Every build stays at its own address, forever. The current one is also at the root.
v0.3.0current2026-09-10Desire first: three capabilities, one of them proven live — v0.2.0 was a connection manual on a shop window: six steps of work before any reason to do them, and three kits picked from what the bridge can do rather than from what anyone wants. Its first kit could not even run — it reviewed overnight positions on a brand-new demo account that has none. Rebuilt around what the trader does not have today: attention that does not run out, a second opinion at the moment of the click, and an idea turned into a bot. Every claim leads with the outcome; the jargon is gone; and the third kit proves itself on the page with a real sweep instead of a screenshot.
v0.5.0rejected2026-09-10v0.3.0's selling, kept whole — plus the quiz — Two attempts at a merge failed in opposite directions: the first buried the selling under more words, the second cut the selling out with them and left a skeleton of labels. This one takes v0.3.0 — the reader's pain named in second person, the tapes showing what you type and what comes back, the 'what you get that you don't have now' panels, the live sweep — and changes as little as possible: the three-question quiz from the club concept becomes the primary call to action everywhere, each kit gets the reader's own question above it, the unmeasured 'free Claude plan is enough' claim is gone, and a handful of sentences lost their throat-clearing. The dark design stays; it was never the problem.
v0.4.0reference2026-09-10Club concept, brought in whole — Ivan's concept built elsewhere, copied verbatim for comparison — only a noindex tag and the version chip added. Light paper palette, 900-weight type, a working three-step quiz that routes to one of three kits, and kit headlines written in the reader's own voice: 'Can you build this for me?'. Its honesty discipline is better than v0.3.0's — every claim is marked as a proposal and it refuses to promise free-plan compatibility. What it does not have is evidence: nothing on the page computes.
v0.2.0previous2026-09-10Club: library, assisted setup, broker-funded access — The product stops being a tool and becomes a club. Library of kits for the visitor's own Claude, assisted first launch, access funded by the broker rather than sold. Three kits picked from the documented cTrader MCP operation map instead of from imagination: morning account review, risk on live positions, and indicators plus alerts on the desktop bridge. Levels follow technical depth, so the lock is honest — the upper kit is gated because it needs our help, not because we are being greedy.
v0.1.0previous2026-09-10Backtester as a hosted MCP — One line connects the visitor's Claude to our own MCP server with five tools: parse_idea, backtest, sweep, position_size, export_cbot. The engine and the honesty are right — the live sweep on the landing page is a real call, and it shows 54 of 60 configurations profitable on the fitted window against 6 on both. What is wrong is the position: a terminal command is not a first touch for cold paid traffic, and on the free Claude plan the single custom-connector slot has to go to the broker's bridge, not to us. Kept: this becomes the upper tier for Pro users.