Tools / Causal

attribute_change()

Key

Distribution-change attribution between two as-of windows (DoWhy-GCM): which upstream causes account for the move in a metric. Capability is partial — when the GCM method can't run, it falls back to a structured candidate-cause analysis over the approved graph, and says so.

attribute_change(metric, window_a, window_b)→ ATTRIBUTION

Callable over MCP as attribute_change(metric, window_a, window_b). REST mirror POST /v1/attribute Planned — alpha is MCP-only

Why your agent calls it

The retrospective twin of simulation: win rate moved — what caused it? Distribution-change attribution between two as-of windows, over the approved graph. And because the capability is honestly marked partial, the agent knows when it's getting the GCM answer versus the structured fallback.

What it returns

attribute_change() · What caused this metric to move?ILLUSTRATIVE

Call it from an agent

Over MCP, your agent invokes it on its own when the moment calls for it — just describe the situation. A Python SDK for headless runtimes is planned; this is what the call will look like.

agent · python sdkSDK · PLANNED — ALPHA IS MCP-ONLY

from cascadian import Cascadian

cx = Cascadian(api_key=os.environ["CASCADIAN_API_KEY"])

result = cx.attribute_change(
    metric="win_rate",
    window_a="2026-03-01/2026-04-15",
    window_b="2026-04-16/2026-05-31",
)

Parameters

Param	Type	Description
metricREQUIRED	string	The variable that moved.
window_aREQUIRED	string	Baseline as-of window, e.g. "2026-03-01/2026-04-15".
window_bREQUIRED	string	Comparison as-of window.

Example response

attribute_change()JSON · ILLUSTRATIVE

{
  "metric": "win_rate",
  "windows": ["2026-03-01/2026-04-15", "2026-04-16/2026-05-31"],
  "attribution": [
    { "cause": "discount_mix", "contribution_pp": -1.4 },
    { "cause": "seasonality", "contribution_pp": -0.5 }
  ],
  "capability_status": "partial — structured candidate-cause fallback"
}

Related — Causal

keypropose_nodes()Propose variables for the graph — status proposed.keypropose_edge()Propose a cause→effect edge, with evidence.keyget_graph()The governed graph, by version or as-of.keyvalidate_graph()Structural checks before anyone trusts an edge.keyidentify()Can this effect be estimated at all? The gate.keyestimate()A causal estimate that survives the skeptic checks.keysimulate_intervention()do(X=x) on the approved graph — the knock-on report.