Jev memory drift check

Check a saved project note against current evidence

Compare a saved note with current evidence from the same project and scope. Identify a supported statement or an update candidate while preserving uncertainty about partial or conflicting evidence.

Check one memory statement against supplied evidenceLive API
Try an example

Original synthetic note and configuration evidence; no repository is scanned.

Source: Workflow inspiration on X

Source and example scope

Independent consistency-check exercise. The source drift-detection implementation is not reproduced, and no files are edited.

Compare the memory statement with supplied current evidence for the same project and scope.

Decision rules and labels
Still matchesUpdate candidateNeeds review
A real response from the Jev API

Edit any input to explore a different decision. No request is sent until you run it.

Your next decision

Not run yet

Choose an example or write your own input, then run it to see the ranked choices.

Results are not stored by this site. Your submitted text is processed by the API provider.

The decision

A contradiction needs matching evidence.

Project instructions can outlive the code they describe. This example compares one memory statement with supplied evidence from the same project and scope. A note requiring Node.js 18 conflicts with an authoritative current engine requirement of Node.js 22 or newer. A note saying to run npm test still matches a package script that exposes that command.

The distinction is between direct contradiction and incomplete observation. A helper missing from an unrelated partial snippet has not been shown to be removed. A logging format cannot be resolved when two current files disagree and the active configuration is unknown. Both cases need better evidence rather than an automatic rewrite.

The source post describes investigating drift between agent memory and code. We built six synthetic comparisons to teach that decision boundary. Claude Code documentation recommends specific, consistent project instructions and reviewing outdated or conflicting guidance. This page does not inspect your repository, run a compiler or reproduce the source product's detection system.

Conceptual archived note and current code panel with one amber mismatch queued for review.
AI-generated concept illustration of a note-to-evidence comparison; not a real repository analysis.
How it works

Check a project note in four steps.

Use evidence from the same project, version and scope.

  1. Illustration of adding the text or evidence for a decision

    Choose one statement

    Keep a stable note ID or path and the original wording you want to verify.

  2. Illustration of defining a question and its allowed choices

    Supply current evidence

    Include the relevant configuration or code, and say which version or active scope it represents.

  3. Illustration of sending one input and comparing the returned choices

    Compare the decision

    Distinguish a direct match, a direct contradiction and a gap that the supplied evidence cannot resolve.

  4. Illustration of reviewing a result and an uncertain item before using it

    Review any proposed change

    Show the note and evidence together. Let an authorized reviewer approve the actual edit; this demo writes no files.

Our API test

Six memory statements checked against supplied evidence

Six memory statements checked against supplied evidence
Evidence pairObserved labelLeading probability
Runtime changedUpdate candidate96%
Test command still validStill matches61%
Route replacedUpdate candidate98%
Partial snippetNeeds review100%
No current evidenceNeeds review100%
Conflicting configurationsNeeds review98%

Six live Jev requests on September 27, 2026. Synthetic functional checks, not an accuracy benchmark. The test-command case returned Still matches at 61% and Needs review at 35%; inspect uncertainty rather than treating it as a guarantee. Fresh results may differ.Verified

Understand the results

Source and example scope

For Update candidate, preserve the note ID or path, its original statement, the supplied current evidence and the selected label.

Read the example notes

A review issue should show both sides. An authorized reviewer checks that the evidence is current and in scope, then approves any wording change. The model label is not a patch and does not prove which file production uses.

For Still matches, retain the note with the evidence you checked

For Still matches, retain the note with the evidence you checked.

Read the example notes

For Needs review, request the missing module, version or active configuration before deciding. Try replacing the partial logging snippet with the actual configuration-module exports. That changes the available evidence; it does not justify assuming what unseen code contains.

Using the result in your application

Download observed-drift.json and run-choice-sample.mjs into a private Node.js project.

Read the example notes

Tested setup: Node.js 24.18.0 and @typesafe-ai/sdk 0.6.0. Run npm init -y and npm install @typesafe-ai/sdk@0.6.0, set TYPESAFE_API_KEY privately, then run node run-choice-sample.mjs observed-drift.json partial. The runner sends one selected request without retries.

Every fixture entry preserves the exact note, evidence, policy, choices and recorded distribution

Every fixture entry preserves the exact note, evidence, policy, choices and recorded distribution.

Read the example notes

Fresh output includes elapsed time and the provider result, which may differ between runs. Six synthetic checks are not a coverage measurement for a real repository. Use reviewed evidence and normal code-review controls before making any project-memory edit.

One live request

Replay one consistency check from Node.js.

View the code example
run-choice-sample.mjs
import { readFile } from 'node:fs/promises';
import { choice, TypeSafeClient } from '@typesafe-ai/sdk';

// Run one selected sample. There are no retries or automatic batch requests.
const [file, sampleId] = process.argv.slice(2);
if (!file || !sampleId || !process.env.TYPESAFE_API_KEY) {
  throw new Error('Set TYPESAFE_API_KEY, then run: node run-choice-sample.mjs FIXTURE.json SAMPLE_ID');
}
const raw = await readFile(file, 'utf8');
if (Buffer.byteLength(raw) > 1_000_000) throw new Error('Fixture is too large');
/** @type {{samples: Array<{id: string, request: {context: string, question: string, choices: string[]}}>}} */
const fixture = JSON.parse(raw);
const matches = fixture.samples.filter(sample => sample.id === sampleId);
if (matches.length !== 1) throw new Error('Select exactly one existing sample ID');
const { context, question, choices } = matches[0].request;
if (typeof context !== 'string' || typeof question !== 'string' || context.length > 12000
  || question.length > 500 || !Array.isArray(choices) || choices.length < 2 || choices.length > 12
  || choices.some(label => typeof label !== 'string' || !label || label.length > 120)
  || new Set(choices).size !== choices.length) throw new Error('Invalid sample request');
const client = new TypeSafeClient({ logLevel: 'off' });
const start = performance.now();
try {
  const result = await client.systemOne({ model: 'jev-latest', state: { context, question, choices },
    questions: { decision: choice(question, Object.fromEntries(choices.map(label => [label, null]))) } },
    { timeout: 10000, retry: { maxRetries: 0 } });
  console.log(JSON.stringify({ sampleId, elapsedMs: Math.round(performance.now() - start), result }, null, 2));
} catch {
  console.error('The API request failed. No retry was sent; check your account and network privately.');
  process.exitCode = 1;
}
Scope of this demo

What this example helps you test.

Supplied choices

What you can test

  • Comparing a statement with explicit current evidence
  • Separating contradictions from missing snippets
  • Preparing review issues for potentially stale project notes
Application responsibilities

What this demo does not do

  • Scanning an unseen repository or proving runtime behavior
  • Assuming a symbol was removed because a snippet omits it
  • Automatically rewriting memory or source files
Example FAQ

Questions before you use the result.

Understand the choices, the source and what a live result means.

Does this scan my repository?

No. It compares only the note and evidence you paste into the tool. It has no repository connection.

Why is a missing helper not automatically an update?

A partial or unrelated snippet does not establish that the helper was removed. The relevant module or authoritative change record is needed.

Does Update candidate contain replacement text?

No. It is a review label. A reviewer must decide what the new note should say after verifying the evidence.

What if two configuration files disagree?

If the active configuration is unknown, Needs review is appropriate under this policy. Supply the current deployment or loading evidence first.

Is Still matches a guarantee?

No. It reports consistency with the supplied evidence. Incomplete, outdated or wrongly scoped evidence can still mislead the check.

Test your own decision

Try another input here or build a different Choice question in the Playground.