backtest_analyze
ChatGPTFacts that explain a backtest result: trades and profit grouped by opening rule, exit reason and side, entries blocked by risk limits, orders rejected by execution (for example for margin), and notes on conditions that need attention. Call it after every finished backtest, check quality first (guide_get backtest-quality), and describe rules by their conditions rather than their ids.
backtest_compare
ChatGPTPuts two backtests side by side with their full results and lists what differed: strategy version, settings and parameter values. Use strategy_get with compare_to_version_id for rule-level differences between versions.
backtest_get
ChatGPTGets a backtest's status and, once succeeded, its full results: return, drawdown, win rate, profit factor, Sharpe, trade counts, an equity summary and rejected orders by reason. Optionally waits for a running backtest.
backtest_options_get
ChatGPTReturns what a backtest can use: the symbols, timeframes, the longest date range per timeframe and the default balance, leverage and costs; defaults apply to every value backtest_run is not given. When the user's plan limits backtests, monthly_backtests gives this month's used, limit, remaining and reset time, and parameter_search.included says whether parameter searches are available to them. Call it before the first backtest_run in a conversation and before a request that may need more backtests than remain. If a request needs more than remain, or a parameter search is not included, say so and when the count resets; do not recommend upgrading.
backtest_run
ChatGPTBacktests a strategy version on historical bars and waits up to wait_seconds for the result. Give symbol, timeframe and either from_date/to_date or range_days; balance, leverage and costs default to backtest_options_get (all costs default to 0). Spread and slippage are in ticks of the symbol (backtest_options_get gives each symbol's tick size); commission is in USD per lot per side. The default leverage is 1, so strategies sized by risk on forex and metals usually need leverage (for example 100). Parameters override declared strategy parameters for this run only. If finished is false, poll backtest_get. Report results as historical backtests, not advice.
backtest_trade_list
ChatGPTLists a succeeded backtest's completed trades in opening order with side, prices, profit, holding bars, the rule that opened each trade and why it closed (rule:<id>, stop_loss, trailing_stop, take_profit or time_stop).
creator_public_get
ChatGPTGets a Trader AI creator's public profile and their published strategies.
guide_get
ChatGPTReturns one topic of the Trader AI guide. Strategy DSL: overview (read before writing or patching a strategy for the first time), expressions, rules, position, risk, diagnostics, examples (complete runnable strategies). Strategy research: data-sources (what to tell users about the price data), backtest-quality (check before judging a result), diagnosis-playbook (evidence-based improvement advice), research-workflow (from result to tested improvement).
indicator_search
ChatGPTFinds the indicators (platform macros such as rsi, macd, atr, bbands) and atomic operators a strategy can use, with inputs, argument ranges, outputs and warm-up. Search by keyword (e.g. 'moving average', 'volatility') or look up exact ids. Use displayName, not the id, when describing a strategy to the user.
leaderboard_list
ChatGPTLists published Trader AI strategies ranked by backtest performance, 20 per page. Use it to discover or compare public strategies. There is no keyword search: narrow results with markets and sort instead.
live_trading_get
ChatGPTGets one AI Trader on the leaderboard by its code: its rank, live performance and state (stopped, starting, catching_up, running, paused, stopping, faulted or lost), and its strategy card when the author published the version it runs.
live_trading_list
ChatGPTLists AI Traders: bots that trade one symbol and timeframe with one strategy on a real-money account. This is the website's AI Trader leaderboard, the same for every user: each comes with its rank and live performance (total and 30-day return, maximum drawdown, win rate, trades, risk, days running), highest total return first unless sorted otherwise. A trackRecord of new means under a week or under ten trades: give its figures as they are, without ranking it against others or judging its strategy. Paste web_url. public_strategy_list is another list: published strategies ranked by their backtests.
mt5_export
ChatGPTChecks that the strategy version behind a succeeded backtest can be exported as a MetaTrader 5 Expert Advisor (EA) and returns web_url, the backtest's page on the Trader AI website, where the signed-in user downloads the EA with its install guide. Nothing is downloaded or installed here. Give web_url exactly as returned. If MT5 export is not included in the user's plan, say so; do not recommend upgrading. Results are historical backtests, not investment advice.
optimization_get
ChatGPTGets a parameter search's status and, once succeeded, every set it evaluated with in-sample and out-of-sample results, rank, overfit flag and the lowest score among neighboring values. Optionally waits for a running search.
public_creator_get
ChatGPTGets a Trader AI creator's public profile and their published strategies.
public_strategy_get
ChatGPTGets a published Trader AI strategy: description, current published version, that version in plain words (rules, risk settings, parameters; never the DSL) and backtest performance (return, drawdown, win rate, Sharpe, profit factor, backtest settings and an equity curve summary). strategy_clone copies it as the user's own.
public_strategy_list
ChatGPTLists published Trader AI strategies ranked by backtest performance, 20 per page. Use it to discover or compare public strategies. There is no keyword search: narrow results with markets and sort instead. For AI Traders and their live results (the website's AI Trader leaderboard) use live_trading_list.
strategy_clone
ChatGPTCopies a published strategy as the user's own new strategy, which they can then backtest, change and export like any of theirs; the original and any AI Trader running it are not affected. Pass ai_trader_code to copy an AI Trader's strategy as it runs, its parameters included (possible when live_trading_get shows a strategy_card), or public_strategy_id for a published strategy. Then backtest the copy, on the AI Trader's symbol and timeframe when copied from one, and compare with its live results. Retrying the same call returns the same copy.
strategy_create
ChatGPTSaves a new strategy from either a DSL document (definition) or a template (template_id plus optional template_parameters). An invalid document returns valid=false with diagnostics and saves nothing. Retrying the same call returns the same strategy. Name the strategy in the user's own words. Lock the user's intent with intent: one entry per rule (by rule id) and per parameter (by name), the risk settings the strategy defines and the direction, in the language of the conversation. Write numbers that come from parameters as {name}, never as digits. For a template, write it from the template's definition in strategy_template_search. Missing or mismatched intent comes back as diagnostics, like a DSL error, and nothing is saved.
strategy_get
ChatGPTGets a strategy with its version history and one version's DSL document (the latest unless version_id is given). With compare_to_version_id it also lists what changed between the two versions. Read the latest version before strategy_update.
strategy_list
ChatGPTLists the user's strategies, most recently changed first, with their latest version and latest successful backtest.
strategy_optimize
ChatGPTTries a few values of at most two declared parameters on the settings of a finished baseline backtest, holding out the last 30% of its range as out of sample, and ranks the sets by out-of-sample return / drawdown. The baseline's own values always run too, and all sets together may not exceed ten; each set counts as one of the user's monthly backtests, and searches may not be available to the user (see backtest_options_get). Before calling, confirm with the user which parameters, which values and how many sets. Nothing is saved; do not resubmit on your own after a failure or a weak result. Waits up to wait_seconds; if finished is false, poll optimization_get. See guide_get research-workflow.
strategy_public_get
ChatGPTGets a published Trader AI strategy: description, current published version and backtest performance (return, drawdown, win rate, Sharpe, profit factor, backtest settings and an equity curve summary).
strategy_publish
ChatGPTPublishes one of the user's strategies. Anyone, signed in or not, can then see its name, the user's profile name, its public description and its backtest results; signed-in Trader AI users can also see its rules in plain words and copy it as their own, including from an AI Trader that runs the published version. Publishes the latest version with its latest succeeded backtest unless version_id or backtest_id is given. Each call publishes again and replaces the published version, so do not repeat a call that succeeded. Before calling, tell the user exactly what becomes public and copyable, and pass risk_acknowledged true only after they agree.
strategy_template_search
ChatGPTLists backtest-verified starting strategies. Use it first when the user asks for a recommendation or describes only a style (e.g. 'a trend strategy'): present two or three candidates in plain trading language with their caveats, then create the chosen one with strategy_create(template_id, template_parameters). definition is the DSL the template creates: write the intent for strategy_create from its rules, params, position_mgmt and risk, and do not show it to the user.
strategy_unpublish
ChatGPTTakes one of the user's published strategies down: nobody can see it in words or copy it any more, also not from an AI Trader running it. Copies already made stay their owners'. strategy_publish publishes it again.
strategy_update
ChatGPTSaves a new version of a strategy by applying changes to base_version_id, which must be the latest version. Each change sets or removes one named element (feature, param, macro, state), one rule by id, or one section (position_mgmt, risk, param_constraints, meta); unchanged parts stay exactly as they were. Use op replace with target definition only when the user asks for a rewrite. Errors return valid=false and save nothing; older versions and their backtests are kept. Lock the intent of the change with intent: changeNote (what changed and why, in the user's terms) plus the entries of the rules, parameters and risk settings you changed or added; unchanged entries carry over. If the conversation language differs from the strategy's, pass the complete intent. A change of intent alone, without a DSL change, is not saved.
strategy_validate
ChatGPTChecks a strategy DSL document without saving it. Fix every error diagnostic (each has a JSON pointer path and usually a suggestion) and validate again before strategy_create; do not show diagnostics to the user unless they ask. Valid documents report their parameters and warm-up bars.