The Registry, folders and run comparison

The Registry, folders and run comparison

The Registry is your lab journal: every configuration, every result, every strategy version. Its history feeds DSR — the honest penalty for the number of attempts (details).

The runs table

Columns (click to sort): Run · Strategy · Status · P&L % · Sharpe · Max DD · Win rate · Trades · Created · Period. In “Run”: ★ favourite, the name (repeat runs get v2, v3…), a rename pencil, an AI badge; coin and type on the second line. Filters: status tabs, a Favourites toggle, search by run/strategy/coin. While runs are active the list auto-refreshes every 5 seconds.

Row actions: folder (move) · rocket (promote: save as strategy / create paper bot / strategy for a live bot) · restart (same config) · cancel · delete.

Deleting removes a run’s trades and charts, but the trial stays in the DSR counter — the attempt history cannot be laundered.

Folders

• Inbox — everything unfiled;
• your folders — name, colour, icon, description (shown as the Registry subtitle when filtered — write the series’ hypothesis there);
• auto-folders — a saved strategy’s run lands in “📊 strategy name”, a one-off run in a date folder;
• Archive and Junk — system folders; their contents are hidden from the main list.

Move by dragging a row onto a sidebar folder, via the folder button in the row, or work inside a folder filter. The researcher’s pattern: folder = idea, hypothesis in the description, one change per run, names that say what changed (“14d”, “60d”, “+price filter”), ★ for candidates, junk to Junk.

Comparing runs

Tick 2+ finished runs → “Compare N runs”. Four blocks:

1. Overlaid equity curves — Normalize (all to 100 — mandatory with different capital) and Linear / Date-aligned (shape vs a common time axis);
2. Delta table — P&L %, $, Sharpe, Sortino, Max DD, PF, win rate, trades; the best value per row is highlighted, deltas vs the first run below each number;
3. Strategy parameters — differing rows are lifted to the top and highlighted: the key to “change one thing at a time”;
4. Correlation matrix — r→1 curves move together, r→0 independent; for a bot portfolio pick profitable strategies with low mutual correlation.

Backtest vs paper and live bots

A bot’s page has a “Backtest on my dates” button — it runs the bot’s strategy over its own period; when done, a comparison opens: a metric table and two curves (backtest vs bot) on a common axis. Don’t expect a perfect match: the bot slightly below is normal; backtest positive while the bot steadily loses — the strategy lives on idealised fills (tighten spread filters); far fewer bot trades — head-to-head metric comparison isn’t valid.