Check up to 32 JSON argument values against a closed, declared scalar tool contract: types, required fields, code-point length and exact safe-integer bounds. Rounded fractions/underflow are not integers. Returns unknown fields and mismatches. No remote references, arbitrary patterns, tool execution or MCP conformance certification.

Split a long text into chunks for retrieval (RAG): split on paragraphs, then sentences, then words, each chunk within maxTokens (default 500) with the overlap you ask for (default 50 tokens). Each chunk has its index, text, start and end character offsets and an estimated token count (characters divided by 4, a rough rule for English). Up to 1 MB and 2,000 chunks.

Inspect one supplied x402-v2 exact-payment challenge against an explicit expected HTTPS resource, EVM CAIP-2 network, asset, payee and maximum price. Checks both URL declarations, structure and payment-policy fields only; no live probe, payment, facilitator verification or server-authenticity claim.

Validate exact UTF-8 SHA-256 hashes for up to 64 public or synthetic passages. No Unicode normalization, source authentication, or semantic validation.

Compute binary Brier score and count-weighted ECE for up to 256 declared held-out score/outcome pairs. Supports probability or percent scores, normalized to probabilities. Equal-width bins use exact rational boundaries: lower inclusive, upper exclusive, final bin includes 1. Empty bins report null means/gaps. No clinical validity claim.

Compare up to 64 total extracted spans from at most 16 labeled tools against exact code-point slices of supplied source text and declared UTF-8 hashes. Group identical source spans to expose text disagreements. No OCR/PDF execution, normalization, or completeness claim.

Exact rational expected-yield and cost arithmetic for up to 16 scenarios and 12 sequential conditional reject rates. All batch units incur unit cost; rejected units incur an additional penalty. Reports differences from the first scenario and undefined cost per good unit at zero yield. Declared assumptions, not a production forecast.

Check one structured public or synthetic low-stakes experiment plan for timing, comparator, outcome and analysis declarations. Returns deficiencies and an integrity receipt. No registration, trusted timestamp, prediction, power analysis or scientific validation.

Classify up to 100 declared measurement intervals against tolerances as pass, fail or hold. Integer millimetres; uncertainty is an absolute interval half-width, not a confidence level. No sensor verification or engineering approval.

Purchase one pinned Markdown section of The Maha Principle with edition identity, word/byte counts and SHA-256 integrity receipt. Choose a section and confirm the edition digest before paying. Not an EPUB/PDF bundle, medical advice or factual certification. Personal/internal use only; no redistribution, resale, training rights or copyright transfer. No stored-result recovery.

Find an exact 0/1 maximum-declared-value allocation for up to 64 evidence items and a 0–4096 integer token budget. Deterministic ID-based tie handling; items are indivisible. Optimality applies only to the additive caller-declared values, not relevance or answer quality.

Compare up to 64 proposed Base-USDC calls with a declared total budget, per-call cap, allowed payees and human-approval requirement. Returns policy mismatches and exact arithmetic totals. Does not authorize, sign or execute payments; caller approval labels are not verified.

Evaluate up to 64 explicitly fictional strategy/game choices against caller-declared integer utilities. Reports best declared actions, chosen utility, regret and observed-minus-declared error. No hindsight counterfactual outcome inference, financial advice, real-world intelligence assessment or strategy fitness claim.

Compare source coverage on exactly the same declared record cohort across two snapshots. Report per-record gained/lost sources, per-source count changes, and global coverage drift. Reject unknown sources or missing/duplicate records. Coverage does not establish source quality.

Check up to 80 declared floor areas and assignable ratios, returning exact assignable/support-area totals. Uses one declared area unit and basis-point ratios. No geometry, zoning, code compliance, rentable-area or structural certification.

Compare up to 80 expected and 80 observed document descriptors against one declared tenant and scope. Flags missing, unexpected and cross-tenant documents; incomplete scope stays undetermined. Public/synthetic metadata only. Not access-control enforcement, corpus completeness or legal verification.

Deterministic four-neighbor BFS on a grid of at most 24×24 cells. Returns a shortest obstacle-free route or an explicit unreachable/step-limit result. Toy discrete geometry only; no hardware, collision certification or physical feasibility claim.

Check presence of baseline, primary measure, acceptance threshold and bounded stopping rule; no efficacy or statistical validation.

Compare bounded room requirements with supplied topology; report unknown and contradictory declarations, not architectural approval.

Compare up to 24 public/synthetic feature summaries: mean differences, reference-standardized shifts and variance ratios. Explicitly flags zero reference variance. Counts and variance convention are declared, not verified; not a distribution test or clinical inference. No raw EEG, participant identifiers or full participant signals.