It trades. You watch.

Most bots have one opinion and a lot of confidence. Kestrel has a chart reader that has to show its working, a gate that says no far more often than yes, an exit that has to cover its costs before it locks anything, and your rules standing in front of every order like a bouncer with a clipboard.

Here is what runs today, on paper. All day, as setups appear, Kestrel's chart reader reads the strongest markets it watches and writes a call: long or short, the horizon, how confident it is, the entry, the stop, the target, and the one price that would prove it wrong. It writes its reason in five languages, and you can read every one on the Eye.

Separately, we are training our own market model from scratch on years of exchange data. Its first run has ended, a new run trains from zero, and no version has passed its tests yet. When one does, it joins on paper beside the reader. Not before, and not quietly.

This is the account the bot trades right now, read straight from the paper desk. Real market prices, pretend money, and nothing on this page is picked out or tidied away.

The figure is the cash plus every open position at a fresh price, or at what it cost where there is no fresh price. A loss is printed in the same size and the same ink as a gain, because hiding the bad days is how trading sites earn their reputation.

Beside it sits Kestrel's own record: every paper call it has made, which coin, which side, which lane, what the round trip cost and what was left after it. A past result says nothing about the next one, and we will never pretend it does.

Kestrel keeps one live book: the paper account whose result is the headline. Every other idea we have for doing better runs beside it as a lane, on its own ledger, taking the same calls with exactly one thing changed: a different exit, a maker entry, only one side, one horizon, one exchange.

A lane has to earn promotion. It needs enough trades to mean something, then it has to beat the live book after real costs, and then it waits for the weekly review on Sunday, where at most one rule changes per book per week. Lanes that fail are demoted in public, with their numbers. Today's lanes, and how far each has got, are on the Eye.

Every exchange Kestrel trades on paper also runs as its own book: the same calls, on that exchange's own prices, fees and quote, including Canadian dollar books. They are tagged PROVING and never count in a headline. Each book's trades, wins and losses are on the Eye.

And there will be one lane nobody is allowed to touch: the control. When the rules are frozen, it starts running them as they were, so every change can be measured against something that did not change. If a "fix" cannot beat the control, it was not a fix.

Here is the sentence most trading sites will never print: no Kestrel lane has yet proven that it beats its real costs over enough trades to call it an edge. Some lanes are up, some are down, and none has passed the bar. We are telling you because you would find out from the record anyway.

The bar is not "it made money last week". It is: after every fee, spread, slippage and funding payment, on trades it never trained on, in rising AND falling markets, over enough independent trades that luck is not the explanation. That is the same test for every lane, every exit change and every model version.

What we have instead is the honest part: a record of every paper call, every loss printed as plainly as every win, every failed idea demoted in public, and a system built so that only what passes gets promoted. Today's numbers are on the record page, straight from the record.

Paper trading exists for exactly this. Better to find out what does not work with pretend money, in public, than with yours.

Each package is one payment. It includes free months of a plan, and those months start on go-live day, the day real-money trading opens, not the day you pay.

Prices are set for your country and shown in Canadian dollars, plus tax. After the free months, most packages keep a discount on their plan for life; the entry package keeps nothing after its months, on purpose.

A pre-order buys months of access to software at a price fixed in advance. It is not an investment, not a security and not a share of anything.

Some of this is still a drawing. We would rather show you the dotted lines than let you find them, so here is what is promised and not shipped.

Each one moves to the changelog the day it ships, and the roadmap shows where it stands now, live from the build tracker. Dates are our best estimate, and if one moves we will say so.

The renderings below are the coming platform as it is drawn today. Nothing loads until you press one, and none of them is a promise of a date.

Kestrel learns on members' machines. Members lend their idle machines to Constellation; every job runs on two of them and counts only when the answers agree. The map below is that network, live, each dot a member's machine shown near its closest large city.

Then there is the noisy part: a monthly draw, a programme for creators, and a registry of the people who backed Atheron first. Bring opinions. The anchors read the chat.

Our accounts on Discord, Telegram, X and Bluesky are the same room with different doors. Same people, same bad jokes about bad trades, in whichever app you already have open.