Run your own file through the ten-stage deterministic pipeline.
Drop a small telemetry-style CSV, JSON or NDJSON file and watch every record move through CalibrAIte Engine: ingest, resolve identity, normalise, classify, correlate, learn normal, calibrate, route, audit and output. Every flag comes with its literal reason, so you can see exactly why a record was classified or flagged. This public sandbox is intentionally isolated. It does not update site documents, customer RAGs or live mine systems; those are connected only inside a scoped paid pilot.
Public sandbox
The upload demo runs in an isolated public path. It proves the Engine logic without touching a customer tenant, site RAG or production document store.
No site docs update before pilot
Site documents, procedures, SharePoint folders and regulatory evidence packs are not updated from this page. They are connected only inside a scoped paid pilot.
Human decision boundary
CalibrAIte drafts, cites and routes. A competent person still reviews the output and decides what happens operationally.
Drop one file. Watch every row complete ten deterministic stages.
The sandbox accepts CSV, JSON or NDJSON telemetry-style events. It runs in memory against the public deterministic Engine path, returns the same result for the same input, and does not write to a customer RAG, production document store or site document register.
No site-doc writes
This page never updates customer site documents, production RAGs, SharePoint, Teams folders or regulatory packs.
Sandbox only
The public path is in-memory classification plus response trace. It proves the Engine logic without touching live mine systems.
Pilot unlock
Actual site documents are connected, governed and audited only after a scoped paid pilot is agreed.
The hot path does not call an LLM. The same uploaded records, in the same order, produce the same classification, correlation, audit trace and output.
Input
CSV, JSON or NDJSON event rows from fleet, telematics, IVMS, SCADA, maintenance or contractor systems.
Engine trace
Every row returns normalised fields, severity, WA basis, anomaly state, correlation and route recommendation.
Output
A visible table, ten-stage trace and API-shaped response for the selected record.
Ingest
Receive raw events from file, API, webhook or controlled sync.
Resolve identity
Map source names to the right asset, equipment type and context.
Normalise
Convert every source into one CalibrAIte event schema.
Classify
Score the event against deterministic rules and WA mining basis.
Correlate
Join related events across the upload session and recent context.
Learn normal
Compare readings with operating bands and file-level patterns.
Calibrate
Flag gaps, drift or anomalies without changing controls.
Route
Draft the right action for the right workflow or role.
Audit
Keep raw, normalised, classification and route evidence together.
Output
Return table rows, cited basis and API-shaped payloads.
Every flag is explainable on its own record. Nothing is a black box.
A fair question when you run a single record: anomalous compared to what? The answer is that CalibrAIte does not need a population of data to justify a single event. The Engine is deterministic. It reaches a verdict from the record in front of it against known rules, equipment operating bands and the WA mining basis, and it writes down the exact reason for every decision. Two different things are happening, and the demo shows both.
Classification & anomaly: rules and thresholds
One record is enough to classify and, where warranted, flag. The Engine assigns a severity band, then flags an anomaly when that band is elevated or a metric falls outside its expected operating range. Each flag carries its literal reason. Examples of what trips it from a single event:
- An explicit severity field, or a keyword like critical, e-stop, fatigue, overspeed, brake fault.
- A temperature reading above 105 C, or a proximity distance under 10 m.
- Hydraulic volumetric efficiency under 85 percent, or pre-charge below nominal: developing faults, not just hard alarms.
- Any numeric metric outside its known operating band, returned as: out-of-range metric X=Y flagged for review.
So with one row, the reason column tells you exactly why it was flagged. It is a rule or a threshold, not a guess, and not a statistical hunch.
Correlation & learned-normal: the stream
The higher-value signal needs more than one event. As records flow in, the Engine joins related events across the session and learns what normal looks like for that asset, so it can spot the patterns a single reading can never show:
- A brake fault on one truck and an e-stop two minutes later read as one cascading story, not two unrelated alarms.
- Repeated proximity events in the same zone become a proximity cluster instead of isolated warnings.
- A reading that looks fine in isolation is flagged because it has drifted from this asset's own operating pattern.
This is why a richer file is more convincing than one row. Drop a handful of related events and you will see correlation groups and learned-normal drift appear, which a single record cannot demonstrate by design.
Try it: upload a few related rows, for example a brake fault and an emergency stop on the same truck seconds apart, or three proximity events in one zone. Watch the reason column on each record and the correlation group that forms across them. One record explains itself; several records tell a story.
Site documents are governed only after pilot scope.
The public upload demo shows the Engine path on your supplied file, but it is not a site document ingestion workflow. A real site connection starts with a paid pilot scope: what documents are approved, which systems connect, who can see outputs, how audit trails are retained, and which human roles approve changes.