What Common Tape is
Common Tape is a neutral market graph of enterprise AI capabilities. It organises what exists by the business job to be done — Finance → Record to Report → Bank Reconciliation — rather than by how a vendor happens to package it. An agent, an app, a plugin, an MCP server, a connector, a workflow and a platform-native feature can all fill the same Position, and a buyer usually cares which one fits their environment, not which category it was filed under.
The thing being tested is narrow and specific: are there decision-critical questions about discovering, comparing and combining AI capabilities that platform marketplaces, standardised metadata, software review sites and general-purpose AI search do not answer reliably? If there are, this graph is worth scaling. If there are not, it should be killed early rather than slowly.
How a claim gets recorded
Every decision-relevant field carries its own provenance: what the value is, which sources support it, what kind of sources those are, and when it was last checked. A vendor saying it integrates with NetSuite and a named customer running it on NetSuite are recorded as different facts, because they are different facts.
- E0
- Unverified marketing claim
- E1
- Builder documentation
- E2
- Demonstrable / testable functionality
- E3
- Marketplace or third-party verification
- E4
- Named customer attestation
- E5
- Independent evaluation
- E6
- Verified production execution evidence
Evidence level says how well a claim is supported. It says nothing about how good the capability is — a capability can be strongly evidenced and weak, or promising and barely evidenced at all. Where sources genuinely disagree, both readings are published rather than one being chosen.
The limits, stated plainly
- 0 canonical records and 14 fictional demonstration records are currently loaded. Fictional records are marked as such everywhere they appear, in the interface and in the API. They exist so the views can be reviewed before the real corpus is authored, and they describe no real product.
- The index is not exhaustive. Counts describe what Common Tape has indexed. Absence here is not evidence of absence from the market.
- 47 fields are recorded as unknown. That is not an oversight to be tidied away — it is the honest state of public information about this market, and closing those gaps is the work.
- There is no overall score. No rating, no percentile, no ranked "best". Those require a defined cohort and a published methodology, and inventing one would make the product worse while making it look more finished.
If your product is indexed here
The useful question is not whether you like the profile. It is: what is wrong, and what decision-critical information is missing? Corrections are recorded with their source and shown in the record’s history, so the profile improves visibly rather than silently.
The machine interface is at /api/v1/players, /api/v1/capabilities and /api/v1/meta. The evidence vocabulary and its rules are published at /api/v1/metaso an automated consumer can read the caveats too.