01 · Most requested
Recording Integrity
Can the integrity of the recording be established?
Wireless acquisition loses packets, and most pipelines join the signal across the gap without saying so.
What it examinesWhere do you stop
trusting the data?
Technical examinations for BCI and neural-data systems. One question, one boundary, one verdict — backed by evidence, with what could not be established said out loud.
An examination can be drawn around any boundary in this chain. Choose a stage.
Most requested
Where it breaks today.
Almost every team works from recordings, after the fact. These three questions decide whether anything built on them holds.
04 · Most requested
Synchronisation
Can the timing between streams be established?
Every multimodal study rests on clocks that agree. Left to their defaults, they rarely do.
What it examines07 · Most requested
Dataset Readiness
Is the dataset fit for its stated downstream purpose?
Models are trained on pooled recordings. One broken file in the pool shapes the model, and nobody sees it.
What it examines
The catalogue
Twelve examinations.
Twelve examinations.
Each answers one question.
Not sure where to start? Run a Preflight on one recording first — in your browser, in seconds.
The recording
Can it be used at all?
01Recording IntegrityCan the integrity of the recording be established?Most requested
Examines
- Packet loss, gaps and truncation
- Duplicated and corrupted samples
- Malformed records
- Timestamp continuity
You receive
A Recording Integrity Map: what happened, where, how much, the evidence, and its impact.
Transport · Raw data
Start with a Preflight02Channel IntegrityCan each channel be trusted, end to end?
Examines
- Missing, flat and saturated channels
- Abnormal ranges and dropouts
- Channel consistency
- Metadata against signal
You receive
A Channel Integrity Map: every channel VALID, SUSPICIOUS, INVALID or UNKNOWN.
Acquisition · Raw data
Start with a Preflight03Event IntegrityCan the events and markers be trusted?
Examines
- Missing and duplicated events
- Event order and timestamps
- Marker against data alignment
- Event consistency
You receive
An Event Integrity Map of missing, duplicated or displaced events.
Raw data
Devices and streams
Did the hardware and the clocks do what they claim?
04SynchronisationCan the timing between streams be established?Most requested
Examines
- Clock relationships, offset and drift
- Buffering and transport latency
- Resampling
- Event alignment across streams
You receive
A Synchronisation Map: reference, stream, offset, drift, jitter, confidence.
Acquisition · Transport · Raw data
05Acquisition ValidationDoes the acquisition system produce what its contract states?
Examines
- Sampling rate and channel configuration
- Resolution and ranges
- Acquisition parameters and metadata
- Device output against documented behaviour
You receive
An Acquisition Validation Report. A file cannot prove what was never measured, and the report says so.
Device · Acquisition · Transport
The chain to AI
What reached the model, and can it be shown?
06ProvenanceCan the origin and history of the data be established?
Examines
- Source and acquisition
- Processing, transformation and export
- Storage and versioning
- Undocumented transformations
You receive
A Provenance Map from device to model input, with its gaps named.
Device · Raw data · Processing · Derived data · AI
07Dataset ReadinessIs the dataset fit for its stated downstream purpose?Most requested
Examines
- Completeness and integrity
- Channel quality and sampling consistency
- Events, artefacts and missing data
- Known limitations
You receive
READY, DEGRADED, NOT READY or UNKNOWN. Ready means no technical blocker within scope, not scientific validity.
Raw data · Processing · AI
Start with a Preflight08ReproducibilityCan the claimed result be reproduced from the evidence supplied?
Examines
- Source, dataset and dependencies
- Environment, versions and parameters
- Configuration and execution conditions
- Undocumented assumptions
You receive
A Reproducibility Boundary: what reproduces, what does not, and why.
Raw data · Processing · Derived data
09Data Contract ValidationDoes the system behave as its data contract says?
Examines
- Schemas, types and units
- Ranges, timestamps and nullability
- Versioning and API behaviour
- Edge cases against documentation
You receive
A Data Contract Matrix: requirement, observed condition, evidence, verdict.
Raw data · Derived data · AI
10Device → AI BoundaryWhat happens to information between acquisition and AI?
Examines
- Preprocessing and feature extraction
- Derived data, transport and storage
- The API in front of the model
- What is observed, measured, declared, inferred or unknown at each stage
You receive
A Device → AI Boundary Map: where information is preserved, transformed, restricted — or lost.
Transport · Raw data · Processing · Derived data · AI
Claims
Where does the evidence stop?
11Evidence ExaminationHow strongly does the evidence support the claim?
Examines
- Documentation, source code and tests
- Benchmarks and measurements
- Formal verification and independent reproduction
- Contradictory evidence
You receive
A Claim → Evidence Matrix: each claim, the evidence for it, and how strong that evidence is.
Device · Acquisition · Transport · Raw data · Processing · Derived data · AI
12Failure BoundaryWhere does the evidence stop supporting the claimed behaviour?
Examines
- Stated claims and assumptions
- Implementation, tests and measurements
- Boundary and failure conditions
- Contradictions and unknowns
You receive
A Failure Boundary Map: where it holds, where it degrades, where it fails, and the mechanism. The DY PROOF method.
Device · Acquisition · Transport · Processing · AI
The DY PROOF method
See one run
Drop a recording.
Drop a recording.
See what holds.
The Preflight preview reads an EEG file in your browser — CSV from OpenBCI or BrainFlow, EDF, BDF — and returns a verdict with its evidence in seconds. Nothing is uploaded.
How a verdict reads
Not everything is pass or fail.
What we conclude
About a claim, within the agreed boundary.
SUPPORTEDThe evidence supports the claim.PARTIALLYIt holds for part of the claim, or under stated conditions only.NOT ESTABLISHEDIt may be true; the evidence supplied does not show it.FAILEDA concrete failure was found inside the boundary.About data: READY · DEGRADED · NOT READY · UNKNOWN
How we know
Every finding carries how it is known, and these are never mixed.
OBSERVEDSeen directly during the examination.MEASUREDQuantified, with the method stated.REPRODUCEDRebuilt by us from what you supplied.FORMALLY VERIFIEDProven by a method that applies to that property — and to nothing else.DECLAREDStated by you, a vendor or the documentation.INFERREDDerived from evidence, not established directly.NOT ESTABLISHEDThe evidence available does not reach it.Every examination ends where the evidence ends.
Early access
Open to early users.
Examinations run on our infrastructure. You send the question and the evidence; you receive the verdict.
APIAccess by API
Early users receive API access by invitation: submit evidence, receive findings as machine-readable evidence.
REPORTA written verdict
Question, scope, findings, evidence, severity, confidence, limits and what was not tested.
NON-DESTRUCTIVEYour data, untouched
The original data is never modified. Any correction is a separate step you ask for.
ON REQUESTPricing on request
Scope and price come back in writing before any work begins, under a mutual NDA if you need one.
- Describethe technical concern.
- Definethe question it raises.
- Agreewhere the boundary sits.
- Sharerepresentative evidence.
- Examineit, without changing it.
- Reviewevery finding technically.
- Deliverthe verdict and its evidence.
Where it helps, an examination checks selected requirements of ISO/IEC TS 27571:2026, the BCI data format for non-invasive collection, or of ISO/IEC 27572, the BCI reference architecture. An examination does not confer conformity with either.