Signal 01
CPU load / execution duration
1× is the calibrated baseline · higher means less CPU available to this page
slowdown baseline
Collecting samples
60 seconds agonow
Open sender ↗
WASM
WORKER
ISOLATED
Receiver setup
Receiver sampling
Receiver settings are locked while a channel test is running. Stop the test to change them.
Calculating total probe cost…
Work and cadence apply to each worker. Changing them or the worker count recalibrates the whole pool. More workers add load and do not guarantee distinct CPU cores. Timed occupancy includes pauses during execution; it is not CPU usage.
Loading graph preference…
Stop visuals, spectrum analysis, local test load, and channel searches. The load probe keeps measuring. Display choices are saved in this browser.
Load periodicity
Spectrogram
Frequency on the vertical axis · time on the horizontal axis · black = weak, white = strong
Black = weak · white = strong · blue-black = no analysis
Move over the plot to inspect time, frequency, and amplitude.
Collecting samples…
New columns appear at the right once a second and scroll left. A bright horizontal band shows when that frequency was active. Each column summarizes the FFT window ending at that time. Shorter windows reveal changes sooner; longer windows separate nearby frequencies but spread activity over more time. At least 32 probe samples are needed per window. Set Max Hz to focus on low frequencies, or lower the intensity ceiling to brighten weak activity. Pauses leave blue-black columns; changing the probe signal, sampling settings, or FFT window clears history.
Peaks suggest repeating load; they do not identify its source. The analyzer itself adds periodic work, so compare with quiet measurements before attributing a peak. Faster sampling extends the frequency range but costs more CPU. Frequencies above the measured sampling limit can alias into lower peaks.
For example, alternating 0101 at 1 bit/s repeats every 2 seconds: look for 0.5 Hz, with possible harmonics. Frame timing and the Smooth 5 setting do not enter this analysis.
Relative CPU availability / live probe
Is the host busy,
or was it our frame?
A fixed-work WebAssembly loop runs off the main thread. Its slowdown is compared with frame timing to make slow-frame attribution less speculative.
Current CPU pressure
—
Warming up
Preparing and calibrating the Wasm loop…
Mean probe duration
— baseline — wake-up delay —Calibration jitter
— collecting signalTotal timed occupancy
— effective rate — timer resolution —Signal 02
Frame delays
Only slow frames · height is time over one refresh interval
No slow frames in the last 60 seconds
0 slow frames this session
0 with CPU pressure
Validation lab
Try to fool the probe
Block only the main thread
Frames should stall. The worker probe should stay close to 1×.
Saturate logical CPUs
Busy workers compete with both rendering and the probe.
Attribution log
Recent slow frames
- No slow frames observed yet.
Remote channel test
Find the fastest reliable symbol rate
Choose receiver workers and a measurement signal in Receiver sampling above. Each trial compares execution duration, scheduling delay, and their combination from the same raw samples. The selected signal determines the search result. Frame delays do not enter the decision.
Advanced decision settings
Fastest reliable observation
—Connect both browsers with the same pair code. The search pauses all visual analysis while measuring and restores it afterward. For a visual check, open the sender ↗ and send a pattern at a fixed rate.
BER is the fraction of known 0/1 states decoded incorrectly. Δarea is the difference between integrated load and idle values of the selected signal. Workers are integrated separately and weighted equally; missing worker data reduces coverage. Comparisons use known bits to fit their thresholds and are supervised estimates.
- No trials run yet.