CEVRA Materials

Open AI materials research
Generating crystals, publishing evidence, keeping failures visible
Building on @RobinhoodCrypto

[CEVRA Materials @CEVRAMaterials]

CEVRA — Official Contract Address
Cevra Materials on Robinhood Chain
CA ↓
0xb8d5eb849bed5a9b3d88f5034e0f800a48dbc7a6

CEVRA builds an open, traceable workflow for AI-driven materials research — from generated crystals to structural screening, relaxation and inspectable evidence
Our research records remain public, including failed candidates and the limits of our findings
Website → https://t.co/0virxQaOkG
Verify the network and full contract address before interacting

[CEVRA Materials @CEVRAMaterials]

One good generation run tells us very little
CEVRA runs independent seeds and compares how many structures pass geometry checks, remain distinct and complete surrogate relaxation
The differences between runs are part of the result
Follow the campaigns → https://t.co/lQgfWO2GHH

[CEVRA Materials @CEVRAMaterials]

CEVRA’s research is easier to inspect now
Our homepage now connects the full screening path to published records — from generated structures and geometry checks to MACE relaxation, including candidates that failed along the way
Each failure category opens the matching records in Candidate Atlas, with evidence timestamps and downloadable files for review
The second campaign is still running, so the counts reflect what has actually been published so far
Explore the record → https://t.co/0virxQagv8

[CEVRA Materials @CEVRAMaterials]

A low MACE energy is not a stability certificate
Raw energies from different compositions or model checkpoints cannot be placed on one leaderboard
That is why CEVRA records the checkpoint, initial and final structure, energy change, atomic forces, optimizer steps and convergence outcome for each relaxation
The useful question is what the same candidate did under a specified method
The evidence is open
https://t.co/rYcEnQsUtr

[CEVRA Materials @CEVRAMaterials]

We have taken one CEVRA pilot candidate beyond surrogate screening into fixed-geometry DFT calculations
Three electronic calculations converged under the registered settings
That is a useful next layer of evidence, with a clear limit: it does not yet establish phase stability or synthesis
The real milestone is a workflow where each stronger claim requires a stronger, inspectable calculation
See the assessment record
https://t.co/MXdGIaxyZx

[CEVRA Materials @CEVRAMaterials]

We’re updating the CEVRA website to make the research easier to inspect
The next homepage section will show where each published candidate stands — generated, structurally valid, unique within its campaign, or surrogate screened
It will also connect failure counts to the underlying records and distinguish the latest evidence archive from the time the page received its data
Research progress should be visible, dated, and traceable
https://t.co/0virxQaOkG

[CEVRA Materials @CEVRAMaterials]

Inside the CEVRA research queue
Our second registered campaign is expanding the materials pipeline across independent generation seeds
At the published snapshot on September 24, 12:41 UTC:
— 640 candidate records published
— 373 passed the initial structural checks
— 363 reached surrogate-screened status
— 277 were recorded as failed, including scope exclusions and downstream screening failures
Each candidate follows the same documented sequence:
MatterGen generation → geometry checks → deduplication → MACE relaxation → evidence archive
Passing geometry checks is only one step
A candidate can still fail during relaxation or subsequent checks — those outcomes remain in the record
We are examining which chemical systems pass screening, where the workflow fails, and how outcomes differ between seeds
The published results are interim
Surrogate screening does not establish material stability
Explore the queue, candidates and evidence
https://t.co/DF1dFYyGvb

[CEVRA Materials @CEVRAMaterials]

We’ve burned the entire dev allocation

https://t.co/ws9oFrQG4E

[CEVRA Materials @CEVRAMaterials]

CEVRA — Research progress
With our CA now announced, here is where the research stands
664 candidate records published across two campaigns
381 candidates recorded as surrogate screened
Behind those numbers is a working pipeline
MatterGen generates crystal candidates
pymatgen checks geometry and identifies duplicates
MACE evaluates and relaxes eligible structures
The resulting structures, measurements and failure records enter our public archive
THE FIRST CAMPAIGN
Our completed pilot generated 24 candidates across three independent seeds
18 passed the registered screening workflow and converged under MACE relaxation
6 were excluded by the campaign’s scope checks
All 24 remain in the research record, including the exclusions
THE SECOND CAMPAIGN
Our second registered campaign has published 640 candidate records so far
363 are recorded as surrogate screened
277 are recorded as failed, including scope exclusions and downstream screening failures
These are interim figures from the published snapshot, not the final campaign results
FIRST DFT ASSESSMENT
One pilot candidate has completed three fixed-geometry DFT self-consistent-field calculations
All three reached electronic convergence, and the selected cutoff and k-point comparisons passed the registered limited checks
This is an initial DFT assessment
Thermodynamic stability and experimental synthesis remain unverified
External reference matching has also begun, with incomplete coverage explicitly recorded
An unmatched candidate is not automatically a globally new material
OPEN THE RECORD
The website now lets you inspect actual atomic coordinates, compare generated and relaxed structures, explore campaign outcomes and download evidence files with SHA-256 hashes
The next work is to complete the registered campaign, review its failures and determine which candidates justify deeper calculations
The CA announcement marks a public milestone
Our research progress will continue to be measured through inspectable results
Explore CEVRA → https://t.co/0virxQaOkG

[CEVRA Materials @CEVRAMaterials]

CEVRA Materials — The tools behind the research
MatterGen — Crystal generation
https://t.co/wyBHUhUcfR
pymatgen — Structure analysis and matching
https://t.co/Auqx14DjNT
MACE — Machine-learned interatomic potentials
https://t.co/PnPLSo4IFU
MACE foundation models
https://t.co/l3K5XXWoah
ASE — Atomistic simulation and geometry optimization
https://t.co/TH5Gr3WdpV
━━━━━━━━━━━━━━━━━━
A generated crystal is the beginning of a research question
Can it pass geometric checks?
Is it distinct within the candidate set?
Can it relax under a specified interatomic potential?
Does it justify more expensive first-principles calculations?
CEVRA is building a public computational materials research pipeline around those questions
We use official pretrained MatterGen models to propose inorganic crystal structures, pymatgen to inspect geometry and identify duplicates, and a specified MACE checkpoint with ASE to evaluate energies, atomic forces and structural relaxation
Each stage produces a record that can be inspected alongside the original candidate
━━━━━━━━━━━━━━━━━━
OUR FIRST COMPLETED CAMPAIGN
The first pilot generated 24 candidates across three independent seeds
18 passed the registered screening workflow and converged under MACE relaxation
6 were excluded by the campaign’s scope checks
All 24 remain in the archive
For the published candidates, the research record connects generation settings and seeds to raw structures, screening outcomes, model provenance, relaxation measurements and downloadable evidence
Failures and exclusions are part of the result
Our second registered campaign extends this workflow across additional independent seeds, with published outcomes available through the website
━━━━━━━━━━━━━━━━━━
MOVING TOWARD FIRST-PRINCIPLES ASSESSMENT
One pilot candidate has completed three fixed-geometry DFT self-consistent-field calculations
All three reached electronic convergence, and the selected cutoff and k-point comparisons passed the registered limited checks
These calculations assess the electronic solution at the chosen geometry
They do not establish thermodynamic stability or experimental synthesis
External reference matching has also begun
Coverage remains incomplete, so an unmatched structure is described relative to the queried reference set
MACE screening, DFT assessment and experimental confirmation represent different levels of evidence
Our records keep those distinctions visible
━━━━━━━━━━━━━━━━━━
WHAT YOU CAN INSPECT TODAY
The CEVRA website connects the research workflow to its underlying files
Explore candidate structures using actual atomic coordinates
Compare generated and relaxed geometries
Inspect cell parameters, energies, forces and optimization steps
Review campaign statistics and failure categories
Download evidence and check its SHA-256 hashes
The visualizations are views into recorded calculations
The scientific claims must remain traceable to those calculations
━━━━━━━━━━━━━━━━━━
THE ROLE OF ROBINHOOD CHAIN
Robinhood Chain is the planned public registration and coordination layer for CEVRA
The intended first step is to register protocol hashes and evidence roots, linking on-chain records to downloadable research evidence
Future participation mechanisms may support research proposals, transparent compute funding and independently reviewed reproduction work
Scientific computation and large evidence files remain off-chain
Core research evidence and failure records remain publicly accessible
A blockchain record can establish that a particular hash was submitted
Scientific confidence still depends on the methods, the evidence and independent scrutiny
━━━━━━━━━━━━━━━━━━
CEVRA’s goal is to make computational materials discovery open to examination — from the first generated structure to the decision about what deserves further study
Explore the research
https://t.co/0virxQaOkG

[CEVRA Materials @CEVRAMaterials]

Introducing CEVRA Materials
AI can generate crystals — which ones survive scrutiny?
We’re building an open pipeline for crystal generation, structural screening and relaxation
Real runs, traceable evidence, failures included