What the production telemetry shows.
These are the actual numbers from our own production database for the 30 days ending 2026-05-07. No extrapolation, no rounding, no marketing-rewriting. The methodology and SQL that produced every figure is linked at the bottom. We publish this page because the most useful thing a small project can offer reviewers is a verifiable account of where it actually stands today, separate from where it intends to go.
01Window
call_events rows29,687Source: call_events on production Postgres
(krolik-postgres-age:17). Country bucket from server-side GeoIP
(MaxMind GeoLite2) at ingest, retained as ISO 3166-1 alpha-2
only; no IP retained. Telemetry is default-on for the free tier;
events recorded are aggregate counters (page_view, room_created,
room_joined, call_connected, call_ended) with no message content,
no IP storage, no fingerprinting input.
02Censored-region cohort (RU + IR + CN + BY + CU)
| Country | page_view | rooms_created | rooms_joined | call_connected | devices |
|---|---|---|---|---|---|
| RU | 24 | 11 | 10 | 1 | 24 |
| CN | 1 | 0 | 0 | 0 | 1 |
| IR | 0 | 0 | 0 | 0 | 0 |
| BY | 0 | 0 | 0 | 0 | 0 |
| CU | 0 | 0 | 0 | 0 | 0 |
| Total | 25 | 11 | 10 | 1 | 25 |
25 page_views, 11 rooms created, 1 call connected. The 24 RU rows
include WebRTC pipeline diagnostics
(candidate_pair_selected, ice_state_change, media_flow_sample) —
the protocol stack is being exercised inside Russia, not failing
at the page-load step. This is the moment before grant funding,
before paid GTM, before partner-edge rvpn fleet conversion.
03Free-world cohort (all attributed countries except RU/IR/CN/BY/CU)
| Country | page_views | calls_connected | devices |
|---|---|---|---|
| US | 1,243 | 176 | 285 |
| NL | 47 | 0 | 24 |
| DE | 22 | 7 | 21 |
| CA | 13 | 2 | 5 |
| SE | 9 | 2 | 11 |
| LT | 9 | 1 | 4 |
| CZ | 8 | 0 | 6 |
| GE | 7 | 0 | 3 |
| PL | 5 | 0 | 3 |
| LV | 5 | 0 | 2 |
PV→call_connected ratio is 13.7% in the free world vs. 4.0% in censored regions — a ~3.4× gap. The gap is not user motivation (RU users open more rooms per page-view: 46% vs. 13%) but transport reliability at the connection step. That is the DPI signal the grant work is designed to close.
04M12 targets, anchored on real distribution channels
| Channel | Subscribers | Geography | Channel type |
|---|---|---|---|
rvpn (commercial VPN partner) | ~350K | China | email-blast to subscriber base |
| Piter Online (founder-owned media portal) | ~250K | Russia | editorial publication |
| Fiesta (partner project) | ~800K | Russia | partner editorial publication |
| Direct-reach total | ~1.4M | CN + RU | founder-controlled or close-partner |
The M12 censored-region MAU target is 6,700 distinct devices: RU 4,000 + CN 1,500 + IR 1,000 + BY/CU 200. That implies ~0.48% conversion against the 1.4M direct-reach surface — well below conventional opt-in launch rates. The PV→call_connected ratio target of 9% equals today's free-world ratio: reaching it means DPI is no longer breaking the call.
| Metric | M0 baseline | M12 target |
|---|---|---|
| Censored-region MAU — total | 25 | 6,700 |
| RU MAU | 24 | 4,000 |
| CN MAU | 1 | 1,500 |
| IR MAU | 0 | 1,000 |
| BY + CU MAU | 0 | 200 |
| Censored call_connected (rolling 30d) | 1 | 2,000 |
| Censored PV→call_connected ratio | 4.0% | 9% (free-world parity) |
| Free-world MAU | 371 | 15,000 |
05Methodology
Every number on this page is reproducible by running the queries
published in docs/strategy/BASELINE-MEASUREMENT.md §11 against the production oxpulse Postgres database.
Read-only credentials are available to grant auditors on request,
NDA-gated. The full methodology — country buckets, retention,
the residual 23.7% NULL country-attribution share — is documented
in the same file. The framework that defines the metric set,
privacy posture, and failure thresholds is in docs/strategy/IMPACT-METRICS.md (v2). Both files are
committed to the project repository and versioned alongside the
code.
Known caveats: GeoIP attribution coverage is 76.3% in this window
(improving — 91% over the trailing 14 days vs. 64% for the
preceding 14). One call_connected from RU is statistically tiny;
we report it as the raw count, not a confidence interval. Direct
TLS-handshake telemetry from a residential vantage inside RU/IR
is not yet measured — it is a grant month-2 deliverable. The
indirect inference today is the 90% attrition between room_joined and call_connected for RU
rows, consistent with TURN-allocation or DTLS-handshake failure
under DPI.