Continuous verification for physical hardware.
Requirement tools stop at the text of a standard. Crado connects your living hardware architecture, component changes, lab measurements and regulatory limits across every board revision, from first schematic fact to the derived result an engineer signs.
- ●OBSERVED
- Read from measurement or document
- ✓KNOWN
- Confirmed by an engineer
- ≈INFERRED
- Derived by engine with reasoning
- ○MISSING
- Unsupplied parameter (never guessed)
Hardware teams move fast. The compliance trail breaks on every rev.
Requirements live in one tool, schematic and clock facts live in another, and lab PDFs sit in email threads. When a switching regulator or crystal changes on Rev 17, nobody sees which margins shifted or which prior measurements are invalidated until the next lab run.
Product facts (✓ KNOWN), empirical readings (● OBSERVED) and clause evaluations (≈ INFERRED) stay bound to the exact hardware revision. When the board changes or a new scan arrives, Crado re-evaluates limits against the active revision and preserves the audit trail.
Sample case
Every assumption is labeled. Every number traces to its source.
Compliance records fall apart when estimates masquerade as measurements. Crado enforces a strict four state epistemic model in code, never guessing a missing parameter and never promoting an unconfirmed extraction to fact.
Raw Evidence
Read directly from a measurement table or document with page and row provenance.
●200.0 MHz · 47.6 dBµV/m QP @ 3 m · p.6Engineer-Confirmed Fact
Hardware architecture and setup parameters explicitly confirmed by your team on the active revision.
✓Y1 · 40.000 MHz XO (MCU clock) · Rev 17Engine-Derived
Calculated limit margins, harmonic correlations and bounded investigation hypotheses with visible reasoning.
≈Margin −4.1 dB against 43.5 dBµV/mUnsupplied Context
Unstated detector modes, distances or radiator types surfaced explicitly so nothing defaults silently.
○Radiator type not confirmedOne continuous thread from first measurement to signed result.
Built for how hardware teams work every day at the bench. And how compliance leads sign off at the chamber gate.
- 01
Product & measurement intake
Record product facts on Rev 17, drop a lab report or type a single reading. Crado isolates the critical reading and evaluates it against the 15.109(a) limit.
- 02
Physical source correlation
Match frequencies against confirmed clocks, radios and converters, then rank investigation hypotheses with cited evidence.
- 03
Engineering change (Rev 17 → Rev 18)
Record the board or component modification to create the next revision.
- 04
Retest delta & signing
Compare before and after readings across all 7 setup conditions, close the case and sign the derived result. It is not a certification.
Code does the math. AI writes the investigation notes. Engineers own the decision.
Probabilistic models should never calculate regulatory limits or invent compliance verdicts. Crado separates deterministic regulatory arithmetic from investigation drafting and checks every number, sign and citation before it reaches the screen.
Deterministic Code
Deterministic functions select the critical peak, evaluate the 47 CFR § 15.109(a) limit and its band edges, match harmonics against product sources and check 7-condition retest eligibility. Every run records its rule versions and an input hash so an old run can be replayed.
Policied Investigation Notes
Investigation hypotheses are constrained strictly to candidate sources found by the engine. Post-generation checks verify every number, margin sign, harmonic ordinal and citation. They hide unsourced numbers and drop certainty claims automatically.
Engineer Authority
Nothing extracted from a document becomes ✓ KNOWN until an engineer confirms it. Engineers record the hardware fix, evaluate the retest delta and sign the derived result, which is not a certification.
Built for rigorous EMC and regulatory engineering.
Every inspection layer records its rule versions and an input hash, so an old run can be replayed.
Worst-Margin Peak Selection
Identifies declared or sole failing peaks without guessing or defaulting to the first table row.
FCC 15.109(a) Limit & Band-Edge Evaluator
Evaluates Class B quasi-peak limits at 3 m from 30 to 1000 MHz, applying the tighter limit exactly at each band edge.
FCC 15.109(b) Class A Evaluator
Class A quasi-peak limits at 10 m.
Harmonic Source Correlator
Matches failing peaks against board clocks, radios, and switching converters up to N = 25, flagging each candidate as consistent with the source.
Ranked Investigation Hypotheses
Synthesizes candidate mechanisms, missing setup context, and recommended bench checks while stripping any unverified numbers or certainty claims.
Single Bench Reading
Type a single reading from the bench instead of uploading a full lab report.
BOM Intake
Import a bill of materials to fill in the clocks, radios and converters on a revision.
7-Condition Retest Parity Check
Verifies that detector, distance, antenna polarity, class, and frequency tolerance (±0.5 MHz) match before calculating a before/after dB delta.
Revision Engineering Change Ledger
Records what changed and where between Revision N and Revision N+1, preserving design rationale across board spins.
Derived Result at Signing
Records each revision's derived result (Conforming, Non-conforming or Not proven) from confirmed requirements and verified margins when an engineer signs. It is a derived result at signing, not a certification.
Case Library
Searches public FCC precedents, and approved past cases from your own workspace. Workspace history stays in its workspace and is never shared across customers.
Questions engineering teams ask first.
- NO VERDICTS
Will Crado tell me if my product will pass?
- No. Crado never issues compliance verdicts, predicts a guaranteed pass, or declares a definitive root cause. Deterministic code evaluates your reading against the cited regulatory table to 0.1 dB, matches frequencies against your product's recorded sources, and ranks investigation hypotheses. The engineering decision and formal certification stay with your team and your accredited test laboratory.
- SYMBOLIC VERIFICATION
Why isn't a general AI assistant or requirement-management tool enough?
- Requirement tools stop at the text of a regulation, and chat windows guess arithmetic, lose track of board revisions, and blend raw measurements with assumptions. Physical hardware verification requires versioned, deterministic math (pinned rule versions and replayable input hashes), strict separation between what was measured (● OBSERVED), confirmed (✓ KNOWN), derived (≈ INFERRED) or left unstated (○ MISSING), and evidence bound to the exact product revision (Rev 17 → Rev 18).
- DETERMINISTIC ENGINE
What runs in deterministic code, and what does the AI step do?
- Peak isolation, regulatory limit evaluation, margin arithmetic, harmonic source correlation and 7-condition before/after retest comparisons run entirely in deterministic code, backed by 1,462 passing unit tests across the engine modules. The AI step writes investigation notes and next-test options constrained strictly to the candidate sources matched by the engine, and post-generation code checks audit every number and citation before display.
- ENGINEER AUTHORITY
Does Crado replace my EMC engineer, compliance lead, or test lab?
- No. Crado structures the technical record they work from and automates clause arithmetic and source correlation across revisions. Nothing extracted from a report or datasheet is marked ✓ KNOWN without a person confirming it. Measurement, chamber judgment, hardware design changes and certification remain with your engineers and accredited laboratories.
- WORKSPACE ISOLATION
Where does my hardware data go?
- Into your isolated workspace. Crado does not use your schematics, product facts or test reports to train models for other customers. Workspace checks learn only from your own revision history inside your workspace. Our public precedent library contains only public regulatory filings.
- LIVE VS. ROADMAP
Which parts are live today versus on the roadmap?
- Live today: radiated-emissions investigation under 47 CFR 15.109(a) (30–1000 MHz quasi-peak, Class B at 3 m), product memory, change and retest comparison, evidence state per revision, typing a single bench reading, the derived result at signing and the public case library. On the benchmark, 34 of 34 comparable readings on public FCC reports match the lab within 0.1 dB. Marked Roadmap: automated BOM file intake, pre-test risk review in the console, Class A at 10 m (47 CFR 15.109(b)), conducted emissions (47 CFR 15.107), radiated emissions above 1 GHz, peak and average detector conversions, distance extrapolation and EU CE / CISPR 32.
Start with your latest lab report or product revision.
See how Crado maps your board's frequency sources, evaluates margins against 47 CFR § 15.109(a), and keeps your investigation and revision history in one technical record.