22 cohorts · 49,222 users · 3,701,902 checks · 13,964 state changes
Generated 2026-09-01 03:35 UTC. Each cohort is a differently-sampled set of chess.com accounts,
polled on a schedule; failing is how many of them the stats endpoint currently 404s.
| list | users | failing | rate | churn | recovered | sampled from | |
|---|---|---|---|---|---|---|---|
▋ |
team-india | 743 | 8 | 1.1% | 0.299% | 49% | /pub/club/team-india/members, weekly+monthly — active members of a 46k club whose endpoint flaps |
▌ |
dev-community | 1,096 | 10 | 0.9% | 0.231% | 51% | /pub/club/chess-com-developer-community/members, weekly+monthly — recently active members of the API's own club |
▍ |
club-admins | 533 | 3 | 0.6% | 0.112% | 51% | /pub/club/club-admins/members, weekly+monthly only — recently active club admins |
▏ |
nm | 2,562 | 6 | 0.2% | 0.853% | 55% | /pub/titled/NM — 100% titled, tests whether failure rate grades with title strength |
▏ |
cm | 2,773 | 6 | 0.2% | 0.401% | 55% | /pub/titled/CM — 100% titled, tests whether failure rate grades with title strength |
▏ |
ee-all | 12,036 | 26 | 0.2% | 0.435% | 51% | /pub/country/EE/players |
▏ |
fm | 4,848 | 10 | 0.2% | 0.500% | 55% | /pub/titled/FM — 100% titled, tests whether failure rate grades with title strength |
▏ |
team-usa | 4,192 | 7 | 0.2% | 0.285% | 53% | /pub/club/team-usa/members, full roster — 1.9% titled, between the untitled and titled cohorts |
▏ |
team-finland | 4,916 | 8 | 0.2% | 0.216% | 54% | /pub/club/team-finland/members, full roster — large, 0.2% titled, independent of the other cohorts |
▏ |
wcm | 662 | 1 | 0.2% | 0.085% | 47% | /pub/titled/WCM — women's candidate master, part of the titled-cohort family alongside gm/im/fm/cm/nm |
▏ |
gm | 1,727 | 2 | 0.1% | 0.386% | 52% | /pub/titled/GM |
▏ |
im | 2,647 | 3 | 0.1% | 0.612% | 53% | /pub/titled/IM — 100% titled, tests whether failure rate grades with title strength |
▏ |
wfm | 958 | 1 | 0.1% | 0.214% | 53% | /pub/titled/WFM — women's FIDE master, part of the titled-cohort family alongside gm/im/fm/cm/nm |
▏ |
team-germany | 7,360 | 5 | 0.1% | 0.236% | 52% | /pub/club/team-germany/members, full roster — 0.4% titled, largest club whose members endpoint works |
|
user | 118 | 0 | 0.0% | 0.345% | 51% | scraped from forum threads already reporting 404s |
|
ee-titled | 32 | 0 | 0.0% | 0.054% | 51% | member-search page, titled filter |
|
int5 | 392 | 0 | 0.0% | 0.669% | 71% | — |
|
streamers | 743 | 0 | 0.0% | 0.432% | 52% | /pub/streamers — 34% titled, all high-traffic |
|
wgm | 266 | 0 | 0.0% | 0.266% | 52% | /pub/titled/WGM — women's grandmaster, part of the titled-cohort family alongside gm/im/fm/cm/nm |
|
wim | 476 | 0 | 0.0% | 0.143% | 50% | /pub/titled/WIM — women's international master, part of the titled-cohort family alongside gm/im/fm/cm/nm |
|
linux-club | 109 | 0 | 0.0% | 0.133% | 51% | /pub/club/linux-users-group/members — a self-selected club |
|
wnm | 33 | 0 | 0.0% | 0.017% | 50% | /pub/titled/WNM — women's national master, part of the titled-cohort family alongside gm/im/fm/cm/nm |
| total | 49,222 | 96 | 0.2% |
Bars are linear to 20%. Sorted by current failure rate.
/pub/club/<id>/members is how cohorts get built from club rosters. The two
largest clubs are exactly the two that fail, and a club's two sub-endpoints fail
independently — team-india serves its size but not its roster.
| club | members | /members |
last change |
|---|---|---|---|
team-india |
46,090 | down (404) | 2026-08-28 12:55 |
chess-com-developer-community |
22,645 | up | 2026-09-01 02:55 |
team-germany |
7,347 | up | 2026-08-31 22:55 |
club-admins |
6,860 | up | 2026-09-01 02:55 |
team-finland |
4,907 | up | 2026-09-01 00:55 |
team-usa |
4,175 | up | 2026-09-01 01:55 |
linux-users-group |
109 | up | 2026-08-28 15:55 |
chess-com-community |
— | down (404) | 2026-08-21 08:41 |
improvers |
— | down (404) | 2026-08-24 16:55 |
kings-clash-royale-club |
— | down (404) | 2026-08-21 08:41 |
failing is the share of users currently at a non-200 status. changes counts
recorded state transitions, excluding first sightings. recovered is the share of
those transitions that went 404 → 200.
churn is changes per individual check — comparable across cohorts of different sizes,
since a raw count scales with both list size and how often a list has been swept.
It is not comparable across cadences. State moves with time, not with polling,
so checking a cohort twelve times as often finds roughly a twelfth as much change per
check. small moved to 5-minute sweeps to catch transient failures, and its churn
will fall accordingly without anything about the cohort having changed.
It divides by checks, not users × runs, because runs are not all the same size:
the recheck tier sweeps only currently-failing users.
Failing users are checked more often than healthy ones. The recheck tier runs
every 30 minutes against whoever is currently at 404, while a full sweep of a large
cohort runs every 6 hours. Recoveries are therefore detected far sooner than
breakages, recovered is biased upward, and churn is a weighted blend of two
different rates rather than one clean number. Compare cohorts with it; do not read
it as an absolute failure probability.
Churn also needs volume. A cohort with a few hundred checks behind it can read 0.000% purely by chance while behaving like one that reads 0.1%.