Asking the fleet what it is doing…
monad-knowledge Wi-Fi sensing lab · FIIT STU
Campaign

c-flat-day-csi — the ray-traced CSI behind the explorer

c-flat-day-csi · exp-csi-crowd, walk-notebook

Archive snapshot, as of 4 h ago — the run corpus is rebuilt once a day, so this page is not a live reading. The fleet panel is the live one; it refreshes every 30 s.

Sessions

state
Session State Runs Synthesis Criticism Figures Verdict
E2S3XCPB 2026-06-25T23:58 finished 5 ✓ ✓ 2 Proxy headline confirmed on ray-traced CSI: the busiest, best-instrumented room is the least observable.

Brief

c-flat-day-csi — the ray-traced CSI behind the explorer

The c-flat-day-ai campaign produced a believable reactive crowd (18 LLM-authored behaviour FSMs) and the CSI explorer visualises a first-order occlusion + free-space-path-loss proxy over the 10 real csi-rp-rx-* receivers. This campaign replaces that proxy with the ray-traced article: couple the same FSM crowd trajectory into the exp-csi-crowd / Sionna chain (the csi-link-resplan-12439-dense10 layout) and emit real CSI tensors + per-link statistics over the evening.

Why

The explorer answers "where does the crowd go and which links does it block" with a defensible but analytic model. The thesis direction (BLE-calibrated CSI crowd sensing) needs the ray-traced signal — multipath, body-as-scatterer EM, per-subcarrier amplitude/phase — to confirm the proxy's headline finding: the densest room (the living-room hub) is the least observable. If the ray-traced signal agrees, the coverage-gap argument is solid; if not, the placement needs rethinking.

Plan

  1. Crowd leg: re-run the c-flat-day-ai FSM cast (resolved_fsm_cast.yaml, on S3 under experiments/ip107-flat-day-ai/) to regenerate trajectory.parquet deterministically (seeds 0–2).
  2. CSI leg: feed it to exp-csi-crowd with the e2e-resplan12439.yaml coupling and the dense-10 receiver layout → links.parquet + csi.hdf5 per seed.
  3. Compare the ray-traced links.parquet to the proxy links_proxy.parquet (same schema) — amplitude-vs-occluder slope, RSSI ranges, the living-0 coverage gap.
  4. Swap the explorer's data source from proxy → ray-traced (identical schema) so the replay shows real CSI.

Execution note

Sionna is not installed in the local venv (CPU-only on Apple Silicon, heavy). This campaign is the systematic / CI path: dispatch via sim-campaign.yml once the walk-notebook image carries the IP-107 agenda+FSM engine (after merge + GHCR build). The crowd leg already runs locally via uv run --with jupedsim; only the Sionna leg needs CI.