Instrument · what the count rests on

What the headcount is made of.

The app on the previous page prints one number per minute. This page opens the back of the instrument and shows what that number rests on: every measured day on one clock, the four ways of computing a level and how often they agree, the minutes with a known headcount the count map passes through, what an empty floor reads night by night, how often each rule would have cried wolf on a night nobody was there, which receivers raise alarms alone, whether the phones in the room move with the radio reading, and whether a warm node reads differently from a cool one.

Also: each receiver and radio configuration, day against night.

Companion labs in Occupancy

In plain words Pick a day on the calendar and see people per minute: the Wi-Fi room level, calibrated that day against the Bluetooth devices the fleet heard, and fused with them. The people scale is a stated assumption you can move. A gap is a gap; nobody fills it with a zero.

The capture format Bluetooth in the room

Method for the level panels: Envelope, our transmitter only · Doppler 20–60 Hz, our transmitter only · Delivered-rate collapse · Envelope, every transmitter

Bystander devices against the envelope

Bystanders are non-public advertisers with no lab identity, from the fleet's continuous Bluetooth scan (since 14 September 2026); our own handsets are counted apart and the static public advertisers are the background. Where a half-hour segment edge cuts a minute, the larger part counts, because each segment re-salts every hash. The level is log2((n + 1) / (floor + 1)) against each receiver's own empty-night count, through the same floor rules, edges and receiver bootstrap as the CSI methods. It counts phones, not bodies.

Bucket agreement between `env_inj` and `ble_bystanders` on shared minutes: κ on four buckets, κ on the two-level collapse (empty+low against occupied+high), and the 2×2 of that collapse. The design's reading: agreement on quiet and disagreement on active means Bluetooth counts phones and the envelope counts bodies.
Scope	Minutes	Same bucket	κ (4)	κ (2)	Both quiet	Envelope only	Bluetooth only	Both active
2026-09-24	841	92 %	0.84	0.94	297	0	25	519
2026-09-25	692	35 %	0.19	0.36	173	0	246	273
2026-09-26	969	66 %	0.05	0.03	645	2	313	9
2026-09-27	609	56 %	0.02	0.01	344	1	261	3
2026-09-28	534	74 %	0.46	0.69	362	3	62	107
2026-09-29	355	98 %	0.53	0.59	348	2	2	3
2026-09-30	299	52 %	0.35	0.84	165	17	6	111
2026-10-01	403	37 %	0.22	0.37	126	97	32	148
2026-10-02	491	80 %	0.49	0.82	373	2	28	88
2026-10-03	360	100 %	1.00	1.00	360	0	0	0
2026-10-04	134	100 %	1.00	1.00	134	0	0	0
2026-10-05	403	53 %	0.35	0.61	204	65	11	123
2026-10-06	195	50 %	0.32	0.66	102	5	27	61
2026-10-07	255	27 %	0.11	0.64	176	32	2	45
2026-10-08	614	81 %	0.63	0.89	451	2	23	138
2026-10-09	135	60 %	0.14	0.50	112	7	7	9
2026-10-10	135	100 %	1.00	1.00	135	0	0	0
2026-10-11	135	100 %	1.00	1.00	135	0	0	0
2026-10-12	376	56 %	0.39	0.75	187	8	38	143
2026-10-13	88	100 %	1.00	1.00	88	0	0	0
2026-10-14	299	98 %	0.00	0.00	293	6	0	0
2026-10-15	211	100 %	1.00	1.00	211	0	0	0
2026-10-16	179	100 %	1.00	1.00	179	0	0	0
2026-10-17	578	97 %	0.49	0.85	562	0	4	12
2026-10-18	343	98 %	0.00	1.00	343	0	0	0
2026-10-20	493	86 %	0.67	0.88	92	4	16	381
2026-10-21	36	100 %	1.00	1.00	36	0	0	0
2026-10-22	180	100 %	1.00	1.00	180	0	0	0
2026-10-23	179	96 %	0.00	1.00	179	0	0	0
2026-10-24	179	100 %	1.00	1.00	179	0	0	0
2026-10-25	179	100 %	1.00	1.00	179	0	0	0
2026-10-26	179	100 %	1.00	1.00	179	0	0	0
2026-10-27	173	100 %	1.00	1.00	173	0	0	0
all	11231	76 %	0.50	0.68	7702	253	1103	2173

JSON: days · agreement · nights · false alarms · receivers · bystanders · temperature · episodes