Shuffle trade PnL order to test path significance. Null hypothesis: the observed Sharpe / max-drawdown is no better than a random ordering of

Nelson-Siegel zero-rate curve. Args: tau: Maturity or maturities in years. beta0: Level factor; the asymptotic long rate.

Fold an option-type string to ``"call"`` or ``"put"``. Public because callers that *store* an option type must fold it with the same rule the

End-to-end: load file, detect format, parse. Args: path: File path. Returns: (format_name, records). Falls back to generic i

Reconcile broker truth against the durable last-known state. Pulls the broker's positions, balance, and open orders through the injected READ

Decide the next safe read-only recovery step for a rejected draft. Returns ``search_symbol`` when instrument identity is unresolved and

Scrub credential-shaped values from a free-form string. Used for plain-text tool results, shell output, error messages and log lines where :f

Return the first *available* loader instance for *market*. Walks the fallback chain and returns the first loader whose ``is_available()`` ret

Call ``fn`` with a bounded retry budget on declared transient errors. Between attempts sleeps ``min(backoff[attempt], remaining_budget)`` so a

Drive a multi-market backtest from a ShadowProfile. Markets are backtested per settlement currency: the composite engine refuses a mixed-curr

Validate a user-supplied broker/export file path. Args: p: User-supplied path. `~` expansion is supported. Returns: Absolute

Return a list of {symbol, market, rule_id, reason} matches. Args: profile: ShadowProfile holding the rules. target_date: ISO "YYY

Full-text search with FTS5 MATCH. Args: query: User search query. max_results: Maximum results to return. Re

Split content into chunks within max_len, preferring line breaks. Args: content: The text content to split. max_len: Maximum leng

Build the weighted average cost of capital. ``Re`` is CAPM plus two optional additive premiums; ``Rd`` is the pretax cost of debt shielded by

Run read-mode against a single mocked response and return the envelope. Args: paper_ids: Ids to request. response: Mocked ``throt

Build a persistent OAuth token store rooted at ``cache_dir``. The directory is created (with parents) and locked down to ``0700`` so only the

SLSQP weights for the penalized objective; updates turnover state.

Helper to create a memory file with extended frontmatter.

Build the DZ BANK original / RESTATEMENT / NEW HOLDINGS chain.

Install a fake ``subprocess.run`` and return (calls, original_run). The pip call is identified by the ``-m pip`` marker, the verification call

Import agent/mcp_server.py without executing main(). Returns: The imported mcp_server module. Raises: ImportError: If the mo

Return the registered MCP tools keyed by name (public async API).

Pre-process tool calls: handle compact, filter duplicates, batch execute. Args: tool_calls: Raw tool calls from LLM response.

Compute correlation matrix for multiple price series. Args: price_series: Mapping of asset code -> DataFrame with a ``close`` column.

Get close price with fallback. Uses array fast-path when available.

Send a raw frame to one connection, cleaning up on ConnectionClosed.

Sanitize skill name to a safe directory slug.

Score one text snippet in [-1, 1]. Returns a dict with ``score`` (float), ``positive`` (int), and ``negative`` (int).

Record what methodology produced this run, beside its trace. Answers "under what system prompt, which skills, and which tool set was

Score a firm's distress risk with one of the three Altman variants. Every argument is keyword-only, deliberately. Six interchangeable float p

Cost of borrowing the short leg for one period. Only negative weights are charged: a long position borrows nothing. The charge is the absolut

Pure core: is ``trigger`` due at ``now_ms``. Deterministic — reads no clock; all inputs are arguments. * INTERVAL: due when ``(now_ms - epoc

Publish a structured progress event from a tool. Silently no-ops when called outside an active tool context (e.g. during unit tests that invo

Evaluate a single set of metrics and return the decay signal. Uses the WORST signal across all provided metrics. For factors: uses ic

Detect peaks and valleys in a price series. Args: close: Closing price series. window: Half-window size; effective window is 2*wi

Fit a Nelson-Siegel or Svensson curve to observed zero rates. The betas are solved exactly by least squares at every candidate decay paramete

Return a goal by id.

Return a loader *class* for *source*, falling back if unavailable. Args: source: Requested data source name. Returns: A Data

Read all messages for a session. Args: session_id: Session ID. limit: Maximum number of messages to return.

Load QVeris config with environment overrides applied.

Strip and validate a required text field. Args: value: User supplied text. field_name: Field name for the error message. Ret

Pair buys and sells per symbol using FIFO to compute per-roundtrip PnL. Args: df: Standardized DataFrame (datetime-sorted). Returns:

Generate a driftless random walk (a unit-root, non-stationary series). Args: n: Number of observations. seed: Seed for the random

Run ONE fail-closed tick in fixed order; return the outcome dict. Ordering (each step short-circuits the tick on failure): 1. **Halt

Strip operator-only keys from API-caller-supplied session overrides. ``mcpServers`` / ``mcp_servers`` define subprocess ``command``/``args``/``en

Send a message to a session and trigger execution. Args: session_id: Session ID. content: Message content.

``sign(df) * |df|**p`` — preserves sign; never produces complex output.

Apply square-root market impact, the best empirically supported form. Impact is ``eta * volatility * sqrt(volume_traded / adv)``. Marginal impact

Turn a raw characteristic into a cross-sectional exposure. Winsorises both tails, then subtracts a cap-weighted mean and divides by an equal-

Single source of truth for where swarm runs are persisted. The swarm store (mcp_server) and the run-dir sandbox allow-list (src.tools.path_ut

Deliver every briefing whose run has reached a terminal state. Idempotent and restart-safe: the outbox row is the only source of trut

Compute both terminal-value estimates and cross-validate them. Args: final_year_fcff: ``FCFF_n``, the last explicit forecast year's free

Return a JSON-serializable profile snapshot.

Build a schedulable job from this playbook. The result is ready for ``ScheduledResearchJobStore.upsert`` and mirrors what ``POST /sch

Rolling sample covariance per column, min_periods=n.

Validate a URL is safe to fetch: scheme, hostname, and resolved IPs. ``allow_loopback`` is intentionally narrow: it only permits literal loop

Guard the model-validation state machine. The only hard rule: a model cannot be marked ``APPROVED`` unless it is already ``VALIDATED`` (re-ap

Walk an on-disk ledger end to end, returning where (if anywhere) it broke. Args: path: Ledger JSONL file path. A missing file verifies as

Split backtest into sequential windows, check consistency. Each window is evaluated independently (returns normalised to window start). Args

Split a distributable amount through the four European waterfall tiers. The tiers, paid strictly in order and only from what remains: 1. **R

Build an AgentLoop with a real (but empty) registry and a stub LLM.

Return the ``api_server`` module for late-access attribute reads. Tests monkeypatch ``ENV_PATH``, ``ENV_EXAMPLE_PATH``, ``_baostock_supported``

Return whether *code* is a symbol this loader handles. Covers US/HK/India/Korea/Canada/Vietnam equities plus Yahoo's own futures (``GC=F``) a

An AgentLoop shell built without running __init__ (we only need the helper).

TestClient whose peer host ('testclient') is treated as loopback.

Build an AgentConfig from a plain server-name → config-dict map.

Build a list of fake MCPRemoteToolSpec for testing.

Point PyPI + install-kind detection at deterministic fakes.

Parse '1-10' / '5' / '1,3,5-8' into zero-based indices.

Build a minimal ShadowProfile for scanner tests.