ENSEMBLE

Identity, plan and collection sizes for the authenticated account. Request billing for metered-work headroom and estimates, or charges for itemized usage with date filters and pagination. Reads never create billing records, credits or charges. An unprovisioned account has no persisted funds.

get_account

Return the profile represented by this request's authenticated credentials. The opaque ID is stable across token refresh, reconnect, and display-name changes. Unavailable optional display fields are omitted. This read does not initialize an account or billing record and accepts no account selector.

get_profile

The catalog record: wrapper, tax form, leverage, issuer, tags, country and the date range of its bars. `price_coverage` separates the catalog seed from stored bars. `tax_evidence` is unclassified when tax form and legal structure are null, which is not a 1099. `country` is set when the name states one country. `backfill` names your rule for the symbol, when you have one, so a history that predates the listing is explained. Whether a particular broker will accept a symbol is not tracked here. Request history or metrics for windowed total returns, data freshness and backfill provenance. This may refresh the market-data cache and return 202 while fetching; it never saves a research model or charges the account.

get_asset

The model's fact sheet: status, the run in progress, what it holds, what holds it, its `inputs_schema`, and `build_id` for the script it serves. Owners may use the ID, slug, symbol, or exact name; anyone may read a public model by its slug. Optional include sections add current source, paginated runs or selected history blocks. Date filters and full_history control the history window; metrics, fees and series describe that same window. get_run reads an exact retained execution. include=builds lists build attempts; build_id selects one historical or failed build with its source and full error.

get_model

The class source, its hash, and the models whose scripts name it.

get_optimizer

One run by id: status, inputs, summary, usage, the program it belongs to (with its `inputs_schema`), and its result on request. Owner or shared run only. The returned model and run IDs identify the retained model history.

get_run

Read-only status of feedback reports, newest activity first. A member sees reports they submitted. An allowlisted admin sees every report's authorization, queue, review link, pull request and closure, without the message. Does not claim, approve or close.

list_feedback

Your backfill rules, by target symbol.

list_backfills

The caller's models, newest first. Favorites, labels and status filter the complete library before pagination. status=trashed returns recoverable models and their recovery deadlines, without code or results.

list_models

Your optimizer classes, by name.

list_optimizers

Your registered receivers and recent delivery status, without signing secrets.

list_webhooks

Read-only status of one report: `fix_authorized`, `review_url`, queue, pull request and closure. A missing id and a report you cannot see are both not found. Does not claim, approve or close.

get_feedback

The contract a supplied script is validated against. Constants, entry point, the harness interface as a stub with every signature and docstring, the validation rules, and a skeleton. Describes the Python interface used by the validator and execution sandbox. Free; cacheable until the harness version changes.

get_runtime

Replace the class source and/or its description. A script edit is live for every future run of a model that names the class. Renaming the class is refused while stored scripts still reference the old name.

update_optimizer

Run the model's stored script (no generation). Pass intent=test for an isolated result, even with default inputs; intent=refresh publishes approved defaults on success. Omission keeps the following legacy behavior for existing integrations. An empty body runs the declared defaults: that run is the live path, its result becomes what the model serves and cascades to the models that hold it. With `inputs`, unset keys take their defaults (see `inputs_schema`) and `from_run_id` starts from another run's inputs with `inputs` overlaid; unknown keys are rejected with the accepted keys named. Any run away from the defaults is a snapshot to compare, share or promote. `inputs.start_date` (YYYY-MM-DD) starts the backtest later than the data allows; earlier prices still feed lookbacks. Metered: one base unit plus one unit per second of sandbox time, charged once per run. 500 manual runs per account per day (429). Answers 202 with the queued run, or 200 with the finished run when `wait=true`. 409 `run_conflict` while a default run or a rebuild owns the model.

run_model

Screen the catalog. Structured filters AND together with `q` (ticker or full text). `tags` matches any of its values, `tags_all` requires every one, `tags_none` excludes any of them; all accept only the closed vocabulary, which `include=facets` returns by dimension. `country` is the single country named by the fund (germany, japan, …); a broader region does not match. `price_coverage` separates a catalog seed from stored bars, and `tax_evidence` is unclassified when tax form and legal structure are null. `min_market_cap` is a size floor on market cap or fund AUM. Sort by `symbol`, `name`, `aum`, `expense_ratio`, `leverage` or `first_price_date`.

list_assets

Send a message to the ENSEMBLE team: a bug, a wrong result, a missing capability, or anything else. Returns a feedback id (`feedback_...` or legacy `f-...`) and a `report` with `fix_authorized`, `review_url` and the queue. 20 per hour. Reporting never authorizes implementation, even for admins. Plans and notes are reviewed and approved in the admin dashboard before a scheduled implementation batch. This does not approve or close a report.

submit_feedback

Extend one symbol's price history with a longer-lived stand-in. Use this when a wrapper has the treatment you want but not the history your backtest needs: OILK avoids a K-1 but starts in 2016, so backfilling it from USO lets a model hold OILK across a window that predates it. A model source lets you shape the stand-in's risk profile first (a lightly levered TLT model standing in for EDV, say). The source's returns are spliced in before the target's first bar, scaled so the two series join continuously. Setting a symbol that already has a rule replaces the rule.

set_backfill

Update an investment model's implementation or settings. Supply prompt or script to replace its implementation through a metered build; the prior working implementation stays live until success. Poll get_model until run is null. Send from_run_id alone to adopt tested inputs after a metered verification run; failure preserves existing defaults. Send trashed=false alone to restore a deleted model without enabling sharing, schedules or emails. No permanent deletion is performed. Rename, change the symbol, favorite, organize with private labels, toggle sharing, set the nightly refresh, or move the default view. `auto_update` and `notify_email` need a built model (409 `not_ready`). `start_date` and `start_rule` only move where the default view begins; nothing is recomputed. The backtest's start date is an execution input, separate from this view window.

update_model

Relabel a run or share it at /run/{run_id}.

update_run