What's new in Ping

Every release of Ping, newest first. To update, run your install line again from the install page; your keys, strategies and paper record are kept. Ping tells you in its banner when a newer version is out.

0.5.3 · October 26, 2026

• Coins moving together. The Watch page now shows which settlement indices moved together over the last fifteen minutes (correlation of five-second moves) and, for moves of two basis points or more, how often the other coin followed within the next five seconds, with the counts beside every number. It is display, not a signal, and the lab's result sits under the table: following the laggard coin ran 1,230 paper trades at about minus 2.3 cents a contract, and the Coinbase to Kalshi lead-lag lost as well, because Kalshi reprices within about 250 milliseconds. The samples are the index values your copy already receives, kept for half an hour on your own machine.

0.5.2 · October 26, 2026

• The pass mark, written before the first run. A strategy now cannot run on paper or be backtested until you write down what would count as passing: minimum trades, cents a contract after fee, whether both halves must be positive, a cap on one day's share of the profit, a loss-rate cap, and a date by which it is judged. Ping stores the mark with a timestamp and shows a Verdict (pass, fail, not yet) computed only from fills made after the mark, so a bar can never be fitted to a result that already exists. The lab's own bar is filled in by default (60 trades, +5¢, both halves positive, no day over 40%). You can change the mark later; every change is kept and shown, including whether fills already existed when you moved it. Strategies already running keep running; they ask for a mark the next time you start them. When live orders arrive, no verdict will mean no "Run live". This is the lab's method made the product's workflow.
• "The lab tested this shape." When a strategy's rule kind and markets match a shape the lab has graded, the editor and the strategy page show the lab's public result for it: runs, trades, cents a contract, thirds, verdict, and where to read more. The data ships with Ping as tested.json, built from the public all-time grade and the public real-money table only; it carries no live-lane settings and no maker parameters, and it is refreshed with each release. It is information, not a block: you can still run the strategy, and the pass mark records that you did so knowingly.

0.5.1 · October 26, 2026

• Your journal. A new Journal page pulls your own fills and settlements from Kalshi with your key (reads only; Ping still has no code that can place or cancel) and scores them the way the lab scores itself: cents a contract after fee, win rate by entry price, first half against second half, the best day's share of profit, taker share, per market family, and the result against simply holding every buy to settlement. Every position is graded from the exchange's recorded result, never from a log, and fees are Kalshi's own charge on each fill. One row per market and side; CSV export. It is the honest answer to "do I have an edge trading by hand?" Kalshi's live endpoint serves recent fills; older ones behind the exchange's historical cutoff are not included, and a position whose opening fills sit behind that cutoff is marked partial and left out of the scores. One thing we learned building it: when you hold both yes and no in the same market, Kalshi pays the matched pairs out at once and its settlement record never shows that cash, so "revenue minus cost" on the exchange's own page understates those markets. The journal works from the fills, which catch it.
• Pages on every long table. Paper orders, fills, journal positions and backtest orders now come 25 to a page with a rows-per-page choice, so a long record stays readable.

0.5.0 · October 26, 2026

• Watch mode, for trading by hand. A new Watch page shows what the bot sees in the last minutes of each market you choose, and places nothing: the line, the live settlement index, the projected settlement, the seconds left, the side the arithmetic favours, the ask, and the edge after Kalshi's taker fee. Beside the edge sits a win rate for that exact situation, measured from history, with its sample size: Ping snapshots every watched market at 60, 40, 30, 20 and 10 seconds before the close and grades each snapshot when the market settles, so the table is built from windows your own copy saw. Until a cell has twenty of its own, the lab's public measurement stands in for the one band it covers (close calls: right 98 of 98 at 10 s, 93 of 95 at 30 s, 88 of 93 at 40 s) and is marked lab. A cell with no measurement says so, and the row says the market's price is the best estimate we have.
• Your thresholds, not the lab's. You choose the markets, how early they appear, the edge that turns a row green (with a short tone), and the ask above which to ignore. The lab's live window, distance and price cap are not in Ping.
• Two findings stated on the page. Where nothing has been measured, the market's price is the best estimate we have: every model the lab tested against Kalshi's price on public information lost to the market. And the last minute edge lives in a few seconds; in the lab's speed study about 83% of offers were still there 650 ms later and 71% after two seconds, and the ones that vanished skewed to the winning side. A person clicking is at the slow end, so the page points to the slower markets where a human can act.
• Settings saved from the interface are now stored in their cleaned form (series tickers upper-cased and filtered).

0.4.1 · October 26, 2026

• Backtests, with honest fills. Every strategy page can now ask the same question of Kalshi's free market history that paper trading asks of the present: at each minute's end inside the rule's window, would this order have rested, and would a real trade have printed through its price before it expired? Decisions come from one-minute candles, fills from the public trade tape, settlement from the recorded result, fees from the series' real maker fee. No key is needed. Pick 3 to 30 days, watch it run, and read the report the way the lab's bars read: fills, win rate, cents a contract after fee, first half against second half, the best single day's share, and the size of the lie a touch-counting backtest would have told you. Everything fetched is cached, so a second run is quick.
• What a backtest cannot see. These markets now quote in sub-cent steps, so the whole-cent spread is one tick at almost every minute's end. Live, Ping rests one tick above the bid when it catches a wider gap between ticks; a minute candle cannot show those gaps, so the backtest joins the queue at the bid instead and counts those orders separately. The fill test is unchanged. A backtest is a best case of a best case: no queue position, no latency, minute resolution. The rule that reads the live settlement index cannot be backtested and says so.
• Your verified record, per strategy. Under "Your paper result": settled fills, net, first half against second half, and the best day's share of profit, scored only on settled fills the way the lab's bars are, with a CSV export of that strategy's orders.

0.4.0 · October 26, 2026

• Alerts to your own Telegram or mailbox on fills, settlements, a strategy starting or stopping, errors, and optionally every resting order. Set up in Settings; the bot token and SMTP password go in your key store; nothing passes through us. A per-minute cap keeps a bad night from becoming a storm.
• Headless mode. ping --headless runs with no browser and no tray. The one-line installer on a Linux box without a desktop writes a systemd user service and starts it, so Ping runs on a $5 server or a Raspberry Pi, with your keys still on that machine. Reach it over an SSH tunnel.
• The settlement index, live. Ping now listens to Kalshi's CF Benchmarks index feed on the same connection, so rules can see the number a crypto market actually settles on, not just its price. Each market's line (strike) is kept too.
• New rule kind: settlement projection. Projects the final-minute average from what has already happened and buys the side it favours when the projection sits far enough from the line. You set the window, the distance and the price cap; the built-in ships with neutral defaults and the lab's public measurement, not the lab's live settings. Your own rules can read the index and the line through the context (ctx.index_latest, ctx.index_window, ctx.strike, ctx.projection()).

0.3.10 · October 25, 2026

• New strategy has a front door: start from a built-in, open a blank strategy with the template rule ready to edit, or open the editor with the AI box. Modify on an existing strategy is unchanged.
• The log uses plain market names and your own time zone, and connection problems are described in words ("the connection timed out") instead of error codes.
• When a side has no offer at all, the panel says so ("no one is offering yes; the market looks decided for no") instead of "yes asks 0".

0.3.9 · October 25, 2026

• Markets in plain words. The editor has a markets picker: search by name ("bitcoin", "temperature", "NFL") or browse a category, and pick from a list that reads "Bitcoin price up down · every 15 minutes" rather than KXBTC15M. Repeating markets are listed first. The "What Ping sees" panel and the strategy cards use the same plain names, with the ticker kept small for anyone who wants it.

0.3.8 · October 25, 2026

• The AI helper can write a rule. Describe a strategy; if a built-in rule fits, it fills the settings as before. If not, it writes the rule's code into the editor's code box, following the same contract as the template. The code is checked like anything typed by hand, you read it, and nothing runs until you save and press Run on paper.

0.3.7 · October 25, 2026

• Rules are edited inside Ping. The strategy editor shows the rule's code next to the settings. Built-ins are read only; Customise makes a copy that belongs to the strategy, you edit it in the box, and Save checks it (syntax, imports, a decide function) before anything is written. Run it on paper when you are happy.
• One Modify button on every strategy, and New strategy in the top corner. The separate rules card is gone.

0.3.6 · October 25, 2026

• Fixes the Create rule file button doing nothing right after an update: the browser kept the previous version's script. Scripts and styles now carry the version number so every update loads fresh.

0.3.5 · October 25, 2026

• Your own rules. Drop a short Python file in the rules folder and it becomes a rule kind any strategy can use. Start from the template or copy a built-in and edit the copy; built-ins stay untouched so their lab record keeps meaning something. Create and reload from the Strategies page; problems show in a banner. The editor and the AI helper know about your rules too.

0.3.4 · October 25, 2026

• Turning on a strategy that watches a new series now looks up its markets at once instead of at the next scheduled refresh, and the panel says "looking up open markets" while it does.

0.3.3 · October 25, 2026

• New built-in: The long shot. Rests a bid on any side priced 5 cents or less. Record: failed in the lab's study (38,565 paper buys, 2.8 cents a trade short on all twelve coins). Shipped so you can watch it rather than take our word.
• Settings links to this page. The update banner links to what changed.

0.3.2 · October 25, 2026

• Ping now knows when a newer version exists and says so in the banner, with a link here. Settings shows your version and the latest.

0.3.1 · October 25, 2026

• What Ping sees. A card on the dashboard and on each running strategy's page shows the market being watched, the price now, what the rule is waiting for, and when the market closes. Updated every few seconds.
• When a market closes without an order, the log now says why it was skipped, in the rule's own words.
• The banner shows whether live prices are streaming.
• Waiting lines explain one-tick spreads: when there is no room to rest a bid below the ask, it says so.

0.3.0 · October 24, 2026

• First member release. One-line install for Mac, Windows and Linux, with the download verified against a published checksum.
• Member key tied to your membership email, confirmed with the site once a day, with a seven-day grace period if the site cannot be reached.
• Keys fall back to an owner-only file if your keychain refuses them.
• Seven built-in strategies with the lab's record next to each; duplicate, edit, or write your own; optional AI helper with your own key.
• Paper engine with through-price fills and Kalshi's real maker fees. No order-sending code in the app.