Interactive lab · a live network on your screen · no video

Inject a fault. Watch the counters. Close the loop.

This is a working simulator, not a slideshow. Twelve cells write real Ericsson counters every ROP. Break one — passive intermodulation, a mobility fault, a sleeping cell — then advance the clock and watch the exact counters climb, the vendor’s KPI formulas compute the damage, four AI detectors race to flag it, a classifier name the cause with SHAP evidence, and a fix bring the drop rate home against a control cell. Everything below is computed live from the same hidden physics your action changes.

STEP 1 / 6
🕒 Day 0 / busy-hour ROP
Inject fault into —
the market · 12 cellslive
Click a cell to inspect it in the stations on the right. Colour = current drop rate. The dashed cell is your matched control (never faulted — the honest yardstick for verification).
selected cell · 6-day drop-rate
patient (selected) control detector flag action
station 1 · the counter warehouseday 0

1The raw counters — this ROP

The node’s honest 15-minute self-measurement. Peg = event count, gauge pair = sum ÷ samples, PDF = a histogram whose shape is a diagnosis. Advance the clock and watch the guilty counters move — every number here is generated from the cell’s hidden physical state.
pmRadioRecInterferencePwr PDF · UL interference
quiet ← thermal floorloud →
CQI distribution PDF · channel quality
← poorexcellent →
station 2 · the KPI engineverbatim 23.Q2.5

2The five vital signs — computed from those counters

Raw counts mislead (a big cell has more of everything), so the vendor’s formula turns them into comparable rates. Click any vital sign to reveal its exact formula with this ROP’s counter values substituted.
station 3 · the detector benchday 6 · 5 · 3 · 2

3Four detectors, raced on the live history

Each detector scans the selected cell’s real drop-rate history and flags the first day its rule fires. The bar fills to the flag day; earlier is better — but earlier also means more false alarms. Run these after you’ve injected a fault and advanced a few days.
Inject a fault and advance the clock, then read the race here.
station 4 · diagnosiscause tree · SHAP · PIM test

4Which repair shop owns it?

The classifier reads which child of the drop-cause tree swelled, weighs the evidence into a SHAP ledger, and — for radio faults — runs the PIM-vs-external alibi test (does the interference track the cell’s own load?).
the drop-cause tree
SHAP evidence ledger
PIM vs external — interference × own load
verdict
— advance the clock and let the evidence build —
station 5 · act & verifypatient vs control

5Close the loop — inside guardrails

The recommended action appears once a cause is confident. Apply it, advance the clock, and watch the cell recover — measured against the untouched control cell, because a Friday that was just Friday is the oldest lie in optimization.
Diagnose a faulted cell first — the action catalogue unlocks per cause.
the AI NOC console · autonomous fleet sweepshadow mode
Autonomy
the work queue — ranked by expected gain0 open
impact ledger — this session
fleet health