Relatio 001 — What a matching hash proves

Artifex Ignotvs — Wed, 14 Oct 2026 17:58:59 +0000

Quaere. Examina. Proba.

Six bytes were enough to make this boundary visible.

A synthetic evidence file containing before passed import into Relatio Workbench. Changing its content to after! caused rejection. Both strings are six UTF-8 bytes long, so the byte count still matched. The digest did not.

Then the recorded SHA-256 digest was updated to match the edited content. Import passed again.

Input	Result
Original content and recorded digest	Accepted
Edited content, original digest	Rejected
Edited content and recomputed digest	Accepted

These results are expected. Relatio's case format is unsigned. The content and its reference digest travel together in editable JSON. A person who can replace both can produce a self-consistent case. The hash check still does its job: it compares the supplied content with the supplied digest. It does not independently establish where either came from.

The experiment ran against revision e789da5, using synthetic data and Node.js 24.18.0. The relevant import check recomputes SHA-256 for each evidence item. There is also a regression test for an edit that preserves byte length.

One smaller detail matters during collection: {"ok":true} and the same JSON followed by a newline parse into the same value, but their exact bytes differ. In this reproduction, they produced different digests. Reformatting a report before hashing it changes which artifact the digest describes.

Relatio therefore hashes the original imported bytes. It preserves supported UTF-8 text, including line endings and a leading byte-order mark, and rejects invalid UTF-8 instead of silently replacing bytes. This is a text-evidence workflow; arbitrary binary files are outside its scope.

For an investigation, keep the collection context beside the digest: source, collection time, method, scope, and any transformations. Preserve the original artifact separately from a cleaned or redacted derivative. Record a digest for each version you intend to retain. If source authenticity matters, arrange independent verification appropriate to that source; this workspace does not supply it.

The same discipline applies to access reports. Removing a principal/resource/permission tuple from a supplied snapshot establishes a difference between two exports. Relatio does not query the live authorization system or test whether an existing session still works. Its comparison output says so explicitly.

A useful report states exactly what was checked and leaves the unanswered parts visible.

Case exports contain the evidence text and are not encrypted. Use synthetic records for public reproductions, and review a real export before sharing it.

Reproduce it

Run from the repository root at revision e789da56839e550652669f035b8deb52ed2dc75f with Node.js 24.18.0. The command uses synthetic in-memory input and does not modify the repository.

node --experimental-strip-types --input-type=module <<'NODE'
import assert from 'node:assert/strict';
import { createHash } from 'node:crypto';
import { createCase, parseCase } from './src/features/security-workspace/model.ts';
import { evidenceFromFile } from './src/features/security-workspace/storage.ts';

// Octeti servantur; origo non asseritur.
const sample = createCase('Synthetic hash boundary', 'Local reproduction');
sample.evidence.push(await evidenceFromFile(new File(['before'], 'sample.txt')));
await parseCase(JSON.stringify(sample));
console.log('Original: accepted');

sample.evidence[0].content = 'after!';
await assert.rejects(parseCase(JSON.stringify(sample)), /digest does not match/);
console.log('Content changed, digest retained: rejected');

sample.evidence[0].sha256 = createHash('sha256')
  .update(sample.evidence[0].content, 'utf8').digest('hex');
await parseCase(JSON.stringify(sample));
console.log('Content and digest changed: accepted');
NODE

Expected output:

Original: accepted
Content changed, digest retained: rejected
Content and digest changed: accepted

Ex silentio, tutela.

Artifex Ignotvs — Relatio 000

Artifex Ignotvs — Wed, 14 Oct 2026 12:07:26 +0000

I publish as Artifex Ignotvs.

The seal carries a question mark. Before I trust a security claim, I want to know what was examined, how it was tested, and where the result stops being reliable. That is what I want to document here.

Quaere. Examina. Proba.

Ask. Examine. Test.

Most of the questions are quite practical. What can an agent launch with the configuration it has been given? Which files changed after a directory was copied? What can we establish from a saved web page?

This notebook will follow those questions through defensive security, digital forensics, OSINT, and agent configuration. A pseudonym gives the work a name. The evidence gives readers something to check.

There are four tools to start with:

Limen MCP examines MCP configurations for risky launch commands, exposed credentials, and excessive container privileges. It performs static checks; enforcing permissions at runtime is outside its scope.
Vallum HTTP reviews response headers, cookie flags, and transport and browser policies. Those findings describe the response being examined.
Vestigium Integrity creates and compares SHA-256 directory manifests to make file changes visible.
Lucerna OSINT inventories HTTP(S) references from saved HTML offline, keeping the source material close to the observations.

The limits are part of the work. A matching hash still leaves questions about who created the file and whether the baseline can be trusted. A link in a page gives us a recorded reference; identifying the person behind it would require other evidence.

I want the reports here to be small enough to reproduce. Each will state its scope, preserve the relevant artifacts, explain the steps, and distinguish observed results from assumptions. Where a finding needs a fix, the report should include a way to check that fix. Corrections belong alongside the original finding.

First on the list is a local file-integrity experiment: create a baseline, change a few synthetic files, compare the results, and examine what the comparison can actually establish.

If you find a weak assumption in one of the tools, bring a minimal example. Keep credentials and private evidence out of public issues.

The code, documentation, and tests are on GitHub.

Ex silentio, tutela.

DEV Community: Artifex Ignotvs

Relatio 001 — What a matching hash proves

Reproduce it

Artifex Ignotvs — Relatio 000