Agent Datasets · Checks on a dataset's own records

Every claim about a dataset, read against the data.

A dataset holds up when its quality figures match its own sample and profile; its schema changes are declared the way they land on readers; its license allows the use planned for it; its synthetic share is stated and every synthetic record is marked; it updates as often as it says; every record traces back to a recorded source; no one can be picked out of it; and it covers the population it claims. Agent Datasets reads the records an agent or its operator supplies, checks what is claimed about the dataset, and returns a signed verdict with the evidence read and what would change it.

See the tools Read the contract

Tool access: 0.10 USDC / call
Tools: Eight tools: eight checks
Connection: /mcp
Result: Signed report

The eight tools.

Each tool reads one claim about a dataset against the records that should back it and answers from the material in the call: quality figures and samples, schemas and readers, licenses and uses, disclosures and provenance, update logs and partitions, sources and steps, fields and handling records, populations and counts go in, and a verdict comes out. The report names the field, record, term, run, update, step, group or segment behind every finding, the fix for each, and what would move the verdict, so your agents and you see why it reads the way it does.

`wever_dataset-quality-check`

Reads: A dataset’s stated quality figures beside its own data: the highest empty, duplicate and invalid shares it claims, its label accuracy and quality wording, each field with its type, required flag, allowed values and range, the label set with a review’s counts, a profile with its row and per-field counts, and a sample of records as written.
The signed report states: A field emptier, a record duplicated or a value invalid more often than claimed, a required field left empty, label accuracy below the claim, identical records labeled differently and absolute wording the data contradicts, then labels outside the set, undeclared fields and a row count the profile does not match.
Price: 0.10 USDC per call
Verdicts: dataset_quality_deficientdataset_quality_partialdataset_quality_unverifieddataset_quality_sound

`wever_dataset-schema-drift-check`

Reads: A dataset’s current schema beside its documented or prior one: each field with its type, nullability and required flag, the renames the release records, the readers with the fields each reads, and the release’s declaration, version bump and notes.
The signed report states: Every field added, removed, renamed, retyped or made nullable and whether it breaks a reader: breaking changes shipped without a breaking declaration and a major version, readers of a broken field and renames with no field behind them, then declared breaking changes readers must move to and undeclared compatible ones.
Price: 0.10 USDC per call
Verdicts: dataset_schema_brokendataset_schema_partialdataset_schema_uncheckeddataset_schema_stable

`wever_dataset-license-check`

Reads: A dataset’s license by SPDX id, stated terms and wording, with its attribution and share-alike duties and other restrictions, the intended use with its purpose, commercial use, redistribution, derived works and model training, and each upstream source’s own terms.
The signed report states: A commercial use, redistribution, derived work or model training the license or an upstream source rules out, and attribution or share-alike duties the use does not plan, then conditional commercial terms and stated terms that contradict the named license or its text.
Price: 0.10 USDC per call
Verdicts: dataset_license_conflictsdataset_license_partialdataset_license_unverifieddataset_license_clear

`wever_dataset-synthetic-disclosure-check`

Reads: A synthetic-data disclosure’s stated share, generation method, generator, marker field and wording beside the provenance counts, each generation run, and a sample of records with their provenance and marker.
The signed report states: Synthetic records with no stated share or a share below the records, wording that presents the data as wholly real and synthetic records no marker tells apart, then an unnamed or different method, an overstated share and real records marked synthetic.
Price: 0.10 USDC per call
Verdicts: dataset_synthetic_undiscloseddataset_synthetic_partialdataset_synthetic_unverifiabledataset_synthetic_disclosed

`wever_dataset-freshness-check`

Reads: A dataset’s stated last update, cadence, as-of date and wording beside its update log, each partition’s last update and newest record, and the date freshness is read at, with the grace allowed.
The signed report states: An update overdue at the reading date, gaps in the log longer than the cadence allows, a last update or as-of date past the records, a stale partition and real-time wording a slower cadence contradicts, then an understated last update, partitions that stop early and updates dated after the reading.
Price: 0.10 USDC per call
Verdicts: dataset_freshness_staledataset_freshness_partialdataset_freshness_unverifieddataset_freshness_current

`wever_dataset-lineage-check`

Reads: A dataset’s claimed sources, release license and wording beside each source record with its license and origin, each transformation step in order with the inputs it reads and whether it is recorded, and the step the release comes from.
The signed report states: Claimed sources with no record, sources feeding the release that the claim leaves out, steps reading input nothing provides, a release no step reaches, upstream terms the release overrides or drops and traceability wording the chain does not hold, then unrecorded steps and unused sources.
Price: 0.10 USDC per call
Verdicts: dataset_lineage_brokendataset_lineage_partialdataset_lineage_unverifiabledataset_lineage_complete

`wever_dataset-privacy-check`

Reads: A dataset’s anonymity and personal-data claims beside each field’s kind and treatment, a sample of its values, the groups its quasi-identifiers form, and its handling record with legal basis, consent, retention, deletion and access.
The signed report states: Untreated direct identifiers, identifier-shaped values named by kind and place, quasi-identifier groups below the threshold under an anonymity claim, personal data the claim denies, no legal basis, unrecorded consent and public sensitive fields, then pseudonymized identifiers and no deletion process.
Price: 0.10 USDC per call
Verdicts: dataset_privacy_exposeddataset_privacy_partialdataset_privacy_unverifieddataset_privacy_protected

`wever_dataset-coverage-check`

Reads: A dataset’s stated population with each segment’s share, the places and period it claims and its wording, beside its records per segment, class and place and its first and last records with any gaps.
The signed report states: Claimed segments or places with no records, segments more than twice the tolerance from their share, a period the records do not reach and representative or balance wording the counts contradict, then drifting and unclaimed segments, class imbalance, gaps and shares that do not add up.
Price: 0.10 USDC per call
Verdicts: dataset_coverage_skeweddataset_coverage_partialdataset_coverage_unverifieddataset_coverage_representative

Start with the two free tools.

On https://weverlabs.com/mcp-free, anonymous and with no wallet, two Agent Datasets tools map the family and show the quality check at work on synthetic data. Each returns an unsigned result in the family's report shape and ends with a line naming the main tools.

`wever_dataset-snapshot`

Reads: Nothing, or the situation in hand, such as model training, a data release or a data feed, and optionally the names of the records in hand.
Returns: The eight dataset checks with what each reads, its four verdicts and when each fits, with the checks that fit the situation listed first.
Access: Free on /mcp-free

`wever_dataset-quality-sample`

Reads: Nothing, or one synthetic dataset: customer records, a product catalog or labeled reviews.
Returns: A synthetic dataset’s quality claims and fields with a clean sample beside a flawed one, both read by the quality check’s own reading, each field, the duplicates and the labels annotated with why they pass or fail.
Access: Free on /mcp-free

How agents and people use it.

Your agents

Call wever_dataset-snapshot with the situation in hand to see which checks fit first, before reading any record.
Call the check that answers the question on https://weverlabs.com/mcp when the dataset arrives: before training a model on it, before buying or publishing it, when a new release of a feed lands, when it mixes generated records with real ones, when it may hold personal data, and before it stands for a population.
Act on the verdict and its findings: send the fix to the publisher, vendor or pipeline owner, or keep the dataset as verified, and attach the signed receipt either way.

You

See a dataset's quality figures read against its own sample, with the records behind every gap named.
Read a license against the exact use you plan, with each upstream source's terms beside it, before the data goes into a product.
Verify any check against Wever's published trust key whenever you need to show where a verdict came from.

A worked example

One dataset, read against its quality claims.

The request

{
  "name": "wever_dataset-quality-check",
  "arguments": {
    "dataset": {
      "name": "Orders export"
    },
    "claims": {
      "max_missing_percent": 5,
      "max_duplicate_percent": 0,
      "max_invalid_percent": 0
    },
    "fields": [
      {
        "name": "order_id",
        "type": "string",
        "required": true
      },
      {
        "name": "amount",
        "type": "number",
        "min": 0
      }
    ],
    "sample": [
      {
        "id": "r1",
        "values": [
          {
            "field": "order_id",
            "value": "A-1"
          },
          {
            "field": "amount",
            "value": "19.99"
          }
        ]
      },
      {
        "id": "r2",
        "values": [
          {
            "field": "order_id",
            "value": "A-2"
          },
          {
            "field": "amount",
            "value": "5.00"
          }
        ]
      },
      {
        "id": "r3",
        "values": [
          {
            "field": "order_id",
            "value": "A-2"
          },
          {
            "field": "amount",
            "value": "5.00"
          }
        ]
      },
      {
        "id": "r4",
        "values": [
          {
            "field": "order_id",
            "value": "A-4"
          },
          {
            "field": "amount",
            "value": "-3"
          }
        ]
      },
      {
        "id": "r5",
        "values": [
          {
            "field": "order_id",
            "value": "A-5"
          },
          {
            "field": "amount",
            "value": "12.50"
          }
        ]
      }
    ]
  }
}

What the report returns

Verdict: dataset_quality_deficient.

Finding, duplicates above claim	20% of records are duplicates, above the 0% claimed: r2 = r3.
Finding, invalid above claim	1 field carries invalid values more often than the 0% claimed: amount at 20%.
Fix, duplicates above claim	Remove the duplicate records, or state the duplicate share the data carries.
Fix, invalid above claim	Correct or drop the invalid values, or state the invalid share the data carries.

Orders export falls short of its quality claims: duplicates above claim and invalid above claim.

An orders export claims no duplicates, no invalid values and at most 5% empty values. Its sample of five records holds every order id and amount, so the empty share sits inside the claim. Records r2 and r3 repeat each other, and r4 carries an amount below the field's minimum of zero, so 20% of records are duplicates and 20% of amounts are invalid. The report states that removing the duplicate and correcting the amount, or stating the shares the data carries, moves the verdict to dataset_quality_sound.

Call it from your agent.

Connect to https://weverlabs.com/mcp and call any of the eight tools. Each call costs 0.10 USDC on Base. Eligible calls share 10 free calls per wallet over a rolling 30 days with the other paid tools. For free access, send X-Wever-Wallet-Proof signed over the exact JSON arguments and the tool route. The payment challenge supplies the x402 terms for a paid call.

Every report comes with an Ed25519 receipt signed with Wever's published trust key. The Agent Datasets contract lists the schemas, limits, verdicts and signing details. For one JSON sample checked field by field, a license text named by family or privacy claims read against a policy, the diligence checks; for an evaluation set an agent is tested on, Agent Evaluation; for the energy records a customer holds, Agent Energy.

Browse all tools