Tools / Causal

estimate()

Key

Requires an identification_review_id — it will not run without one (invariant I3). Builds the dataset through the as-of repository (no future rows), hashes it, runs the estimator, then runs the deterministic skeptic checks: lookahead, post-treatment control, missing adjustment member, overlap, sample size, stale context, Granger-misuse, factor-mirage, regime-nonstationarity, multiple testing. Blockers abort with an explanation. Estimators: diff_means, ols_adjusted, logistic, ipw with bootstrap CIs; aipw/did/its are registered interface_only — never faked.

estimate(identification_review_id, estimator, as_of)→ ESTIMATE + SKEPTIC REPORT

Callable over MCP as estimate(identification_review_id, estimator, as_of). REST mirror POST /v1/estimate Planned — alpha is MCP-only

Why your agent calls it

Where the number comes from — and where bad numbers go to die. The dataset is built as-of (no future rows), hashed, estimated, then run through ten deterministic skeptic checks; blockers abort with an explanation instead of shipping a confident artifact. What survives is an effect with intervals you can defend in a post-mortem.

What it returns

estimate() · A causal estimate that survives the skeptic checks.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.estimate(
    identification_review_id="idr_0271",
    estimator="ols_adjusted",
    as_of="2026-06-11T17:00:00Z",
)

Parameters

Param	Type	Description
identification_review_idREQUIRED	string	The review from identify(). No review, no estimate.
estimator	enum	diff_means · ols_adjusted · logistic · ipw. Default chosen by the review.
as_of	timestamp	Build the dataset as of this moment. Default the review's as_of.

Example response

estimate()JSON · ILLUSTRATIVE

{
  "identification_review_id": "idr_0271",
  "estimator": "ols_adjusted",
  "effect": {
    "treatment": "discount", "outcome": "win_rate",
    "estimate_pp": -2.1, "ci95_pp": [-3.5, -0.7]
  },
  "dataset_hash": "sha256:91ce…",
  "skeptic_checks": {
    "lookahead": "pass", "post_treatment_control": "pass",
    "overlap": "pass", "n": "pass (n=1,412)",
    "regime_nonstationarity": "warn", "multiple_testing": "pass"
  },
  "capability_status": "implemented"
}

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.keysimulate_intervention()do(X=x) on the approved graph — the knock-on report.keyattribute_change()What caused this metric to move?