zFiniazFinia
Research

Research

Current market briefs, protocol comparisons and defensible data investigations. Claims stay attached to sources; missing data stays missing.

Editorial

Latest research · 15 Sep 2026

MCP · observability · tool calls · agent operations

MCP tool-call health: observe agent runs without inventing protocol sessions

Current MCP is stateless, but operators still need to understand multi-call agent runs. Separate request, trace and application workflow identifiers; classify precondition failures distinctly; and measure recovery, loops and abandonment instead of raw error counts alone.

Read research →

Related

Field validation · 21 Sep 2026

10 repositories · public GitHub outcomes

What happened after two CrossCheck follow-up waves

One owner confirmed a pnpm-only policy and implemented the cleanup; four repository-specific PRs and three intent questions remain open; two contributions were avoided because the state was documented or already covered.

Inspect every public outcome →

Evidence boundary

Original benchmark · 21 Sep 2026

1,027 frozen repositories · raw JSON · replayable

CrossCheck's package-manager drift benchmark

The default rule produced 47 emissions across 947 untouched holdouts, and the cited configuration evidence was manually verified. A post-publication intent audit found deliberate multi-manager cases, so the benchmark does not treat all 47 emissions as confirmed errors.

Inspect the benchmark →

In this edition

Index · IN DEVELOPMENT

zFinia Machine Economy Index

A future citeable measure of machine-economy activity. No index value is published tonight. The intended sentence — “According to the zFinia Machine Economy Index…” — is valuable only if the methodology is real.

DataData collection in progress

zFinia Machine Economy Index

In development

Measurement is being established. No public headline number is published until coverage and classification allow.

Period
2026-09-01 → 2026-09-30
Coverage
No index value is published until inputs, weights, classification rules and revision policy are defensible.
Updated
8 September 2026

Do not write “according to the zFinia Machine Economy Index” until a value exists.

Methodology · Cite

  • Transaction activity

    How many machine-initiated payments were observed in the window.

  • Verified payment volume

    Independent third-party settlement, after quality filters.

  • Active paid services

    Services with a live challenge that a machine can complete.

  • Active purchasing agents

    Distinct independent payers, not owner tests.

  • Protocol adoption

    Which payment and discovery rails show repeat use.

  • Marketplace activity

    Machine-purchasable catalogues with evidence of use, not listing count.

  • Machine-priced resources

    Resources that advertise a price a wallet can pay exactly.

  • Network diversity

    Whether activity concentrates on one chain, card network, or facilitator.

  • Repeat-use activity

    Whether the same independent payers come back.

  • Ecosystem growth

    Net new verified services and rails, after de-duplication.

Full index rules →

Research-grade explainers

Feature

x402 vs MPP

Two HTTP 402 families. Different payloads, facilitators, and settlement stories.

Read comparison →