RuField adds its first modality adapter for camera free sensing

What RuField is

Camera free sensing is fragmented: WiFi people use one pipeline, radar people another, acoustics a third, and none of them agree on what an event is. RuField is the specification that tries to fix that — one typed description of a field observation, whatever produced it.

The repository is unusually honest about its own numbers. The v0.1 benchmark figures come from a deterministic synthetic simulator and are labelled SYNTHETIC in the README.

What changed

Commit 9955672 added UltrasonicReplayAdapter, the first adapter connecting an actual modality to the specification rather than a simulated one. The day before, commit 09a677b governed field evidence, uncertainty, standards and RuVector integration, and PR #5 added authenticated BLE and channel sounding evidence.

A CI commit in the same window pinned GitHub Actions to immutable commits, which is the sort of supply chain hygiene worth copying.

Get started

A Rust toolchain and a checkout. This is a standalone Cargo workspace:

cargo run -p rufield-bench

Expected result: the deterministic benchmark runs on the default seed and prints a labelled synthetic report.

There is a viewer for watching the field state:

cargo run -p rufield-viewer

Expected result: a local server on http://127.0.0.1:8088/ showing the synthetic field.

No documented MCP server. The supported surface is the Cargo workspace, the bench and the viewer.

Use it today

Practical case: evaluating whether a sensing idea is worth hardware money. Input is a seed and a modality adapter. Workflow is run the bench, read the JSON, compare configurations. Output is a reproducible comparison before you buy anything.

Acceptance test: run the bench twice with the same seed and confirm identical output. Then change the seed and confirm it moves.

Push it further

Experimental commentary. The real prize is cross modality fusion: WiFi for coverage, ultrasonics for fine detail, radar for motion, all producing the same typed event so a consumer does not care which sensor saw it.

Limitation: synthetic benchmarks say nothing about real rooms. Falsifiable test: replay a real ultrasonic capture through the adapter and check whether the confidence scores track ground truth you recorded yourself.