160 earlier updates, newest first. Showing 21–40 on page 2 of 8.
Released
Android
Android 1.1.12 is now available
The five main navigation tabs stay available while using Tools and calculators, including loading and error states. Home, Tables and Me links inside Tools open their corresponding native pages.
Google Play public listings in the US and Singapore show 1.1.12.
Released
H5
Desktop side panels can now be collapsed and reopened
The permanent desktop side panels could only stay open: the docked panels have no close button, and the phone entry buttons are hidden on desktop. Each rail now has an arrow handle on its table-facing edge — one click collapses that side to a thin strip, another restores it. The two sides are independent, the default is still expanded, and your choice is remembered in that browser. Collapsing only hides the panels, so an unsent chat message and the loaded hand history are still there when you reopen.
Each rail carries its own arrow handle, so the two sides collapse and reopen independently.
A collapsed rail leaves only a thin handle at the screen edge; the table keeps its exact size and layout.
Expanded stays the default, and the choice is remembered in that browser.
Collapsing keeps an unsent chat message and the loaded hand history intact.
Released
H5Backend
The tools page drops its duplicated title and link
The tools page used to print “Table Tools” twice and end with a “Back to the human tables” link that pointed where the bottom navigation’s Lobby tab already goes. The title now appears once, in the same style as Me and Lobby, and that link is hidden whenever the bottom navigation is present. The public tools page opened from an external link has no bottom navigation and is unchanged — it still offers one tap back to the tables. The same release also trims the internal daily operations report.
“Table Tools” is named once, using the same large-title style as Me and Lobby.
The duplicated “Back to the human tables” link at the foot of the page is hidden while the bottom navigation carries that destination.
The publicly shared tools page (no bottom navigation) keeps its way back to the tables, and the educational-purpose notice stays on every screen.
The internal daily operations report is trimmed to the few metrics that are actually read; table behaviour is unchanged.
Released
H5
Desktop tables gain permanent side panels
On wider desktop screens the empty space beside the table now permanently shows the table leaderboard, hand history and table chat, while the table itself keeps its existing size and layout. Phones and tablets keep their existing slide-in panels.
The table leaderboard sits on the left, with hand history and table chat on the right.
Panels no longer cover the table, and the buttons that opened them are hidden on desktop.
Permanent chat no longer steals keyboard focus from the betting controls, and an unsent message survives a window resize.
Released
H5
Keep the tab bar in the tools hub
Opening Tools from the app keeps the bottom navigation, with direct access to Home, Tables, Me, and the central play button.
The Tools tab stays highlighted and its central play action remains available.
Individual calculators keep their back action, while standalone tool links retain their focused layout.
Released
Marketing site
A clearer introduction to BluffKing
A green-felt hero, clearer copy, and direct links to play, review, analysis, and training introduce the product alongside real screenshots.
Shorter English and Chinese introductions retain practice-chip and dealing-verification details.
Real table, hand-review, and analysis screenshots are visible on desktop and mobile.
Released
Marketing site
Update the latency article with the September 9 results
Add the two-vCPU reproduction, the server p99 change from capping the analysis thread and batching relay writes, and the production re-measurement with the September 8 on-host method; explain that the public ladders bounded the test route, not the service.
Production server p99: HTTP 7.08→2.57ms, WebSocket all-recipients 12.20→4.48ms, server CPU 76.5%→50.0%.
Editor's note on the public capacity tables: a single client's route degraded with concurrent connections while the server stayed nearly idle.
Production count for the analysis cap: of the run's 834 hands, 485 completed background analysis and 349 were skipped with the queue full; the queue emptied less than 100 seconds after the window closed.
Re-align the read-cache article with the current server source: 12 of its 17 file:line citations relocated (including the session-deadline derivation that moved from auth/session.rs into read_cache.rs), and the header and session-deadline descriptions rewritten.
Released
Backend
Cap background analysis CPU and batch table message writes
The optional post-hand analysis thread no longer occupies a whole core, each table connection writes one action's messages in a single flush, and messages that need no rewriting skip a repeated parse, so live requests reach a processor sooner.
Background analysis defaults to at most a quarter of one CPU (configurable); excess work queues or is skipped without affecting hand persistence.
Per-connection session leases refresh at most once per second, removing a global lock and timer registration per message.
The performance-test operator lease now also covers HTTP per-address budgets for public-path verification.
Update the bilingual latency article with public-path capacity tests, separating rapid and slower play and reporting passing loads, threshold breaches and measurement limits.
Measure action p99 until all nine players receive the result, alongside maximum latency.
Preserve failures against the original stricter criteria and disclose the scope of the temporary fixture lease.
Released
Marketing siteBackend
Enable scoped public-path capacity tests
Add expiring connection-accounting leases for designated test accounts and source addresses to measure gameplay over the public path, retaining normal account and global connection protections.
Leases bind fixture accounts, source address, runtime version and expiry; missing or invalid leases retain ordinary admission.
Synchronize verified mobile version information on the website.
Released
Android
Android 1.1.11 is now available
This update improves connection recovery and keeps you signed in through temporary server interruptions. Profile names and avatars display more consistently, session details include squid-game penalty settlements, and room setup and post-hand pacing are refined. Games continue to use practice chips only.
Google Play public listings in the US and Singapore show 1.1.11.
Released
Marketing site
Latency article adds a public nine-player test
Add the post-publication public-path measurement: nine-player action p99 was 355.33ms, with connection and required persistence checks passing; maximum capacity remains outside the measured scope.
Retain the missed 200ms criterion and per-IP connection limit, distinguishing admission settings from performance guarantees.
Released
Marketing site
New article on latency optimization and concurrent play
Publish a bilingual engineering article covering latency diagnosis, database and scheduling improvements, and the results and limits of a 100-table, 900-player production test.
Distinguish measured action p99 from the still-unverified public-network capacity limit.
Released
Backend
Improve request scheduling in load verification
The performance tool now processes responses and table messages between batches of overdue requests, improving latency comparisons.
All scheduled requests, deadlines and persistence checks remain in place; latency still includes every server-side wait.
Released
Marketing site
Crypto Quant joins the studio project list
The studio page and footer now link to Crypto Quant, the quantitative project’s public monitoring dashboard.
Correct the server container's CPU scheduling hierarchy so its existing priority also applies when competing with other host workloads, reducing waits after requests arrive.
Compare complete receive-to-write latency at the same offered load while checking connection retention, successful requests and required persistence.
Preserve per-request timing breakdowns for application handling, waits outside the application timer and the remaining nine-player broadcast interval.
Released
Backend
Reduce CPU scheduling waits for the live server
Increase the live server container's CPU scheduling weight under contention to reduce waits between request arrival and application reads.
Measure the complete receive-to-write interval under production load alongside request success, connection retention and required persistence.
Keep table queries at least as current as the committed hand count, avoiding briefly stale values.
Released
Backend
Reduce repeated work in river equity analysis
Once the river board is fixed, equity analysis reuses hand ranks within each estimate to reduce background CPU work.
Preserve seeded draws, trial counts and analysis results.
Keep the existing computation path before the board is complete.
Released
Backend
Give live tables priority for CPU time
Move post-hand analysis to a dedicated background CPU worker and reduce repeated construction of shared hand data and live table views.
Lower the priority of optional analysis on Linux to give live gameplay more CPU time under contention.
Build shared hand fields once per hand while preserving private cards and atomic persistence.
Released
Backend
Reserve connection capacity for concurrent tables
Raise the table service file descriptor limit to leave room for WebSocket, HTTP and database connections when concurrent tables exceed the inherited connection allowance.
Increase the container file descriptor limit from the inherited 1024 to 65536.
Retain memory limits, connection admission and API rate limits.