Verified publication track
Verified is an interactives track for papers whose core contribution can be tested. Authors submit scoped claims, runnable evidence, and a pinned environment; interactives returns a public record of what ran, what passed, and what remains out of scope.
Authorship is not the gate. The verification record is.
Verified submissions are being prepared; this page defines the track, not an open intake form.
Contract
Verified does not certify an entire paper. It certifies named claims, checked under named conditions, with the evidence and limits attached.
01
The author's identity, reputation, or writing process is not the gate. The declared claim is.
02
A Verified badge applies to specific claims, not to every sentence in the paper.
03
Tests, fixtures, seeds, data references, and environment specs carry the evidence.
Protocol
The workflow turns a paper's testable contributions into a record a reader can inspect: claim, evidence, environment, run, verdict, limits.
State one contribution as a specific, testable claim.
Attach tests, fixtures, seeds, baselines, data references, and a pinned environment.
Execute the package in a network-isolated CPU sandbox with fixed resource limits.
Check whether the evidence supports the stated claim, not merely whether the tests pass.
Attach the verification record to claims that pass both gates.
Two gates
Gate 1 asks whether the evidence ran. Gate 2 asks whether that evidence is adequate for the claim. The second gate is where the badge earns its value.
The submitted package runs in a pinned, deterministic, network-isolated environment. This gate produces the raw execution record: logs, exit status, resource use, and reproducibility signals.
Outputs
Passing tests are not enough. The faithfulness gate checks for tautological tests, cherry-picked fixtures, overfit answers, and manuscript claims that exceed the evidence.
Outputs
Verified result
Passing submissions are published with the claim, environment, outputs, verdict, and limits attached. Failed runs return logs and costs, but not the Verified publication outcome.
sortpartial returns a sorted array and stays under the stated comparison budget on the pinned workload described in C1.
def test_matches_oracle():
for seed in range(10_000):
xs = random_ints(seed, n=1_000)
assert sortpartial(xs) == sorted(xs)
def test_comparison_budget():
ops = count_comparisons(sortpartial, n=100_000)
assert ops <= 3 * 100_000 * log2(100_000)Scope and policy
Verified starts with claims that machines can actually check. The same boundary applies commercially: authors pay for verification work, never for acceptance.
Billed
Sandbox execution, verification work, and metered compute.
Not sold
Acceptance, badge status, or publication outcome.
Public record
The claim, environment, run result, verdict, and limits.
Executable publishing
For authors, Verified is a submission path for runnable evidence. For readers, it is a record of what was checked, how it ran, and what was not claimed.