Loading ToolBench
Model Context Protocol (MCP) server for AETRE epistemic triage, Kingman queue governance, and VOI routing
AETRE defines 18 tools with complete input schemas and descriptions. Naming follows verb_noun convention (triage_proposal, evaluate_queue_congestion, optimize_reviewer_assignment). Descriptions are domain-specific and substantive (avg ~150 chars), explaining the mathematical/queueing concepts. However, output schemas are NOT documented, the source shows input schemas only. Parameter descriptions are present but often technical/jargon-heavy without actionable constraints (e.g., 'Coefficient of variation' lacks range guidance). No error handling guidance visible. Tools are READ_ONLY except run_shadow_pilot (WRITE), but no confirmation/dry-run patterns for destructive operations. Composition is strong: tools chain logically (triage → VOI boundary → sequential stopping). Security: no secrets in params (good). Overall: solid domain-specific definitions hampered by missing output schemas and lack of error recovery guidance.
Analyze submitter congestion dynamics and anti-spam staking elasticity under capacity constraints
Run venture capital and equity dealflow triage benchmark comparing allocation strategies
Calculate the value-of-information boundary for sequential review stopping decisions
Compute randomized exploration audit estimator for discovering novel proposals in a pool
Evaluate Kingman heavy-traffic queueing dynamics and governor actions for review queue management
Evaluate dynamic Bayesian sequential review stopping rule with confidence thresholds
Get a specific AETRE prompt template with argument substitution
Output schemas not documented. Source shows input schemas only; return types for all 18 tools are not visible. LLMs cannot plan downstream tool calls or extract required fields without knowing response structure.
Parameter descriptions lack actionable constraints. E.g., 'Coefficient of variation for arrival process' does not specify valid range (0.0 - 1.0? 0.0 - ∞?). LLMs cannot validate inputs or self-correct without explicit bounds.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
List all available AETRE prompt templates
List resource URI templates for dynamic resource access
List all available AETRE resources (datasets, catalogs, reference materials)
List all available AETRE MCP tools with their schemas and descriptions
Optimize reviewer-to-proposal assignment using congestion-aware matching under capacity constraints
Read a specific AETRE resource by URI
Run held-out review allocation backtest comparing 8 triage policies
Run Level 4 prospective shadow-mode pilot, prediction freeze, or randomized 3-arm trial
Evaluate triage routing on domain benchmark datasets or custom JSON files
Evaluate a proposal using multi-attribute Bayesian epistemic triage with value-of-information routing
Evaluate frozen test-split predictions against labels and baseline policies
No error handling guidance. Tools return READ_ONLY or WRITE but no recovery hints (e.g., 'If threshold is invalid, try values 0.0 - 1.0'). Agents cannot self-correct on failure.
run_shadow_pilot (WRITE) lacks confirmation/dry-run pattern. Destructive operations should support preview before execution to prevent accidental data loss.
Domain-specific jargon in descriptions ('Kingman heavy-traffic queueing dynamics', 'multi-attribute Bayesian epistemic triage') may confuse LLMs unfamiliar with operations research. Descriptions should include plain-English summaries before technical details.