Yes — markedly more than this morning. The 08:17 sample found no occupied French cell at any useful precision. The 17:58 sample found four, including a live French-language conversation in northern Paris and a greeting exchange in Montpellier.
The change is in kind, not just degree: this morning all activity sat in continental cells spanning six countries, where location carries almost no information. This evening, people appeared in cells precise enough to name a suburb.
| Measure | 08:17–08:42 | 17:58–18:28 |
|---|---|---|
| Window length | 25 min | 30 min |
| Occupied local cells (4+ char) | 0 | 4 |
| France-overlapping cells, all precisions | 2 | 12 |
| Chat messages | 4 | 17 |
| Distinct speakers | 1 | 5 |
| World events seen | 4,211 | 3,295 |
| World distinct cells | 36 | 75 |
Worldwide event volume actually fell while distinct cells doubled — the evening traffic is spread across more places rather than being louder overall. Heartbeat beacons dominate the raw event count, so it is a poor proxy for how many people are talking.
These are the four cells precise enough to locate. Extents are the geohash cell bounds, not a user's position; nearest places are given for orientation.
| Cell | Approx. location | Extent (lat / lon) | Ident. | Talkers | Events |
|---|---|---|---|---|---|
| u09w | Northern Paris — Saint-Denis / Clichy | 48.9–49.0 / 2.1–2.5 | 4 | 1 | 22 |
| spfb004 | Montpellier — cell is ~150 m across | 43.6 / 3.9 | 1 | 1 | 2 |
| gbrp | Loire-Atlantique, north of Nantes | 47.6–47.8 / −1.4–−1.1 | 1 | 0 | 2 |
| u0k1 | Jura, near Lons-le-Saunier | 46.6–46.8 / 5.6–6.0 | 1 | 0 | 1 |
Only u09w and spfb004 carried conversation. The other two are silent presence beacons.
Cells of three characters or fewer are too coarse to attribute to a country. They are listed because people are in them, and because one of them is easy to misread.
| Cell | Covers | Ident. | Talkers | Events |
|---|---|---|---|---|
| gc | Ireland, UK, N Atlantic — touches France only at the Channel coast | 42 | 0 | 233 |
| u0 | E/N France, Switzerland, S Germany, Benelux, N Italy | 5 | 2 | 29 |
| u1 | Benelux, Germany, Denmark — France only at its northern edge | 2 | 1 | 19 |
| sp | S France, Corsica, E Spain, Italy, W Mediterranean | 2 | 1 | 15 |
| g | N Atlantic, Greenland, British Isles | 1 | 0 | 27 |
| gcp | Southern England | 1 | 1 | 2 |
| ezz | Bordeaux / Arcachon basin | 1 | 0 | 1 |
| u09 | Paris basin, broader | 1 | 0 | 1 |
Do not read gc as French. It is the largest cluster in the whole sample — 42 identities, 233 events — but it spans Ireland, the UK and the North Atlantic, and meets France only along the Channel coast. It appears here because its bounds clip the French border, not because anyone in it is in France. It also produced no chat at all.
Summarised rather than quoted. These are public relay events, but they are ordinary people's conversations tied to small geographic cells, so this report characterises them instead of reproducing them.
| Cell | Msgs | Speakers | Character |
|---|---|---|---|
| u09w | 5 | 1 | A sustained personal argument in French. The only real conversation in either sample. |
| sp | 3 | 1 | French, confrontational, aimed at another user; includes one image link. |
| u0 | 5 | 2 | Connectivity checks and a greeting, plus two client-generated block notices. |
| spfb004 | 2 | 1 | A French greeting pair — someone opening a channel in Montpellier. |
| gcp | 1 | 1 | An English greeting, same speaker as Montpellier moments earlier. |
| u1 | 1 | 1 | A single emoji. |
Two structural points. Most of the French-language traffic is one person each in two cells, not a group — only u09w shows anything resembling a dialogue. And one identity recurred across both samples ten hours apart, which means these channel keys persist rather than rotating per session; that is what makes repeat observation of an individual possible at all.
Relays retain these ephemeral events for roughly five minutes, so there is no way to look back over days. Every run sees only its own window. Both samples here are from the same day; this is a morning-to-evening difference, not an established trend.
Cells at different precisions carry different pubkeys with near-identical event counts, which points to per-channel derived keys rather than distinct users. One person can appear as several identities. Do not sum the identity columns.
Bitchat is primarily a Bluetooth mesh app; these geohash channels are the internet-bridged portion, observed through nine specific relays. Anyone using the local mesh without bridging is invisible here. Every figure is a floor.
This traffic is bursty: one person joining produces most of a window's messages. The 17 evening messages come from five speakers, and nearly half from one argument. A different half-hour could plausibly show much less.
At 4+ character precision this method pairs a persistent key, a timestamp and a small area with message content — enough to follow one person across sessions. Aggregate occupancy is a reasonable thing to measure from public relays. Message text and keys are deliberately left out of this report, and should stay out of anything shared onward.
Rather than enumerating French cells into a relay filter — roughly 700 at 4-character precision, which risks filter-size limits and misses whatever precision clients actually use — the capture subscribes to all kind-20000 and 20001 geohash traffic, decodes every g tag to coordinates, and classifies it. Cells of 4+ characters are tested against a 17-point France outline plus a Corsica box; coarser cells are reported as overlaps rather than dropped. An earlier version tested cell centres only, which wrongly excluded sp because its centre falls in the Mediterranean; that is fixed here.