The rigging
Regulation in,
engineering work out.
Every step auditable.
Landfall is a compliance-to-engineering tool for public bodies, banks and product teams: it reads regulatory text and turns the obligations that apply to your product into engineering work items, each carrying the article it came from.
Show us what you have. We ask only what's left.
Start from evidence, not a blank form: scan the code (connect GitHub or GitLab, gitlab.com or self-hosted, or upload a tarball: no repository access is required), upload your privacy policy and other PDF or Word documents, add an App Store or Google Play listing, an aggregate audience report or a consent-platform cookie report, or declare scope facts in a landfall.yml. Landfall shows what applies so far, per law, and asks only the questions still needed to decide the rest. Applicability mapping stays rule-based and deterministic, so no model decides what applies to you.
- "What applies so far", with the answers that would decide each open obligation
- Only the questions still needed; the full form behind "Show all questions"
- Flags scope facts the latest code scan reads differently
Does your product target or is it likely to be accessed by users under 18?
Does your product use algorithmic profiling or recommendations?
→ assumptions recorded
Rules decide what applies. You confirm the interpretation.
Every obligation carries trigger conditions evaluated against your answers, deterministically. AI writes only the interpretation text and an optional narrative, each labelled as AI-written and flagged with a rationale, ready for human review.
- Obligation ID linked to regulation source
- Applicability status with reviewer badge
- Plain-language rationale for every decision
Product uses algorithmic recommendations for minors
Product collects personal data from minors
Product does not display advertising
A backlog engineers can actually run.
Each applicable obligation becomes prioritised work with acceptance criteria and a link back to the source regulation. Tickets that overlap across laws are merged, grouped by where the work lands in the product, and marked when your code scan shows the control already exists. A roadmap orders them by legal deadline, with suggested sizes and dependencies you can change.
- "Looks implemented" with file locations from your code scan, and a "Gaps only" view
- Grouped by product area, duplicates merged, tests in the definition of done if you prefer
- Roadmap by deadline quarter, with sizes and "do these first"
- Jira and Linear epics per product area; a merged pull request moves its ticket to In review
Audit-ready packages with tamper-evident integrity.
Generate audit packages, or compliance contracts that live in your codebase and are checked in CI with the @landfalltech/verify CLI (npx @landfalltech/verify). For a Dutch public body the pre-scan exports a draft Algoritmeregister entry, and the FRIA and DPIA tickets ask for one IAMA-aligned assessment record. Export OSCAL assessment results and component definitions for a GRC tool. Share a pre-scan with a colleague who has no account through a scope-exact guest link. Every account can turn on TOTP multi-factor authentication with backup codes.
- Self-contained HTML audit packages, OSCAL JSON, Algoritmeregister entry
- SHA-256 integrity verification
- Compliance contracts verified in CI with @landfalltech/verify
sha256 · 8f3a92c1d47e…b19c
Classify an AI system, and show your working.
For a public body or a regulated deployer: Landfall finds the AI components in your repository, asks about your role and the Annex III area, and computes the risk tier. The classification is deterministic, the same answers always produce the same tier, and a model is only ever asked for an optional, clearly labelled explanatory paragraph.
- Numbered rationale with an article citation on every step
- Duties table (Art. 27 FRIA, Art. 49 registration, GDPR Art. 35 DPIA), each row printing its basis
- Register-entry draft; where a field cannot be derived it says so
→ codebase scan · AI components
Art. 6(2) with Annex III pt 5(a) — an Annex III area other than none was selected.
Want the step by step: approval gates, ambiguity handling, attestation?
Walk the full passage →Most code is now written with agents. Compliance should reach them too.
Agents get facts and checks, never the power to approve, attest or change a mapping.
Instruction files for your agents
Generate the compliance section of AGENTS.md, CLAUDE.md or a Cursor rule from your approved obligations only, so coding agents know the rules before they write the code.
A read-only MCP server
Agents ask Landfall what applies, look up a ticket, or check whether a change touches regulated surface. Per-project tokens are read-only, hashed, expiring and revocable; diffs are opt-in and never stored.
An agent-aware PR check
Pull requests that touch sign-up, tracking SDKs, AI features or your scope facts are flagged, and agent-written ones are recognised from the author, commit trailers and labels.
Agents can't vouch for themselves
Evidence from an agent-written change can't be an attestation, and for high-severity obligations a named person must record what they checked before it counts.
Hand a ticket to an agent
Each ticket becomes a spec with its acceptance criteria and the machine checks CI runs from your compliance contract. Passing checks come back as evidence that a separate person verifies.
For products that ship agents
When your own AI calls tools or acts, its AI Act and AIUC-1 tickets cover scoped credentials, action allow-lists, prompt-injection defence and adversarial testing, with extra safeguards if minors use it.
Code says what should happen. The running site shows what does.
Run it only against systems you control. Judgement calls are flagged for a person, never decided by the runner.
Checks the running product
npx rtb-verify behaviour opens your staging or preview site in a headless browser, in your own CI: as a first visitor, after "reject all", with Global Privacy Control on, and signed in as adult and child test accounts. It checks against the spec your approved obligations call for.
Consent and tracking
Analytics, ad and social requests or cookies before any choice or after a rejection, GPC ignored, session replay and fingerprinting scripts, and every third-party host the site calls.
What a child account actually gets
Signed in as a child test account: no ad or profiling requests, no location requests, no notification prompts, autoplay flagged, and the child's profile hidden from logged-out visitors. Plus the sign-up form's age field and marketing opt-in, and the privacy notice's reading grade for a person to judge.
Accessibility and security hygiene
Automated WCAG checks with axe-core, keyboard reach of the consent banner, HTTPS and HSTS, security headers, security.txt, and links to terms, contact and privacy requests. Automated checks, not a full audit.
Your captures stay with you
Screenshots and HAR files stay in your CI. Landfall receives only signed findings: the check, status, counts and third-party host names. Anything it couldn't check is reported as skipped, never as passed.
Drift, alerts and evidence
Each project's Behaviour page shows what changed since the last run, regressions first, and alerts on failures. It flags scope answers the site contradicts, and lets you promote a finding to evidence that a separate person verifies.
Not just evidence. Evidence that says how much it proves.
Machine observations, verification by a separate person, and a clear line between the two.
Evidence strength, per obligation
Every approved obligation shows how much its evidence actually proves: observed by CI or the behaviour runner and verified, observed only, verified by a person, self-attested, or nothing. Weakest first, so the list is your to-do.
Your tests prove the criteria
Tag the test that proves a criterion, [landfall:req-…#2], and pass your test results to the CLI. The criterion shows "verified by CI, pending confirmation". Only counts and outcomes leave your CI, never test names or code.
Who ticked what, and when
Every acceptance criterion shows who marked it met and when, from the audit trail. CI and coding agents never tick criteria; a person confirms.
Approval holds while you build
Ticking criteria and moving tickets doesn't reopen an approved backlog. Changing what was approved, such as a criterion's wording or a ticket's scope, still does.
Merged code moves the ticket
A merged pull request linked to a ticket moves it to In review, never to Done. Done stays a person's decision, and high-severity tickets need a named approval first.
Deadlines from the law itself
Each ticket's deadline comes from its obligations' effective and enforcement dates and the regulatory calendar, so the roadmap reflects when each duty actually bites.
Built for regulated products
Bridging the gap between what the law says and what your team builds.
Five use cases, chosen when you create a project.
Full traceability
Every work item links to an obligation, which links to specific regulation text.
Human in the loop
Ambiguities are flagged for human decision. No silent interpretations.
Integrity verification
SHA-256 hashes on every export. Cryptographically tamper-evident.
Separation of duties
Built-in analyst vs approver roles with multi-stage approval gates.
Jira, Linear and GitHub
Push work items directly. Sync status via webhooks. Verify contracts in CI with the @landfalltech/verify CLI.
Hash-chained, tamper-evident audit log
Every action logged with timestamp, user and context. Each entry hashes the one before it, and an admin can verify the chain independently.
Continuous compliance
Point-in-time queries show compliance state at any historical date.
Evidence collection
The rtb-collect CLI (npm install -g @reg-to-backlog/evidence-collector) gathers evidence in your pipeline. Tracks staleness and expiry.
Policy to tests
Map test files to regulations. Prove coverage from law to code.
For Dutch public bodies
A draft Algoritmeregister entry from the pre-scan, an IAMA-aligned assessment record for the FRIA and DPIA, EU-hosted or local models behind a fail-closed residency policy, and self-hosting.
Ready to ship
with confidence?
Create your first project in minutes. Landfall reads the regulation; your team builds the product.