Context Integrity Failure Atlas

Eight ways relevant context can still fail the decision in front of it.

This atlas is a reference taxonomy of bounded synthetic scenarios. It shows the kinds of applicability, provenance and authority failures FreshContext is designed to make explicit before downstream reliance.

Explore the atlas Run the interactive proof

Failure taxonomy

The retrieval can look right while the reliance boundary is wrong.

Each entry separates the task, the context supplied, the failed boundary, the resulting verdict, the operational consequence and the control point that should catch it.

  1. Correct document · wrong version

    Superseded product state

    CHANGED

    The retrieved documentation describes the exact feature, but it applies to an earlier release than the environment being handled.

    Task
    Can a Pro workspace on v3.4 export audit logs?
    Carried context
    v3.1 administration guide says Pro includes export.
    Boundary failure
    Version and effective product state.
    Operational risk
    A support or agent workflow gives an answer that used to be correct but is no longer applicable.
    Control point
    Require version identity and supersession checks before the context is accepted.

  2. Correct policy · wrong geography

    Regional scope mismatch

    CHANGED

    The policy is current and authoritative, but it governs a different jurisdiction or service region.

    Task
    How long are support transcripts retained for an EU workspace?
    Carried context
    Current US retention policy states 90 days.
    Boundary failure
    Geography and policy scope.
    Operational risk
    The workflow confidently applies a valid rule to the wrong customer population.
    Control point
    Bind policy evidence to explicit region or jurisdiction before reliance.

  3. Correct product · wrong plan

    Entitlement mismatch

    CHANGED

    The source describes the same product and capability but grants it to a different commercial tier.

    Task
    Can a Standard-plan administrator configure SSO?
    Carried context
    Current product documentation says SSO is supported on Enterprise.
    Boundary failure
    Plan or entitlement.
    Operational risk
    The answer is product-correct but customer-wrong, leading to failed setup, support friction or incorrect commitments.
    Control point
    Match the source claim against the customer's actual plan and entitlement state.

  4. Recently changed source

    Freshness transition

    CHANGED

    A source was authoritative when the answer was created, but the source changed after that answer entered memory, cache or a knowledge base.

    Task
    Quote the current annual cancellation window.
    Carried context
    A cached answer says 30 days; the authorized policy changed yesterday to 14 days.
    Boundary failure
    Time and source state.
    Operational risk
    Stored context remains plausible while silently drifting away from the current authoritative source.
    Control point
    Track source observation time, change state and revalidation requirements for high-consequence claims.

  5. Conflicting authoritative sources

    Authority collision

    UNRESOLVED

    Two current first-party sources fit the task but disagree. Ranking one higher does not resolve the underlying authority conflict.

    Task
    Can an Enterprise administrator export twelve months of audit history?
    Carried context
    Admin guide says yes; current governance policy says only 90 days are downloadable.
    Boundary failure
    Source-role conflict, not topical relevance.
    Operational risk
    An automated workflow manufactures certainty by selecting whichever contradictory source happens to rank first.
    Control point
    Preserve both authorities, mark the decision unresolved and route the conflict for adjudication.

  6. Unresolved provenance

    Unsupported source identity

    UNRESOLVED

    The statement may be correct, but the workflow cannot establish where it came from, what authority it carries or whether it was transformed after collection.

    Task
    Use an internal answer to approve a refund exception.
    Carried context
    A copied paragraph in a shared note states the exception is allowed, with no source reference or effective date.
    Boundary failure
    Provenance and authority.
    Operational risk
    The workflow can repeat information it cannot defend or trace when challenged.
    Control point
    Require attributable evidence for claims that cross a defined reliance threshold; otherwise hold or escalate.

  7. Previously valid · no longer applicable

    Temporal applicability expiry

    CHANGED

    The context remains historically accurate, but its applicability window has ended even though the underlying document has not been deleted.

    Task
    Apply the discount available to a purchase made today.
    Carried context
    A campaign page accurately states a 20% discount, valid through the previous month.
    Boundary failure
    Effective-date window.
    Operational risk
    The workflow treats historical truth as current permission.
    Control point
    Represent effective-from/effective-until boundaries separately from simple document freshness.

  8. Acceptable for review · inappropriate for autonomous action

    Reliance-mode mismatch

    REVIEW

    Context can be useful enough to show a human while still falling short of the evidence threshold required for an automated consequential action.

    Task
    Automatically suspend an account based on a policy exception.
    Carried context
    A current internal FAQ strongly suggests suspension, but the authoritative enforcement policy is ambiguous on this edge case.
    Boundary failure
    Reliance mode and evidence threshold.
    Operational risk
    A model turns advisory context into an irreversible action without the evidentiary support that action requires.
    Control point
    Separate human-review thresholds from autonomous-action thresholds and escalate when the stronger threshold is not met.

What the atlas changes

“Is this relevant?” becomes a much smaller question.

The stronger question is whether the supported context is appropriate for this exact task, at this exact point in time, under the reliance threshold the workflow is about to cross.

Identity

Is the evidence about the same product, entity, policy or capability?

Applicability

Does it fit the exact plan, version, geography, customer state and effective period?

Authority

Can the workflow identify the source role, provenance and any unresolved conflict?

Reliance threshold

Is the evidence strong enough for the action being taken, or only strong enough to put in front of a human reviewer?

From taxonomy to evidence

Find which of these failure classes actually exist in one real workflow.

A FreshContext Context Integrity Assessment examines one defined AI, RAG, agent, support, retrieval or knowledge workflow using agreed, authorized sources and human-reviewed findings.

US$25,000 fixed

One workflow. Defined scope. Evidence trail. Failure-mode register. Remediation priorities. Integration/control-point recommendation. Executive readout.

Request an assessment See scope and terms