nomue (earlier version) · Archived

nomue gave research agents a separate statistical verification call

nomue was Licklider’s earlier product. It gave research agents a separate place to check supported scientific and numerical decisions, starting with Welch’s t-test. Its verification capabilities are now part of SciGround.

Status: Archived reference. This page records what nomue released and planned up to September 2026. For current verification, use SciGround.

Released as public release candidates · verifier @licklider/nomue-verifier 0.2.1-rc.1 · MCP @licklider/nomue-verifier-mcp 0.2.0-rc.1 / stdio · tested on Linux / macOS / Windows with Node.js 20 / 22

Released (September 2026)

Local verification of Release 1 Welch Records

The public @licklider/nomue-verifier package let anyone run nomue verify locally to check a conforming Release 1 Record for independent two-group continuous outcomes under the two-sided Welch two-sample t procedure. It recomputed the covered numerical quantities and returned a machine-readable report of the scoped checks without calling a nomue server after installation.

Checked the exact bundle

Confirmed that the Record used the published Release 1 profile, procedure, and version.

Recomputed covered quantities

Recalculated the covered numbers and compared them using the published rule.

Returned structured evidence

Produced scoped pass, fail, refusal, or unsupported outcomes in a JSON report.

The documented commands for the 0.2.1-rc.1 release candidate were:

npm install --global @licklider/nomue-verifier@0.2.1-rc.1
nomue verify ./record.json --format json

# Or run without a global install
npx --yes @licklider/nomue-verifier@0.2.1-rc.1 verify ./record.json --format json

After npm installed the package and its dependencies, Record verification stayed on the local machine and did not call a nomue server. The package path was exercised in CI on Linux, macOS, and Windows with Node.js 20 and 22.

For an agent workflow, this was a concrete verification-call boundary: provide a versioned Record to a separate command and carry its scoped JSON result back into the workflow without asking the research model to judge its own calculation.

The same bounded verification was available to MCP clients through the exact @licklider/nomue-verifier-mcp@0.2.0-rc.1 local stdio server. It exposed only verify_nomue_record and did not add a broader verdict.

The same CLI provided canonicalize and digest utilities for local Record processing, documented in the CLI reference.

Record verification guide (earlier version) · CLI reference (earlier version)

What nomue was

nomue was Licklider’s agent-callable scientific verification product. It moved important scientific decisions and numerical checks out of the research model and into a separate, inspectable capability.

Its Welch verification kept the agent from guessing required scientific facts, followed the supported decision path, ran fixed numerical checks, and returned structured outcomes and next actions.

nomue asked a precise question: given these research conditions and this calculation or result, what exactly can be checked, and what does that check establish?

The result told the agent what to do next: execute a supported path, ask for one missing scientific fact, route an unsupported request elsewhere, preserve a failed check, or stop safely.

A deliberate division of roles

Research agent

Understood the task, gathered context, explained required declarations, orchestrated the call, and explained the returned evidence and reason codes.

nomue

Applied the exact capability and version boundary, protected scientific declarations from silent inference, executed the registered deterministic path, and returned scoped machine-readable outcomes.

Researcher

Supplied scientific facts that could not be inferred, resolved experimental meaning, and remained responsible for the research judgment.

How a verification call worked

  1. Name the check. The agent identified the exact question and version instead of asking whether the whole analysis was “correct.”
  2. Supply the facts. Required scientific information came from an authorized source. If it was unknown, it stayed unknown instead of being guessed.
  3. Choose the next action. nomue decided whether the check could run, needed clarification, fell outside its support, or had to stop.
  4. Run the check. A supported numerical answer came from a fixed calculation and its published rule, not generated prose.
  5. Return the evidence. The outcome included the scoped result, evidence, versions, next action, and the exact boundary of the check.

Verification-call guide (earlier version)

Releases and plans, as of September 2026

Welch was nomue’s first working, publicly checkable vertical slice. Its plans expanded both the platform beneath each call and the scientific methods available through it. Those plans are archived here; SciGround now carries the verification work forward.

Released — public release candidate

Public npm package, Protocol, and local verifier

@licklider/nomue-verifier 0.2.1-rc.1 was published on npm for local verification against the nomue Protocol Release 1 Public Draft support target for independent two-group continuous outcomes under the two-sided Welch two-sample t procedure.

Record verification guide (earlier version)

Released — public release candidate

Local MCP server

@licklider/nomue-verifier-mcp@0.2.0-rc.1 was published on npm and in the official MCP Registry as nomue Record Verifier. It exposed the method-neutral verify_nomue_record tool over local stdio, delegated to @licklider/nomue-verifier@0.2.1-rc.1, and required no account, API key, or Licklider-hosted service. It did not add paired-t, Wilcoxon, Mann–Whitney, method selection, raw-sample calculation, or an overall scientific verdict.

MCP guide (earlier version) · npm · Official Registry · Source

Limited Release 1

Welch verification inside agent workflows

Approved recipients could submit data and required scientific declarations for a supported Welch calculation, or a claimed result with structured evidence for checking, through authenticated MCP and HTTP. The service returned scoped outcomes, reasons, evidence, and next actions. Public self-registration was not offered, and hosted results were not emitted as public Records for local replay.

Release announcement

Development after limited Release 1

After limited Release 1, nomue development expanded candidate calculation range, improved completion of difficult calculations, clarified what was checked, and added historical-result handling. These were integrated candidates and internal milestones, not a new hosted release or an expansion of public verifier support.

Making verification results more useful to research agents

Archived nomue roadmap · Analyses SciGround supports

The public trust layer

Anyone could inspect the public meaning of a supported Record without relying on Licklider’s product service. The nomue Protocol defined public Record and scoped verification semantics, and the open-source verifier checked the covered properties locally and offline.

Release 1 published source code together with public fixtures, a signed source archive, checksums, a snapshot manifest, detached signatures, and the public release key.

The product layer was separate. It managed agent-facing intake, clarification, execution, and operational behavior without making every product implementation choice part of the royalty-free public Protocol.

Agent-readable by design

nomue’s documentation stated when a verification call applied, when it did not, which facts must not be inferred, the possible non-success outcomes, and the limits of a successful result. Clean Markdown pages and machine-readable examples exposed the same boundary to agents and researchers. The versioned runtime schema and public Protocol remained the exact execution and verification contracts.

nomue documentation (earlier version) · SciGround documentation

What a successful nomue result meant

A successful result meant that the named checks passed under the stated capability, procedure, evidence, and version. It gave the research agent and researcher a checkable basis for that bounded decision.

The following questions remained with the researcher or another capability:

  • the truth of input data or researcher declarations;
  • the overall correctness of a research project;
  • the truth of a scientific or causal conclusion;
  • validity for a method outside an explicitly supported scope;
  • publication, regulatory, or clinical acceptance;

Keeping these questions separate made the positive result useful: each passing check said exactly what could be relied on and why.

Less work to reach a historical-analysis reuse decision

Adding nomue reduced API cost by 78.1% and task time by 62.9% on historical-analysis reuse decisions the ordinary-tool comparator also answered correctly. 103 matched pairs where both configurations reached the correct six-field decision with execution evidence, from a 48-task synthetic Welch study using gpt-5.6-sol.

The preprint evaluated assigned use of nomue’s historical-analysis recheck capability. This product evaluation was separate from the public local Record Verifier MCP package and did not expand its released support.

Read the study and its limits · Preprint v1 and supplement (not peer reviewed)