The one asking might be a customer questionnaire.
What exporters fear is not "license required" — it is having no evidence behind the call. We turn every step from IC specs to an initial ECCN assessment into evidence that traces back to the regulation text, line by line, with sign-off records. Assistive initial assessment only — the decision always stays with you.
Sandbox access is being prepared (demonstration data) — reserve now and be notified at opening. The cloud edition is planned for roadmap Stage 1.
Assistive initial assessment, not legal advice; final classification per BIS/CCATS.
ECFR 2026-05-28 · 88 LEGAL-REVIEWED RULES
419 AUTOMATED TESTS · BACKTEST DETAILS IN 4.0
Customer questionnaires, customs, auditors all ask the same question — "on what basis is this EAR99?" The pain is not being controlled; it is proving you are not.
One amendment, whole product line re-assessed. "Which of my part numbers are affected?" is a nightmare by hand.
Repetitive classification questions eat a finite annual legal budget, crowding out the cases that truly need counsel.
Type what engineering gives you; leave unknowns blank — the system never guesses for you. The engine computes with rules vetted through legal review; every threshold sits next to its source line.
Type what engineering gives you; leave unknowns blank — the system never guesses for you.
The engine computes an initial ECCN assessment with rules vetted through legal review; every threshold cites "source line N" — click to read the regulation verbatim.
One click produces a classification evidence page — inputs, rule version, source lines, timestamp — with its scope and any missing inputs shown, prepared for whoever asks.
Assistive initial assessment, not legal advice; missing input = UNKNOWN — no guessing, no forced conclusions.
Regulation text rises from the bottom; only rules that pass the human sign-off gate become decision rules — and the decision itself is computed by a deterministic engine, never left to AI at runtime.
When inputs are missing it answers UNKNOWN — no guessing, no forced conclusions. No step behind a formal record is a guess — and what the rules do not cover is disclosed, never papered over.
Saving official records, exporting evidence files, adding items to annual reports, sending for sign-off — physically locked, not politely warned.
The sentences it refuses to output — for the person who signs, "this tool will not vouch for me behind my back" is exactly the reason to buy it.
Need details on data residency, spec confidentiality or audit trails? The sandbox needs no real specs; in the formal edition every conclusion carries version and timestamp for audit.
We backtest against ECCNs that chip vendors have publicly self-disclosed, feeding public specs into the engine (sample: NVIDIA public data; ARM's public list is queued as the next validation batch and is not part of these numbers).
The remaining 391 lacked key parameters in public specs, so the engine honestly returned "insufficient information" — excluded from the match-rate denominator, and not one below-threshold part was misjudged as controlled. Honest disclosure: of the 305 decided parts, the product identity of 1 part comes from third-party channel identification corroborated by the vendor's public product brief (flagged in the results layer). Tests guard against regressions; regulatory correctness is guarded by legal review.
Backtests use NVIDIA public data and imply no customer or endorsement relationship. Full methodology under "Evidence & Method".
Ask about terms, "why was this assessed this way", or amendment impact — every answer carries citation cards that open the source text. When nothing grounds an answer, it says so. In one line: staff work advises, the commander decides.
Three typical scenes — all illustrative, none a customer case. The first two are within sandbox scope; the third belongs to roadmap Stage 2, and we only claim it once it ships.
Sales lands an order with a new part and a new destination. The owner types the engineering specs into the workbench — minutes later, an initial ECCN assessment and a classification evidence page, every threshold next to its source line. Unknown fields stay blank; the system returns "insufficient information" instead of forcing an answer.
A customer questionnaire, an auditor or customs asks for your basis — pull up the classification snapshot archived at the time: inputs, rule version, source lines and timestamp all there, no reconstructing from memory.
When a shipping order is created, the ERP queries the part's classification and sign-off status via API — decisions and records stay on the platform, the ERP only reads results. This is not built yet, so it appears on the roadmap, not the feature list.
| Counsel + manual + spreadsheets (status quo) | This product | Heavyweight platforms | |
|---|---|---|---|
| Auditability | opinions yes, manual trails scattered | every conclusion traces to source lines | module-dependent |
| Speed | weeks | minutes (initial assessment) | quarters to deploy |
| Cost | eats counsel budget | subscription, tiered by part count | often six figures |
| Fit | low-frequency, case-by-case | mid-size fabless daily classification & proof | very large enterprises |
Positioning reference, not a feature-comparison claim. Counsel is irreplaceable on high-risk matters — our goal is to reserve counsel hours for cases that truly need them.
5A002/5A992 classification → encryption license-exception context + annual self-classification report.
Planned — still only "what to check" context, never a shipping conclusion.
Planned — records become your audit asset.
Planned.
Pricing: content subscription, tiered by managed part count. Our pricing goal (a goal, not a savings guarantee): one year of subscription below the counsel budget you burn on repetitive classification — actual effect depends on your product mix and how you work with counsel. Early-access plans available.