762"
quote:
exact: "Z-score normalization yielded higher accuracy."
location: { page: 8 }
findings:
best_model:
claim: Random Forest reaches 96.2% with standard scaling.
created_at: "2026-04-20T17:00:00Z"
derived: true
evidence:
- id: ev_rf_run
artifact: accuracy ← output of THIS analysis
quote:
exact: "accuracy = 0.962"
decisions:
scaling:
options:
standard:
insights: [scaling_svm] ← option cites prior insightPrior insights
Knowledge brought IN from literature. Evidence = doi + verbatim quote + page anchor. These inform your decisions.
Findings
Knowledge taken OUT of the analysis. Evidence = artifact (an output of this analysis) + quote. What the run produced.
Shared model
Both are the same Insight object — claim + evidence[]. Placement (prior vs. finding) sets the direction.
astra validate --verify-evidence fetches DOIs and checks quotes are real text. No fabricated citations.
How ASTRA turns methodological choices into an explorable analysis space.
Decisions
Each choice has named options with rationale and evidence.
Universe
One complete set of selections — a single path through decision space.
Multiverse
The full space of decision combinations — for testing robustness to analysis choices.
# universes/baseline.yaml
id: baseline
description: "Default configuration"
decisions:
scaling: standard
model: random_forest
test_size: smallA universe is just a YAML file — one option selected per decision. Running it produces all declared outputs.
The multiverse is the full space of decision combinations. Run alternative paths to test whether your conclusions are robust to the choices you made.
Purpose: robustness. Do your results hold when you swap a scaling method, change a model, or shift a prior? The multiverse tells you.
Building on Steegen, Tuerlinckx, Gelman & Vanpaemel, “Increasing Transparency Through a Multiverse Analysis,” Perspectives on Psychological Science 11(5), 702–712 (2016), doi:10.1177/1745691616658637; and Yu & Barter, Veridical Data Science: The Practice of Responsible Data Analysis and Decision Making (MIT Press, 2024) — PCS (Predictability, Computability, Stability) framework, vdsbook.com.
MyST is Jupyter’s open document engine for science — Markdown with equations, citations, cross-references and executable content, built into websites, PDFs, or the JATS XML journals actually ingest.
It is the engine behind Jupyter Book 2, and behind AGU’s Notebooks Now! — publishing computational notebooks as first-class elements of the scientific record.
MySTRA, our MyST plugin, lets the prose
reference the analysis instead of restating it:
Under the {astra}`decisions.bao_fit_method`
we measure α = {astra:value}`outputs.alpha_iso`.
:::{astra} outputs.bao_fit_plot
:::The decision links to its full record, the number is read live from the result table, and the figure is the materialized artifact. One source of truth — re-run the analysis and the paper updates itself.
mystviewer.org/eiffl/myst_proto — DESI DR1 BAO measurement reproduction
A human-authored argument beside a navigable ASTRA research record.
We use the in the baseline universe. This assumption is supported by the verified methodology record.
For the LRG3+ELG1 bin, the angle-averaged distance is . The complete result lives in .
◆ output · hubble_diagram
Selected: Gaussian. Alternative: flat. Five prior insights support the selected option.
Source: outputs.bao_distance_table
Cell: DV_over_rd · tracer=lrg3_elg1
Machine-readable analyses, artifact references, decisions, alternatives, findings, and evidence—composable across projects.
Human-authored argument, context, mathematics, citations, and composition—without duplicating the scientific record.
Select any result → inspect every declared input, decision, assumption, alternative, and evidence source behind it.
Technical deep dive
The execution layer for ASTRA.
lc init · lc materialize · lc status
Turns an astra.yaml into enforced, reproducible execution —
making it mechanically impossible for an agent to produce a result that isn’t traced.
The agent describes the analysis and drives the CLI — but it never writes a result itself. Every artifact is made by the engine, under a sandbox, and committed with the record of how it was made.
# The daily loop
$ lc init my-analysis # idempotent: converges a project
✓ astra.yaml · pyproject.toml · uv.lock · .venv
✓ git + git-annex, so results travel with the repo
$ lc materialize # every output, every universe
$ lc materialize robust/fit # one universe’s output
→ validate + resolve · Dask graph · sandboxed recipes
→ each result committed with its manifest + run record
$ lc status # reads only; always exits 0
sandbox: landlock (fs: declared, network: allowed)
· current baseline/fit a3f1f11
· behind robust/fit 00cc14e earlier environment
! stale baseline/fit_plot 9f2ab04 edited by handA tracked, reproducible environment
lc init converges the project on uv: the agent adds dependencies with ordinary uv add, and the resulting uv.lock is committed beside the spec. The environment an analysis ran in is part of the record.
Recipes run sealed in
lc materialize executes every recipe under a sandbox — Landlock on Linux, Seatbelt on macOS. A recipe cannot read a file that isn’t a declared input, and cannot reach a dependency outside the locked environment. Undeclared reads fail loudly instead of passing silently.
A hand-edited result gives itself away
Each output is committed with a content-addressed manifest and a replayable run record. Touch one outside its own run and it is a foreign write: the next lc status calls it stale and names the offending commit. The agent cannot write a number into a figure the engine didn’t produce.
An analysis is only reproducible if it stays executable — on the machine it was written on, and on whatever the next person has.
Laptop & workstation
LocalCluster
Write and iterate on the machine in front of you, using every core it has.
HPC — SLURM
pilot jobs
The very same lc materialize, with one Dask scheduler running inside the allocation. Nobody writes a batch script — least of all the agent.
Kubernetes & cloud
dask‑kubernetes · JupyterHub
New backends attach behind the same interface — a Kubernetes cluster, a JupyterHub deployment — and the analysis itself never changes.
One plugin — lightcone — carries everything an agent needs to work on a project:
how to scope it, how to record it, and how to run it.
# one-time: register the marketplace
$ claude plugin marketplace add LightconeResearch/agent-skills
# install the plugin — it bundles everything
$ claude plugin install lightcone@lightcone-research
# then just ask
/lightcone:lightcone scope, run, debug, publish
/lightcone:astra the spec itself
Open Agent Skills standard — the same plugin in every harness.
github.com/LightconeResearch/agent‑skills
Scopes the project with you
It opens with an interview, not code: what the question actually is, which choices are genuinely open, and what the literature already settles — searched and quoted.
Documents every decision for you
Each methodological choice lands in astra.yaml as a decision — its options, its rationale, its evidence. The documentation you never sit down to write.
Drives the CLI to make the artifacts
It writes the recipes, calls lc materialize, reads the status back and diagnoses failures — so every output comes out of the engine, with its provenance, never out of the conversation.
Let’s start a new analysis.
I’m setting the project up with lc init — astra.yaml becomes the durable structure of your research, right alongside the code.
What prior should we take on the optical depth τ?
I’m recording it as a decision in astra.yaml — with the alternatives you rejected and the evidence behind the one you kept.
“The low-ℓ EE polarization likelihood provides the tightest CMB-only constraint on the reionization optical depth, τ = 0.054 ± 0.007.”
Planck Collaboration, 2020 · A&A 641, A6
Go ahead and produce the results.
lc materialize — I read the inputs, the choices and the outputs from astra.yaml, and the engine makes each one against that same structure.
Where does this figure come from?
Here is its full provenance: what produced it, what it consumed, and the choices that shaped it.
plot.pydataPut to the test
in observational cosmology — on the DESI survey
Building these tools with scientists, on a real research project — to make sure they hold up against the complexity of the real thing.
arXiv:2404.03000
DESI 2024 III: Baryon Acoustic Oscillations from Galaxies and Quasars.
Interactive render of the DESI survey — NOIRLab/DESI
Lightcone
DESI 2024 III
From raw catalogs to the Hubble diagram.
damping_prior_gaussian_fiducial
Tight Gaussian priors (width 1–2 h−1 Mpc) on the BAO damping parameters centred on theory values are recommended; fully fixing them risks biases and flat priors can also bias α and weaken constraints on noisy data.
damping_prior_width_justification
Biases of order 0.1% in αiso and 0.2% in αAP appear when the damping parameters are misspecified by more than ∼2 h−1 Mpc from their fiducial values — justifying Gaussian priors of width 1–2 h−1 Mpc.
sigma_perp_parallel_formula
Σ⊥ and Σ∥ follow Σ⊥ = Σ0·G, Σ∥ = Σ0·G·(1+f), with Σ0 = 12.4 h−1 Mpc for σ8 = 0.9 at z = 0 — the theoretical prediction that anchors the Gaussian prior centres.
damping_prior_gaussian_fiducial
Flat priors can bias α and weaken constraints on noisy data; the same Chen+ 2024 analysis that motivates the Gaussian fiducial also rules flat priors out as the headline choice.
bias_weak_effect_on_sigma
Because Lagrangian displacements are dominated by bulk flows, galaxy bias has only a small effect on Σ⊥/Σ∥. With a flat prior the fit has no anchor against this near-degenerate direction.
Insight
damping_prior_gaussian_fiducial
Tight Gaussian priors (width 1–2 h−1 Mpc) on the BAO damping parameters centred on theory values are recommended; fully fixing them risks biases and flat priors can also bias α and weaken constraints on noisy data.
● Evidence 1 · verbatim quote
We recommend choosing the BAO damping parameters based on theoretical calculations and allowing these to vary within reasonably tight priors (1−2 h−1 Mpc).
page 22 · right column · bullet (iii)
Reference
Three Members of Technical Staff — distinct roles, one team. Joining now means real ownership of the abstractions, tools, and norms that will define Lightcone Research.
Invent how scientists actually work with research agents — scoping problems, steering long-running work, reading progress and uncertainty, and trusting what gets produced.
Build the infrastructure underneath it — open specifications, agentic tooling, evaluation systems, scientific workflows, HPC integration, provenance and reproducibility.
A broad, founding role spanning agentic systems, scientific infrastructure and the researcher experience — building the European side of Lightcone Research.
Berkeley at BIDS, in the new Bakar Gateway building — surrounded by the Scientific Python and Jupyter communities. Paris in the city, alongside our CNRS AISSAI colleagues.
lightconeresearch.org/careers or write to info@lightconeresearch.org
Work with us
Help us build the open substrate for scientific research
in the age of AI.
Pilot it
Bring us a real analysis. We are looking for groups willing to put the stack against their own science.
Shape the spec
ASTRA is developed in the open, by RFC. The schema is still alpha — the time to argue with it is now.
Join the next summit
The second AAI4Science Developer Summit lands in Paris, December 2026.
info@lightconeresearch.org · lightconeresearch.org · Discord
