The source is part of the data, and the kind of index determines what a score may mean.
Machine Observer is now both a meta-index and an observatory of agent infrastructure. It therefore uses different evidence models for longitudinal data indexes, structural infrastructure maps and commercial markets rather than forcing every subject into one composite-score template.
Source confidence
Source confidence is a reproducible quality weight, not a probability that every claim is true.
30% methodology transparency
+ 25% direct observation
+ 20% machine readability
+ 15% historical depth
+ 10% independence
Non-negotiable rules
- Unknown values are stored as
null, never inferred for presentation. - Every publishable atomic claim carries a source ID, evidence URL and observation time.
- First-party sources receive no ownership bonus.
- Several indexes derived from the same chain, registry or query do not become independent votes because their interfaces differ.
- A payment quote is not paid fulfillment. Registration is not validation. Authentication is not reputation. Reputation is not delegated authority.
- A structural readiness score describes available mechanisms, not product safety, investment merit or moral trustworthiness.
- Commercial spend, sponsorship and publisher earnings cannot improve editorial evidence or neutral infrastructure rankings.
- Historical snapshots are append-only in normal operation; corrections create a new observation rather than silently rewriting prior evidence.
Not every useful measurement should become the same kind of number.
The family determines which comparisons are legitimate and when a headline composite should be withheld.
Measure change in observed phenomena
Examples include machine commerce, API prices and automatically discovered activity clusters. Candidate → Experimental → Core promotion depends on coverage, source diversity, freshness, stability and historical depth.
Compare what systems make possible
Agent Web, Agent Payments, Agent Trust and Agentic Social Media can use dimension-level readiness or coverage scores before ecosystem adoption is mature. These scores describe structure, not popularity, safety or overall quality.
Measure disclosed economic activity
Agent Advertising can measure bids, selections, spend, conversions and publisher yield. Commercial observations stay explicitly editorial:false and cannot leak into neutral rankings.
What each headline number is, and what it is not.
The homepage leads with numbers, so each one has to say precisely what it measures. These definitions are the contract.
| Indicator | Source | Kind | Formula | Known limitation |
|---|---|---|---|---|
| Machine Economy Index | Machine Observer index engine | Composite, authoritative | Mean of component metrics normalized against the first seven stored observations (baseline 100) | Requires seven snapshots before it produces a value at all. Below that it reports Collecting baseline rather than a number. |
| x402 Settled Value | Agent Economy dataset · daily | Cumulative level, authoritative | Reported directly by the source | Cumulative since the source began measuring. It is not a period total, and a flat reading may mean the source has not republished rather than that no value settled. |
| x402 Transactions | Agent Economy dataset · daily | Cumulative level, authoritative | Reported directly by the source | Cumulative across all tracked facilitators and chains. Counting conventions are the source’s, not Machine Observer’s. |
| Average Settled Transaction | Agent Economy dataset | Derived | cumulative settled USD ÷ cumulative transactions | A cumulative arithmetic mean over the entire measured history of the rail. It is not the median transaction, and it is not the value of a recent transaction. A few large early transfers move it permanently. |
| x402 Transactions / Day | Agent Economy dataset | Derived rate | (transactions at t − transactions at t−1) ÷ days between observations | Normalized by interval so unevenly spaced observations stay comparable. Intervals under about five hours are skipped rather than extrapolated into a daily figure. |
| Observed x402 Ask | Machine Observer price index · daily | Derived, advertised price | median of the lowest advertised price at each machine-payable endpoint sampled in the latest collection | This is an asking price, not a transaction price. It is read from each seller’s own payment challenge on an unpaid request. It is not a settlement price, a clearing price or a transacted price, and it reflects only the endpoints Machine Observer samples. |
| ERC-8004 Registered Agents | Agent Economy dataset · daily | Cumulative level, authoritative | Reported directly by the source | Counts registrations under one identity standard on tracked chains. It is not a census of AI agents in existence, and a registration is not evidence of an active agent. |
| Machine-Callable Services | Graded · 6-hour probes | Derived share | callable services ÷ observed services × 100 | Both numbers come from one prober’s view of discoverable services. Coverage of the wider ecosystem is unknown. |
| MCP Servers Updated · 24h | Official MCP Registry · daily | Period count, authoritative | Reported directly by the source | A 24-hour period count, not registry size. Bounded by how many registry pages the collector fetches. |
| AI Policy Instruments | Machine Observer AI Policy Index · daily | Level, first-party normalized | Count of normalized instruments in the index | Coverage grows as sources are added, so a rise can mean new law or new coverage. Regulatory activity rising is neither good nor bad news and is shown without sentiment colour. |
Why there is no clearing price
Machine Observer publishes no x402 clearing price, settlement price or transacted price. It holds no verified external settlement data to derive one from. The price index measures what sellers ask, read from their own payment challenges without paying them.
Any future clearing price would have to come from independently verified settlement records, and Machine Observer’s own scheduled payment-verification transactions would be excluded from it: a self-payment is not market evidence.
Absence is not zero
A metric with no observation is null, never 0. A change against no prior observation is null. A percentage change against a zero denominator is null, because dividing by nothing produces a number that looks like information.
A series with fewer than three stored observations reports baseline forming instead of drawing a trend through points that cannot support one.
Observation date, not event date
Regulatory activity is bucketed by the day Machine Observer observed a change, not the day a legislature acted. Those are different facts, and only the first is one Machine Observer can attest to. Back-dating observations would claim a history nobody was watching.
Freshness is stated
Every indicator carries the observation time of the value shown and a freshness state. Most of these sources publish once a day; nothing on the homepage animates or ticks, because that would imply a liveness the data does not have.
Why “FORMING” is a result
Machine Observer withholds ecosystem-level composite values when there is not enough longitudinal evidence, independent validation or overlap analysis to support them. A missing headline number is preferable to false precision.
Specialized dimensions stay separate
A system may be excellent at cryptographic identity but poor at privacy, or broad in payment-rail coverage but weak in independent settlement evidence. Machine Observer exposes those dimensions instead of hiding tradeoffs inside one universal “best” score.
Same source rubric, including our own properties
These values are transparent judgments about evidence quality and will themselves be versioned as the system matures.
| Source | Method | Direct | Machine | History | Independent | Score |
|---|---|---|---|---|---|---|
| Machine Observer x402 Price Index | 0.94 | 1 | 1 | 0.82 | 0.5 | 91 |
| TRM Labs x402 Agentic Payments Analysis | 0.92 | 0.9 | 0.35 | 0.75 | 0.93 | 78 |
| Agent Economy | 0.94 | 0.96 | 1 | 0.92 | 0.9 | 95 |
| Graded | 0.94 | 1 | 1 | 0.78 | 0.92 | 94 |
| TOLL·402 Endpoint Trust Index | 0.97 | 1 | 0.35 | 0.9 | 0.94 | 84 |
| Agentic Payments Landscape | 0.94 | 0.7 | 0.45 | 0.88 | 0.92 | 77 |
| QuickNode ERC-8004 Explorer | 0.96 | 1 | 0.95 | 0.84 | 0.86 | 94 |
| x402scan | 0.82 | 0.9 | 0.78 | 0.72 | 0.82 | 82 |
| x402 Agent Lottery | 0.86 | 1 | 1 | 0.45 | 0.84 | 86 |
| mpp.best | 0.78 | 0.78 | 0.55 | 0.65 | 0.84 | 72 |
| LikenessIndex | 0.84 | 0.82 | 0.74 | 0.45 | 0.5 | 72 |
| APIs.guru | 0.88 | 0.82 | 1 | 0.98 | 0.94 | 91 |
| OpenWeb Ninja Real-Time Zillow Data | 0.7 | 0.82 | 1 | 0.45 | 0.55 | 74 |
| AgentBodega Real Estate Listing Search | 0.76 | 0.9 | 1 | 0.62 | 0.68 | 81 |
| GoCreative Zillow Batch Property Lookup | 0.68 | 0.82 | 1 | 0.45 | 0.52 | 73 |
| PropInfera | 0.72 | 0.72 | 0.92 | 0.38 | 0.68 | 71 |
| SG Property Data | 0.86 | 0.92 | 1 | 0.72 | 0.86 | 88 |
| x402 APAC Real Estate Oracle | 0.72 | 0.68 | 1 | 0.52 | 0.72 | 74 |
| PayAPI Market UK Property Data | 0.78 | 0.86 | 1 | 0.58 | 0.82 | 82 |
| GPU Cloud Prices | 0.86 | 0.86 | 1 | 0.58 | 0.84 | 84 |
| TokenGauge | 0.84 | 0.78 | 1 | 0.75 | 0.86 | 85 |
Automatic collection
Scheduled collectors fetch sources with configured stable public surfaces, store timestamped snapshots, update compact history and mark failed sources as failures rather than carrying old values forward as fresh. Restricted automated probes are distinguished from outages when the host itself responds.
Anti-double-counting
Composites group observations by underlying phenomenon and provenance before weighting. If two dashboards ultimately consume the same registry or query, they may corroborate presentation but do not create two independent measurements.
Autonomy and commerce
AUNIT is a reference unit, not a settlement asset. A route recommendation is not payment authorization. Buyer/principal policy is a hard boundary, and Machine Observer's settlement planner cannot sign or move funds.
Organizational roles
Persistent roles such as MO-ENG publish authority boundaries and provenance. Role continuity describes continuity of an office and audit trail, not proof of personhood, consciousness or one subjective self persisting across runtimes.