Campaign

BLE/CSI co-registration — does the RSSI anchor channel resolve occupancy where temporal CSI collapses?

c-ble-csi-coregistration · exp-csi-static

Archive snapshot, as of 5 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

Session	State	Runs	Synthesis	Criticism	Figures	Verdict
Z77Q1ANT 2026-06-24T14:34	finished	30	✓	—	1	Claim 2's enabling premise is partially met in sim: the co-registered BLE anchor robustly resolves occupancy PRESENCE (empty vs occupied, d=2.3-3.7, noise-robust) in the seated regime where temporal CSI collapses, but the response saturates after first occupants — coarse presence yes, graded headcount no. Motivation only, not proof. H1 NOT bumped (stays supported on literature weight); H2 stays plausible.
3R9REFP4 2026-06-04T13:46	finished	14	✓	—	—	First BLE/CSI co-registered corpus: the anchor channel resolves occupancy where temporal CSI collapses (ρ = −0.83/−0.66) but weakly (−0.115 dB/person) — resolvable ΔN spans ≈3 (fixed crowd, noise-only) to ≈10 (placement-averaged; ≈16 at 4 dB). Motivation for ble-periodic-calibration with a concrete design consequence: anchor placement near occupant zones is the next lever, before hardware. Synthesis inlined by the supervisor (15-run corpus, IP-084 cost rule); circularity boundary (self-authored noise) stated.

Brief

Question

With CSI and BLE/RSSI synthesized from the same ray-traced channel (co-registered by construction), does the BLE anchor retain occupancy information in the static regime where the temporal CSI counting feature provably collapses?

Framing discipline (per the 2026-06-04 survey verdict): this is in-silico motivation for ble-periodic-calibration / the L7.4 evaluation spine — not a hypothesis test. The BLE measurement noise is a self-authored Gaussian (σ = ble.noise_db) standing in for per-event fading/MAC jitter; any accuracy-style claim against it would be circular. The honest deliverable is a sensitivity table (what ΔN is resolvable at what noise, given the event count), and the real defeater is multi-week hardware capture on ≥2 sites. The dishonest "≥30% MAE reduction" headline is explicitly out of scope.

What we already know

c-csi-crowd-temporal (sealed 01KT99FX2Z…): the temporal CSI feature is motion-driven — walking ρ(N, CV) = +1.0, but seated-window CV collapses to ≤ 3.7e-6. A seated/static crowd is where CSI-only counting starves; that is the regime a periodic BLE anchor must cover.
The static occupancy corpora show mean |H| falls with N (blockage) — the RSSI anchor inherits exactly this channel, narrowband, at 2.4 GHz.
sionna-csi-runner synthesizes BLE since 2026-06-04 (ble.enabled): per-(frame, link) RSSI from a ~2 MHz centre slice of the same CFR + per-event Gaussian noise; smoke-verified (01KT9BJYNV3TTH7K9PC2XYVQAB, 40 adv events).
Caveat inherited from c-csi-fidelity-material (sealed 01KT9BJ82A…): the palette is load-bearing; this session runs the same uniform-drywall default, so absolute dB magnitudes are conditional on the tag.

What the supervisor does

30 runs on the through-wall rig (static placement = seated-crowd proxy): main grid N {0,2,4,6,8,12} × seeds {0,1,2} at noise 2.0 (18 runs) + the noise arm {4.0,6.0 dB} × N {0,6,12} × seeds {0,1} (12 runs) — the wider noise sweep traces how the resolvable step degrades as the self-authored event noise grows. Reduction pools ble_links.parquet per run: per-N mean RSSI ± per-event σ, Spearman ρ(N, mean RSSI) per seed, and the resolvable-ΔN table (effect size d = |Δmean| / SE after event averaging) as a function of noise_db. Cite zhao2023_b424 (NeRF2 — learned RF synthesis as the adjacent approach), demrozi2021 and longo2019 (BLE occupancy sensing) where the vault carries them.

Out of scope

Any BLE-vs-CSI estimator MAE comparison (circular against the self-authored noise model).
Moving crowds (the coupled kind: groups BLE variant is the follow-on once this floor exists).
Hardware noise models, MAC scheduling, advertising-channel hopping.

Expected interpretation

Anchor informative (ρ ≤ −0.6, ΔN ≤ 4 resolvable at noise 2.0) → "in the seated regime where temporal CSI collapses, a co-registered BLE anchor still resolves coarse occupancy — the in-silico motivation for hybrid BLE-anchored calibration." Strength of ble-periodic-calibration stays plausible (motivation, not proof).
Anchor uninformative (flat RSSI or ΔN > 12) → the anchor idea needs link-geometry rework (anchor placement near occupant zones) before hardware investment — a cheap, valuable negative.
Criterion 1 fails → BLE plumbing finding (artefact/manifest bug), fix before interpretation.