honest_variance

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations provide readOnlyHint=true, which the description aligns with—variance computation is read-only. Beyond that, the description adds substantial behavioral context: statistical assumptions (unconfoundedness, overlap/positivity, honesty with separate subsamples), pre-conditions (data requirements), failure modes with concrete recovery actions, and a typical minimum N. This goes well beyond the annotations and gives an agent actionable expectations about when the tool will malfunction and how to respond.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Is the description appropriately sized, front-loaded, and free of redundancy?

The description packs a large amount of information into one dense block using labeled segments (Validation, Assumptions, Pre-conditions, Failure modes, Alternatives, Typical minimum N), and the main purpose is front-loaded. However, it is a single unbroken wall of text with no paragraph breaks, and the 'Validation: validated evidence tier' sentence is somewhat cryptic without elaboration. It is economical but would benefit from structural formatting for an agent to parse efficiently.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool's statistical complexity and the presence of a documented output schema, the description is remarkably complete: it covers the validation evidence tier, the three key identifying assumptions, pre-conditions, failure modes with recommended remedies, alternative estimators, and a typical minimum sample size. An agent has enough context to decide whether to call this tool, anticipate failure, and plan the next step without extra round-trips.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the baseline is 3 with no penalty. The description adds only indirect parameter context—the pre-condition about needing covariates/treatment/outcome maps loosely to X and forest, and the failure-mode advice hints at n_splits/tree count—but it does not enrich any specific parameter beyond what the schema already documents. The schema carries the full parameter-semantics burden, so baseline 3 is appropriate.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific verb+resource: 'Half-sample bootstrap variance of the ATE/GATE estimate', which identifies the computation clearly. It distinguishes itself from forest-growing tools like causal_forest and from forest_diagnostics by describing its specific estimation technique, and it names sp.dml, sp.auto_cate, and sp.tmle as alternatives. It loses a point because it doesn't explicitly contrast with variance/bootstrap siblings (e.g., bootstrap, jackknife_se, wild_cluster_bootstrap), so some differentiation is left to inference.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description gives implied usage guidance through pre-conditions ('Covariates, treatment, and outcome with enough data to grow an honest forest'), a minimum sample size (N: 1000), and failure-mode recovery actions ('Calibration test rejects -> Increase the sample / number of trees, or fall back to a doubly-robust learner'). However, the alternatives (sp.dml, sp.auto_cate, sp.tmle) are listed without conditions for choosing them, and there are no explicit when-not-to-use statements. Guidance is present but mostly implied.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.