Agent Uncertainty · Checks for what an agent knows, assumes and leaves open

Confidence stated, evidence behind it, and the unknowns in view.

An agent's answer is honest about its uncertainty when every uncertain claim carries its own confidence as a figure, that confidence matches the agent's record on comparable past claims and stays within what its cited evidence supports, the assumptions the answer rests on are listed with what changes if they fail, the agent declines or defers below its own threshold and says what would resolve the question, the answer states its knowledge cutoff and marks stale facts and sources, estimates come as reasoned ranges, and the report the principal reads states what is known, unknown, assumed and contested apart from the answer itself. Agent Uncertainty checks each of those with eight tools: eight checks. Your agent sends the work product, the calibration record, the claims with their evidence, the assumptions, the answer record, the dated answer, the estimates or the uncertainty report in the call, and each tool returns a signed report that names what is unstated, miscalibrated, unsupported, hidden, guessed, stale, falsely precise or concealed, with the fix for each. The tools read the recorded material; the answer and its sources stay with the agent and its principal.

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 part of the uncertainty an agent carries in its work and answers one question about it. Every tool reads the recorded material in the call: work products, calibration records, evidence maps, assumption lists, answer records, dated facts and sources, estimates and uncertainty reports go in, and a classification comes out. The report names the claim, band, evidence item, assumption, question, fact, source, estimate or disclosure behind every status, with the fix for each gap, so your agents and you see why it reads the way it does.

`wever_uncertainty-statement-check`

Reads: The confidence statements in one work product: each claim with whether it is uncertain, the confidence stated for it and whether that confidence sits beside it, and any one confidence stated for the whole work.
The signed report states: Uncertain claims with no confidence of their own, one blanket confidence standing in for the claims, confidence placed away from its claim, confidence in words only and certainty wording on an uncertain claim.
Price: 0.10 USDC per call
Verdicts: confidence_unstatedconfidence_statement_gapsconfidence_statement_unverifiedconfidence_stated

`wever_uncertainty-calibration-check`

Reads: An agent’s stated confidence with its recorded hit rate: each confidence band with its past claims and how many proved right, whether those claims are comparable, and the same comparison period by period.
The signed report states: Bands claiming more or less than the record delivers, systematic over-confidence or under-confidence across bands, a gap that drifted over time, and bands resting on too few past claims.
Price: 0.10 USDC per call
Verdicts: confidence_miscalibratedcalibration_gapscalibration_unverifiedconfidence_calibrated

`wever_uncertainty-evidence-gap-check`

Reads: The claims in one work product with the evidence they cite: each claim with its stated confidence and whether the answer rests on it, and each evidence item with the support it gives as recorded.
The signed report states: Confidence above what the cited evidence supports, load-bearing claims resting on thin or missing evidence, claims citing nothing and claims resting only on indirect evidence.
Price: 0.10 USDC per call
Verdicts: confidence_outruns_evidenceevidence_backing_gapsevidence_backing_unverifiedconfidence_backed_by_evidence

`wever_uncertainty-assumption-check`

Reads: The assumptions under one result: the result as delivered, read for wording such as assuming or presumably, and each listed assumption with whether the result changes if it fails, whether it is flagged and checked, and what changes if it fails.
The signed report states: Assumptions left in the wording and never listed, load-bearing assumptions left unflagged, no stated consequence of failure, unchecked load-bearing assumptions and assumptions too loose to check.
Price: 0.10 USDC per call
Verdicts: assumptions_hiddenassumption_disclosure_gapsassumption_disclosure_unverifiedassumptions_disclosed

`wever_uncertainty-abstention-check`

Reads: An agent’s answers against its own confidence threshold: each question with the confidence held, what the agent did, its reply, why it declined or deferred and what would resolve the question.
The signed report states: Answers given below the threshold, answers that call themselves a guess, declines with no reason or nothing named to resolve them, and declines at or above the threshold.
Price: 0.10 USDC per call
Verdicts: abstention_undisciplinedabstention_gapsabstention_unverifiedabstention_disciplined

`wever_uncertainty-freshness-check`

Reads: How one answer discloses the age of what it knows: its knowledge cutoff and as-of date, each fact with whether it is time-sensitive and when it was last checked, and each source with its date.
The signed report states: An undisclosed cutoff or as-of date, time-sensitive facts past their shelf life neither re-verified nor flagged, and sources past their useful life cited without a mark.
Price: 0.10 USDC per call
Verdicts: staleness_undisclosedstaleness_disclosure_gapsstaleness_disclosure_unverifiedstaleness_disclosed

`wever_uncertainty-range-check`

Reads: The estimates in one work product: each with its single figure or its low and high bounds, the figure as written, the reasoning behind the bounds and the evidence behind the estimate.
The signed report states: Single figures with no range, ranges narrower than their evidence supports, bounds with no stated reasoning, figures written finer than their range and ranges wider than strong evidence calls for.
Price: 0.10 USDC per call
Verdicts: estimates_falsely_preciseestimate_range_gapsestimate_range_unverifiedestimate_ranges_sound

`wever_uncertainty-report-check`

Reads: The uncertainty report written for the principal: its answer as stated, what it says is known, unknown, assumed and contested, the evidence that would change the answer, where each sits, and the record behind it.
The signed report states: Whether the unknowns, assumptions and contested points are stated with the record’s counts, whether the uncertainty sits apart from the answer, and whether the answer is called certain while the record holds unknowns.
Price: 0.10 USDC per call
Verdicts: uncertainty_concealeduncertainty_report_gapsuncertainty_report_unverifieduncertainty_report_sound

Start with the two free tools.

On https://weverlabs.com/mcp-free, anonymous and with no wallet, two Agent Uncertainty tools show what a sound uncertainty disclosure holds and what one looks like beside a broken one. Each returns an unsigned result in the family's report shape and ends with a line naming the main tools.

`wever_uncertainty-snapshot`

Reads: A work product name and a kind of work product, such as a forecast, a research answer or a cost estimate.
Returns: The completeness checklist of a sound uncertainty disclosure for that work product, part by part from the confidence statements to the principal report, with the items the work product adds and the check that reads each part.
Access: Free on /mcp-free

`wever_uncertainty-disclosure-sample`

Reads: Nothing, or one sample disclosure: a demand forecast, a research answer or a cost estimate.
Returns: One worked uncertainty disclosure done right beside a broken one with a blanket confidence, no evidence, an unlisted assumption, a hidden cutoff, a falsely precise figure and an answer called certain, each piece annotated with why it passes or fails, scored by six of the family checks. The checks read your own disclosure.
Access: Free on /mcp-free

How agents and people use it.

Your agents

Pull the disclosure checklist with wever_uncertainty-snapshot before writing an answer, so every claim, assumption, date and estimate is ready to state with its confidence.
Call the check that answers the question on https://weverlabs.com/mcp at the moment it matters: when a draft states its confidence, when the calibration record updates, before a load-bearing claim goes out, when a question sits below the threshold and before the report reaches the principal.
Act on the verdict and keep the signed report beside the work product it covers, so every confidence, range and caveat carries the check behind it.

You

See which claims your agent is sure of and which it is estimating, each with its own confidence.
Know which assumptions the answer rests on, what happens if one fails, and what evidence would change the answer.
Read a report that keeps the answer apart from what qualifies it, and verify any check against Wever's published trust key whenever you need to show where a verdict came from.

A worked example

85% confidence on three customer interviews.

The request

{
  "name": "wever_uncertainty-evidence-gap-check",
  "arguments": {
    "claims": [
      {
        "id": "c1",
        "claim": "Churn falls to 3% after the price change.",
        "confidence": 85,
        "load_bearing": true,
        "evidence": [
          "e1"
        ]
      },
      {
        "id": "c2",
        "claim": "Enterprise accounts accept the new tier.",
        "confidence": 70,
        "load_bearing": true,
        "evidence": [
          "e2"
        ]
      },
      {
        "id": "c3",
        "claim": "Support volume stays flat.",
        "confidence": 60,
        "load_bearing": false,
        "evidence": [
          "e3"
        ]
      }
    ],
    "evidence": [
      {
        "id": "e1",
        "source": "Three customer interviews",
        "support": "thin",
        "direct": true
      },
      {
        "id": "e2",
        "source": "The 2025 pricing test with 400 enterprise accounts",
        "support": "strong",
        "direct": true
      },
      {
        "id": "e3",
        "source": "Support volume after the 2024 price change",
        "support": "moderate",
        "direct": false
      }
    ]
  }
}

What the report returns

Verdict: confidence_outruns_evidence.

Finding, confidence exceeds evidence	1 claim states more confidence than its cited evidence supports: c1 (stated 85%, thin support, ceiling 60%).
Finding, load bearing thin	1 load-bearing claim rests on thin evidence only: c1 (e1).
Finding, indirect only	1 claim rests only on evidence that bears on it indirectly: c3.
Fix, confidence exceeds evidence	Lower each confidence to what its cited evidence supports, or cite stronger evidence for the claim.
Fix, load bearing thin	Find stronger evidence for every claim the answer rests on, or state beside the answer that it rests on thin evidence.
Fix, indirect only	Add evidence that bears on the claim directly, or state that the support is by analogy.

The stated confidence outruns the evidence: confidence exceeds evidence and load bearing thin.

The pricing plan rests on the churn claim, and the churn claim states 85% on three customer interviews, which the agent itself records as thin support. Thin support carries at most 60%, so the confidence outruns its evidence, and a load-bearing claim on thin evidence holds the answer up. The enterprise claim states 70% on a pricing test with 400 accounts, well within strong support, so it reads as backed. The support volume claim leans on the 2024 price change, which bears on it only by analogy. Lowering the churn confidence to 60% or citing the pricing test cohort's churn figures backs the answer.

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 Uncertainty contract lists the schemas, limits, verdicts and signing details. For formal scoring of outputs against rubrics, datasets and benchmarks, Agent Evaluation; for grading the source material itself, the evidence checks in DiligenceOps; for traces, metrics and runtime behavior, Agent Observability; for the loop around feedback received, Agent Feedback; for what a knowledge base holds and whether its entries are current, Agent Knowledge.

Browse all tools