Know whether a paper is buildable, what credentials it needs, and which claims can be tested before spending credits.
Feasibility verdict<3mLIVE LABLISCENSKQMFive agents. One evidence trail.
Paste a Paper. Watch It Run.
Know What Actually Holds.
Watch five agents rebuild the experiment, verify every claim,
and turn the paper into a reusable research twin.
RECRA LABFive agents. One office.
OFFICE DEMOOpening the research office…
DEMO SEQUENCE
Meet the five-agent research team
LISCENSKQM
FIVE AGENTSOne recorded research teamEvery research paper becomes a living, verifiable, remixable experiment—with measured evidence behind every green verdict.
Librarian, Scout, Engineer, Skeptic, and Quartermaster each own a separate part of the evidence chain.
Stochastic claims are repeated so variance is visible instead of buried behind a single favorable run.
Standard minimumn≥3The live lab is a renderer of real work. No fake progress and no theatrical busyness.
From static PDF
to living evidence
Feasibility, autonomous reconstruction, independent verification,
and transfer experiments in one system.
Feasibility before spend
Paste an arXiv paper. Recra finds the code, extracts numeric claims, and tells you honestly what can be built.
A watchable live lab
Five agent scientists read, fetch, rebuild, audit, and preserve the paper. Every visible action maps to a real backend event.
Measured verification
A mechanically separate Skeptic compares paper values with instrumented results across independent runs and confidence intervals.
A reusable research twin
Keep the environment, spec, assumptions, artifacts, and run history—then swap the model, data, or configuration on your stack.
From paper to a
measured verdict
Recra scopes the claims before the build, reconstructs the experiment, and returns evidence you can inspect and reuse.
Request accessSubmit the paper
Paste an arXiv URL. Recra locates the repository and extracts testable numeric claims with their source locations.
Approve the scope
See feasibility, assumptions, estimated time, and the exact claims Recra will attempt before the build starts.
Watch the lab work
Five agents rebuild the method in an isolated environment while every action is recorded and inspectable.
Get the twin
Receive claim-level verdicts, fidelity labels, artifacts, and a workspace ready for transfer runs.
A research twin keeps the method alive
after the PDF ends
Every build leaves behind a portable Research Spec, environment snapshot, assumptions log, run history, verdicts, and artifacts.
Code, data pipeline, dependencies, and snapshots restored inside an isolated sandbox.
Watch the build.
Trust the event stream.
Switch from the live lab to claims, assumptions, raw events, artifacts,
replays, and the final report without losing the evidence trail.
SGC
SGC 2019Standard verification
Seed-level verificationMeasured results against paper values
95% CIClaim verdicts•••
3 / 3reproduced
- Cora .810
- Citeseer .719
- Pubmed .789
- Variance Pass
Five-agent research lab
Agents assigned5
Independent runs3
FidelityStandard
Green means measured.
Never vibes.
Recra is designed around a few hard rules: fidelity is always labeled, blocked work fails loudly, and the agent that builds the experiment never grades its own answer.
Loud blocked beats silent wrong. If the build needs a credential, dataset, or decision, it stops and tells you exactly why.
The Skeptic reads instrumented artifacts, recomputes metrics, audits variance, and can disagree with the Engineer.
FREE: 1 QUICK BUILD · 10 FULL PUBLIC TWIN VIEWS
Start with one paper.
Scale to a research program.
All builds run on Recra-managed credits. No API keys to configure,
and paid work stays private unless you choose to publish it.
Author
$149/paperPre-flight a paper before reviewers—or the internet—do it for you.
Verify my paperSTANDARD VERIFICATION- Free Quick self-check first
- Paid multi-run verification
- Full Skeptic findings
- Claim-level evidence report
- “Runs on Recra” badge if passed
Pro
$39/month + creditsFor researchers and builders evaluating published techniques.
Request accessPRIVATE RESEARCH WORKSPACE- Private research twins
- Standard verification tier
- Model and data transfer runs
- Unlimited registry viewing
- Credit top-ups when needed
Team
$99/seat / monthFor research and applied-AI teams building a shared evidence base.
Talk to usEVERYTHING IN PRO, PLUS- Pooled build credits
- Shared private registry
- Roles and permissions
- SSO and audit logs
- Usage and cost controls
Enterprise
CustomFor venues, publishers, and organizations running verification at scale.
Contact usORGANIZATION-WIDE- Batch paper screening
- Custom model endpoints
- Dedicated infrastructure
- Field-pack support
- Service guarantees
Common Questions
Recra fetches the paper, locates its code, extracts the numeric claims and source locations, then returns an honest feasibility verdict before any paid build begins.
It means Recra measured the claim numerically against the paper’s reported value. Standard verification uses at least three runs for stochastic experiments, confidence intervals, a fidelity label, and links to the exact artifacts.
The Skeptic is mechanically separated from the Engineer and reads instrumented run artifacts—not the Engineer’s prose. It audits parsing, variance, robustness, and baseline fairness before issuing a verdict.
A twin is the paper’s Research Spec, executable environment, assumptions, artifacts, versioned run history, and verification verdicts in one reusable workspace. It can be exported; the artifact is yours.
Yes. Transfer runs can swap the model, dataset, or configuration. Paid results and uploaded data are private by default and only enter the public registry if you choose to publish them.