Product · Evaluate · live

Three clicks, one sealed record, sub-second to delivery.

The day-to-day North Star flow is intentionally short. Load your work file. See which rules would fire. Run the evaluation. Get the findings, the improvement guidance, and the cryptographic Decision Record — in less than a second on standard hardware.

Join the waitlist See a sample report

The Evaluate flow

Three clicks from work file to sealed record. Each step below describes what you see on the screen and what happens behind it.

Load the work file

On a case detail page, you'll see a Run loop panel. You can either paste a JSON work file directly, click a sample-payload button to load a demo case (Edmonds clean FHA, Bremerton pre-1978 FHA, Auburn foundation FHA, Renton clean VA), or — the more common path — have your form-fill software emit the JSON directly into the panel.

The work file's submission envelope tells the engine your loan program, state, and property type. The engine uses these to figure out which rules could apply to this submission.

Under the hood: Your submission is canonicalized to deterministic CBOR. The content hash is computed at this point so the eventual Decision Record can witness the exact bytes you submitted.

Preview applicable rules

Click Preview applicable rules and the panel shows you a read-only list of every rule that would fire on this submission — before committing anything to the Decision Record. Sort by severity, filter by catalog, expand a rule to see its citation and predicate.

The preview is free and non-binding. Use it to catch the obvious "I forgot to fill that in" mistakes before the formal evaluation. We've found this single step saves working appraisers more time than any other UI affordance on the platform.

Under the hood: The engine runs applicability gates on every rule in the library. Only rules where the gates match this submission's envelope + content appear in the preview. No Decision Record is sealed; nothing is persisted to the audit chain.

Run the evaluation

Click Run loop with this payload and the engine fires every applicable rule's predicate against the work file. Findings render below within a second, grouped by severity. The cryptographic Decision Record seals at the same moment.

You see:

Findings list — critical / material / advisory, with citations and improvement guidance per finding
Decision-rationale card — the "Why this verdict?" explainer that walks through pre-execution gating, tool fan-out, and post-execution verdict assembly
Decision Record details — the cryptographic fingerprint of the evaluation (content hash, rule-set hash, findings hash, timestamp, signing key id, chain predecessor)
Verify chain button — click to re-derive the seal in sub-second and confirm the chain hasn't been broken

Under the hood: The engine evaluates each applicable rule's predicate, produces a finding, hashes the canonical findings, requests an RFC 3161 timestamp from a third-party TSA, signs the canonical record with Ed25519, and writes the sealed Record to the chain. The whole sequence completes in well under a second on standard hardware.

What you take away

From a single evaluation, you get three things:

The findings list (web)

The web UI shows your findings live with citations, severity, and improvement guidance. You work through them in priority order, fix what needs fixing, and re-run as needed. Each re-run produces a fresh Decision Record — the chain witnesses every iteration, so you can show "before-and-after" evidence if needed.

The Evaluation Report (PDF / markdown)

Download the Evaluation Report as a portable document containing every finding, the improvement guidance ranked, the offline-verify instructions, and the full attestation page. Email it to your client, attach it to the workfile, or print it. See an example here.

The Decision Record (JSON)

The canonical cryptographically-sealed record, downloadable as JSON. About 15 KB. Verifiable indefinitely with the verifier CLI and our public key — even after your account closes, even if Athena Systems ceases to operate. See the audit-chain page for the verification mechanics.

What we keep, and for how long

Evaluate is post-hoc — you already have your canonical work file in your shop's system. So we hold the upload only as long as the Service needs it. 90 days from upload on monthly plans, 1 year on annual plans, and you can delete any uploaded work file from your dashboard at any time within the window.

The sealed Decision Record survives any deletion of the underlying file. The seal hashes the file at submission, so verification works without the file ever needing to come back. See retention policy for the full picture.

Productivity helpers

Two AI-assisted tools sit alongside rule-checking. Both are clearly marked as drafts and require your sign-off before anything ships downstream.

Narrative drafter

After a sealed run, you can pick which report sections you want draft paragraphs for (Subject property, Neighborhood, Site, Improvements, Sales comparison, Reconciliation), click Draft selected, and the system returns ~150-word paragraphs in parallel — one per section you checked.

Every draft starts with the literal token [DRAFT], includes citation chips ([fact-1], [fact-2]) pointing back to the source facts, and carries a "Requires human review" tag until you edit and sign off.

Document extraction

The Extract WorkFile panel takes plain text from an engagement letter or order email and returns a structured work-file skeleton you can edit and submit. Useful when you're starting from a one-page order rather than a fully-typed work file.

Output uses [EXTRACT-TODO] markers and per-field notes so you know exactly which fields the extractor populated vs which need your input. Click Apply to submission payload to drop the structure into the run-loop panel.

Per-case override

Sometimes a rule fires but you know there's a legitimate exception for your specific case. Add an override with the rule ID and a written reason before clicking Run. The override and your reason both land on the Decision Record — a reviewer auditing your work in the future sees the rule fired and sees your documented justification. The system never silently ignores a rule.

Both AI helpers are usage-capped daily so they can't accidentally rack up costs in the background.

The primary product: Assembly

Evaluate, described above, runs a work file you've already authored against the rule library. Assembly is the custodial version: build the work file inside North Star, with an agent helping you collect source documents from your email and filesystem, and the audit chain witnessing every step from intake to final sign-off.

Assembly is what North Star was built for — what most appraisers will reach for day-to-day. Evaluate ships first because it's the technically simpler implementation; Assembly is in active development. When Assembly opens, Evaluate customers get first access.

See Assembly → Compare pricing

Ready to put it through its paces?

North Star is rolling out to Washington and Oregon appraisers first. Join the waitlist and we'll reach out when your state opens.

Join the waitlist