Evidence that travels with the claim.
A result is more useful when its inputs, its checking mechanism, and its limits arrive alongside it. Verifier Standard Labs, Inc. builds the standard and the tooling that keep those together — and keep what was never established visibly unknown.
pip install verifier-standard
Python 3.10+ • five releases published • view on PyPI ↗
Each figure above is reproducible from the install command. The run that produced them is reproduced in full below.
A run you can repeat
On 17 September 2026 the published package was installed into a clean environment and its adversarial demonstration was executed twice. What follows is that output. It is not a mockup, and no part of this page simulates a verification.
The recorded session below executes 1.4.2, installed directly from the Python Package Index into an isolated virtual environment.
$ pip install verifier-standard==1.4.2
$ vstd demo
VSTD flagship adversarial demo
4/4 scenarios behaved as required.
[DEMO OK] Valid-looking proof, wrong artifact
expected: REJECTED
observed: REJECTED
[DEMO OK] Bound exhausted without a false answer
expected: ACCEPTED/UNKNOWN
observed: ACCEPTED/UNKNOWN
[DEMO OK] Inflated verification-cost claim
expected: REJECTED
observed: REJECTED
[DEMO OK] Revoked ancestor behind valid descendants
expected: GRAPH-CANDIDATE-0; REVOKED blocker
observed: GRAPH-CANDIDATE-0; REVOKEDReproducibility check
The same command was run twice and both outputs were hashed.
- Command
vstd demo --json- Runs
- 2, independent
- Output size
- 37,393 bytes, both runs
- SHA-256
cdccc0458870c99a4509b0bea2eea0202149e3055394ee2e35747e49a3d552b6- Result
- IDENTICAL
This establishes determinism across two runs on one interpreter and one operating system. It does not establish determinism across platforms or Python versions — an open question this company intends to measure rather than assert.
What each scenario actually asks
Each is an attack the checker is supposed to catch. The question, and the machine's own explanation, are reproduced verbatim.
Does the grounded proof refer to the same artifact as the claim?
grounding rejected: clause 2 claims role 'member' is about 'artifact:somebody-else', but variable 2 is grounded in a fact about 'artifact:ancestor-1'
Can bounded exhaustion stay explicit instead of becoming PASS or FAIL?
indeterminacy transcript accepted: PROOF_BOUND_EXCEEDED
Can a linear Horn check be dressed up as general resolution?
formula is Horn, so tier UP is admissible and therefore mandatory; certificate declares RES. A linear-time check may not be dressed in general-resolution machinery.
Does a poisoned transitive ancestor cap the collection's profile?
collection:demo computes to candidate Graph profile 0 from caller-supplied ratings; conformance is not established. Graph profile 1 is refuted by: STATUS_ADMISSIBILITY: artifact:source is REVOKED.
Inspect the receipt bytes
The binding from the first scenario, exactly as the run emitted it — evidence root, policy root, declared bounds, and the hashes of the verifier that produced it.
wrong-artifactThese are bounded reference-implementation checks over the included specimens, not a claim of empirical truth, complete provenance, or adoption.
No independent third-party evaluation, customer, certification, or adoption is claimed anywhere on this page. Where something has not been established, it is left unknown rather than filled in.
The Verifier Standard (VSTD)
VSTD is a vendor-neutral, portable specification defining how computational systems state claims, delineate bounds, isolate assumptions, and emit reproducible evidence receipts.
How we label evidence
These labels state the basis of a statement. They are used throughout this site, and in the company's internal records, with the same meanings.
| Status | Meaning |
|---|---|
OBSERVED | Directly inspected through a current tool, environment, or command. |
REPRODUCED | Independently rerun under declared coordinates, yielding identical results. |
DOCUMENTED | Asserted by written specification or comment. Not assumed true. |
INFERRED | Derived from stated evidence and assumptions, not directly observed. |
UNKNOWN | Unmeasured or unestablished. Preserved as unknown, never zero-filled. |
The run above is REPRODUCED. The market projections any company puts in a pitch deck are INFERRED. Both belong on a website; only one of them should look like the other.
Fail-closed by construction
Missing evidence must not become a passing result. Refusing an operation because a prerequisite is absent does not, by itself, falsify the underlying proposition — and the difference is recorded rather than collapsed.
Input evidence: NOT_ATTACHED
Checking mechanism: NOT_RUN
Proposition result: UNKNOWN
Operation: REFUSED_PENDING_EVIDENCE
No successful verification is claimed.Independent research, inspectable evidence
Verifier Standard Labs, Inc. is a Delaware corporation working on tools that make the relationship between a claim and its support explicit — while keeping unestablished claims visibly unknown.
A small lab, stated plainly
The company is led by founder and chief architect Tyler Roost (@timelordraps), bringing together software engineering, research, and creative practice in one independent laboratory.
Automated tooling assists research and implementation. Scope and disclosure decisions are the founder's, and every published result must identify the evidence that supports it.
What a check does and does not settle
A digest identifies bytes. It does not establish that a scientific or business claim is true.
A passing check applies to its stated mechanism and scope. It does not imply independent validation, certification, or fitness for a purpose nobody checked. Saying so on a marketing page is the point, not a caveat buried beneath one.
Why Verifier Standard Labs, Inc. — Distancing and the Garden
We are often asked why the company was incorporated as Verifier Standard Labs, Inc. instead of Claimgarden, Inc., especially given that our primary commercial engine and main source of revenue early on will be generated from claimgarden.com.
The decision is rooted in community neutrality. ClaimGarden is meant to be a welcoming home for the broader community — not any one place captive to or associated with any single corporation. Founder Tyler Roost's purpose was to provide an open garden for computational advancements across the verification ecosystem to bloom and flourish freely. The best way to achieve that was deliberate institutional distancing between the corporate laboratory and the community hub.
Ambition far beyond a single platform
Verifier Standard Labs, Inc. represents far more than an operational attachment to ClaimGarden. Our research horizons and company goals are ambitious to say the least.
Beyond commercial hub services, the laboratory's enduring mandate spans foundational claim-boundary specifications (VSTD), symbolic ordinal computing architectures (ordinatics), closed formal mathematical universes (Hypermath in Lean 4), and unforgeable evidence verification pipelines. ClaimGarden is one fertile soil where verified computational advancements bloom; VSTD Labs is the deep research engine driving the foundational science forward.
The pipeline, and what stage it is at
Three stages turn an input into a receipt. The reference implementation covers them today for the specimens it ships with. Generalising them is the research programme, not a finished result.
Provenance
Binds an artifact to content-addressed roots, so later claims trace back to specific bytes rather than to a description of them.
- Content-addressed evidence roots
- Declared exclusions recorded, not dropped
- Revocation propagates to descendants
Adversarial audit
Checks the certificate against the claim, including whether the proof is even about the right artifact, and whether the declared cost tier is admissible.
- Grounding checked against the claim subject
- Bound exhaustion reported as UNKNOWN
- Inflated cost tiers refused
Canonical receipt
Emits a deterministic record of claim, artifact, mechanism, outcome, and limits, so two parties inspecting it reach the same bytes.
- Deterministic serialisation
- Verifier and parser hashes embedded
- Limits carried in the receipt itself
Artifact
Exact bytes and declared environment
Mechanism
A named check with stated assumptions
Outcome
Pass, fail, or explicitly unknown
VSTD receipt
Claim, artifact, mechanism, outcome, limits
The ecosystem
Three repositories, each at a different stage, each with its own documentation. They share one commitment: the evidence for a claim is published alongside the claim, and the limits of that evidence are stated rather than implied.
verifier
The reference implementation of the Verifier Standard. It produces receipts that bind a result to the bytes, environment, and checks that produced it, and records what was not established. Installs as verifier-standard.
ordinatics
Exact ordinal arithmetic, symbolic value operations, and bounded typed satisfaction for Python — the computable part of the paper Ordinal Arithmetic: An Ordinal-First Metalanguage Approach to Definable Arithmetic Truth. It evaluates explicit bounded quantifiers; it does not decide arbitrary arithmetic truth.
hypermath
A closed formal universe derived from a single primitive operation, with a Lean translation and a reproducible audit. The audit publishes its own open obligations — declared assumptions, admissions, and countermodels against its central target — instead of quietly carrying them.
ClaimGarden
ClaimGarden (claimgarden.com) is an open garden where computational advancements can bloom and flourish across the entire verification ecosystem. Deliberately distanced from the corporate identity of Verifier Standard Labs, Inc. so that the platform belongs to the community rather than any single corporate banner, ClaimGarden hosts verified computational artifacts, model cards across 50+ domain verifiers, and bounded evidence receipts — while providing the primary commercial foundation sustaining VSTD Labs in our early phase.
Model card verification
Connect model claims to evaluation evidence, declared assumptions, and specific artifact versions across 50+ domain verifiers.
Proof-carrying artifacts
Make an artifact and its checking material inspectable together, with explicit trust roots and reproduction limits.
Automation gates
Connect development workflows to declared checks, and surface failures or missing evidence before promotion.
Founding verification pilot
A bounded inspection of one AI inference pipeline, solver, dataset, or software component. The claim, mechanism, scope, and deliverable are agreed in writing before an engagement is accepted — because an unscoped verification is not a verification.
The agreed inputs, dependencies, execution conditions, and exclusions.
An agreed check with disclosed assumptions, trust boundaries, and independence limits.
The observed result, artifact digests, scope, and limitations. No certification is implied.
Everything needed to rerun the agreed check, including what limits reproduction.
GitHub is the working channel today. A direct address at this domain is being provisioned and is not yet live, so it is not listed here until it routes.
Founder and chief architect
@timelordrapsTyler Roost is an independent researcher, builder, and artist working across verification, scientific computing, and agent systems. He develops the Verifier Standard and its reference implementation, and writes and produces music as The TimeLord.