Changelog
What we shipped
</and who asked for it>
TRKR is in beta, which means this list is the product roadmap running backwards. When a change comes from something a tester asked for, their name is on it — with their permission. If something here is wrong, missing, or annoying, the feedback button on every page is the fastest way to end up in a future entry.
</agents>Your AI assistant can test playbooks, leverage and trading costs
The AI connector now runs portfolios exactly as the playground does. It charges trading costs, can borrow, and has a new tool for playbooks, which reports whether the rules timed the market or just held a different mix. Slippage now defaults to 25bp of every trade, in the playground and for your assistant alike, so the two give the same answer for the same portfolio.
</engine>Interest rates in the simulated futures no longer drift toward zero
The engine now learns where interest rates tend to settle from 1990 onward instead of from 2006, a window in which rates sat near zero for about eight years. Before, rates ended the decade below today's level in almost every simulated future; in our tests they now do so in about seven futures out of ten. In practice cash earns more in the simulations, bonds earn less, and the engine's max-Sharpe portfolio moves away from bonds toward equities and gold. Datasets published before this change are not comparable on rates, cash or bonds.
</playground>Start any portfolio from a max-Sharpe mix, a benchmark or your own
The engine's max-Sharpe portfolios used to be listed among the assets, so you could put a portfolio inside a portfolio. They are starting points now. A "Start from" list, in Simple mode and in every playbook mix, fills the weights from either of them, the S&P 500 or 60/40 benchmarks, a playbook preset's mixes, or any strategy you have saved. They are still there to compare against, and the datasets and the AI connector now list only assets you can actually hold.
</leverage>Leverage, with what borrowing really costs
A portfolio, and each playbook mix, can now hold up to three times its equity. The borrowed part pays each future's own cash rate plus a spread (100bp a year by default), every rebalance resets the positions to target, and a margin call sells down when equity falls below 25% of what is held. The risk table says how many futures met a margin call, and how many were wiped out.
</playbooks>Playbooks can move to cash, stay put, and start where you choose
A rule, and the fallback when no rule fires, can now move to cash or keep whatever the playbook already holds, and you choose where it starts. That gives a playbook memory: start in cash, hold the base mix while VIX is below 20 and the storm shelter above 30, and in between keep whichever it last switched to — the new Volatility band preset. You also choose which mix the playbook is compared with.
</timing>Did your rules time the market, or just hold a different mix?
A playbook that shelters in bonds holds less stock than its base on average, so beating the base on drawdown can come from that alone. The rules panel now also compares your playbook with its own average mix — what your rules held on average, kept constant — through the very same futures. Beating that mix is what the switching added; whatever the mix gains over the base needed no rules at all.
</costs>Trading now costs something: slippage and turnover
Every rebalance and every switch between mixes now pays slippage: the bid/ask spread, market impact and commissions, 10bp of what is traded by default and yours to change. The risk table shows turnover, how many times a year trading replaces the whole portfolio, so a busy playbook's cost is visible next to its returns. The base you compare against can be rebalanced on its own schedule, and a saved strategy is named for what it holds — rename it whenever you like.
</playground>Compare any portfolio against a base, in every chart
Pick a base to compare against — the engine's max-Sharpe portfolios, the S&P 500 or a 60/40 — or, in a playbook, your base mix held on its own. The outcome fan, the final-value histogram and the risk table show it beside your portfolio through the very same futures, in teal, so you can see where the whole spread moves, not just the median. Risk & return is drawn as boxplots with the exact figures one click away, and the fan's bands are stronger and labelled. Presets, mode switches and loaded strategies now re-run on their own, so the charts always describe what is on screen.
</open>Explore the data without an account
The dataset pages and the playground are open to everyone. Read any asset, run a portfolio against every simulated future, and decide afterwards whether you want an account — which is now only for saving a strategy and downloading the dataset archive. The endpoints behind those pages gained rate limits at the same time, so an open door does not mean a slow one.
</agents>A public page for connecting your own AI
How to point Claude, ChatGPT or Antigravity at this data now lives on a page you can read without logging in, including the install command for each. The simulations stay here; the thinking happens in the assistant you already pay for.
</links>Shared links arrive as a card, not a grey URL
Every page carries a preview image and its own description, so a TRKR link posted in Slack, LinkedIn or a message renders as something worth clicking.
</nav>The methodology is one click from anywhere
How the engine builds a future — regimes, macro paths, and how it is scored against real history — was reachable only by direct link. It is in the top navigation now.
</chatgpt>ChatGPT can connect to TRKR
ChatGPT cannot send an API key to a data source, so the endpoint learned to sign you in instead. Add the connector, approve it here, and the tools appear — no key to copy.
</antigravity>So can Google's Antigravity
One command with your key, or none at all if you let it sign you in the way ChatGPT does.
</scope>An approved assistant reaches the simulations, and nothing else
Your saved strategies and your profile are out of reach for a connected app — enforced by the database itself rather than promised on a consent screen.
</keys>One agent key at a time, retired in the right order
Creating a key now retires the previous one first, so a key you have been told is dead cannot still be working somewhere.
</beta>The beta opened
The dataset explorer, the playground, accounts, and the feedback widget that decides what gets built next.
Tell us what to build next.
Every entry above started as someone using the thing and saying what was missing.