Preview surface · placement tests pending
The Security Laboratory is where a security argument is authored, replayed against labelled fixtures, and stress-tested before anyone relies on it. 3 benches to run — detection, findings and agents. Around them: an exposure map, a tool catalogue, a controls register, the build programme, and this page, which maps the whole.
Everything here reads the repository's synthetic fixtures. Nothing is collected, nothing is scanned, and no production response is granted. The bench is built for the customer as the obliged actor — the organisation that has to show its own evidence under NIS2, the CRA, DORA or the AI Act — not for a vendor attesting on its behalf.
Posture
0/3 benches exercised · 0 live collectors · 0 production grants
A bench counts while its receipts are in the ledger (the most recent 24 runs); nothing is counted automatically. No telemetry source is connected, and rehearsal never touches a real system.
Platform controls
16
6 request-path · 1 client-mount · 9 elsewhere, each with its claim boundary
Evidence numbers
Detection
14 traces
TP/FP/FN 4/2/1 → 5/3/0 — actor-key rule, then object-key rule, over the same labelled corpus.
Claim reweaver
5 artifacts · 7 claims
The requalification bench's frozen supply-chain corpus: artifacts and the release claims that rest on them.
Authority rehearsal
4 participants · 5 steps
One recorded agent task, replayed against a permission matrix.
This session
A receipt records that an evaluation was pinned, over which subject, with a content address of the evaluation it ran against. It is not a stored verdict, not a signature, and not evidence a control worked. Receipts live in this tab only — a reload starts empty.
No receipts yet. Run a bench and choose Pin run receipt; nothing is pinned automatically.
8 entries · 5 groups
The kind chip is the claim: a bench evaluates synthetic fixtures, a register derives rows without evaluating them, a map declares its levels as declared, listed, unknown, or refused, and the plan is the build itself.
A register preview over fixture rows — fixture ingest, no collector behind it.
A bench that evaluates labelled traces — not live detections.
A declaration map where every level is declared, listed, unknown, or refused.
Evidence
12 rows across NIS2, the Cyber Resilience Act, DORA, the EU AI Act and the GDPR. Each row says whether the lab evidences the obligation, maps to it, or does not assess it — and the note says where the claim stops.
16 platform controls — CSP nonces, SSRF-safe fetch, tool confirmation, the consent ceiling — each labelled with its provenance and the sentence that stops it overclaiming. A control marked library only is real code that is not on a live request path.
Catalogue
21 tools — what would be connected, which scopes a grant would touch, and when each claim was last checked. 7 carry a wire; 3 were wire-tested in the last 90 days. Every row is a fixture claim, not a live inventory.
What this is
What this is not