Truthound · All articles · 3/24
Evidence-led Truthound Core guidance on data contracts, check inputs and operational limits.
480 articles
An AI employment record needs a correction route as well as a consent field
Connect the decision, evidence and responsible reviewer before treating a record as settled.
A production time record should show the work outside the camera take
Keep activity and contractual treatment separate when reviewing missing production time.
No recorded disaster loss is not proof of low risk
Check the event-to-impact coverage before using absence as evidence.
A climate-data promise depends on the observing network that can keep supplying it
Trace a data series to its maintained network and successor responsibility before promising continuity.
A seasonal forecast needs an issue date and a target interval
Keep forecasts out of observed-data evaluation samples unless their role is explicit.
A biodiversity bond's issuance is not yet a conservation result
Trace one allocation to its project before interpreting the environmental outcome.
A missing contact route should create rural-support work, not erase an applicant
Keep the application visible while an authorized team confirms an alternative contact path.
A 30% cross-border target needs a defined denominator before a progress chart
Obtain the official counting rule before calculating a regional-cooperation share.
An antigen-data submission needs the assay roles, not just a variant name
Review the requested version and scientific comparison conditions before accepting a data bundle.
A small AI field experiment needs both power and connectivity readiness
Do not attribute an infrastructure interruption to the AI model's usefulness.
An AI trade forecast is not a chip company's revenue forecast
Name the series, price basis, period and geography before using a macro figure in a demand plan.
A new trade portal entry still needs an original source before a transaction decision
Keep source, verification date and unresolved conditions attached to the opportunity.
Reproduce a climate-finance change as a revision plus a new-period result
Reconcile publication vintages before interpreting an increase or decrease as current delivery.
Before scaling a health-AI tool, name the evidence for its intended setting
Return an expansion request when the required setting-specific evidence has no owner.
A health-data trust discussion should preserve disagreement, not manufacture a score
Record unresolved views and follow-up responsibility without turning silence into consent.
Triage a digital health document exchange failure before changing its metadata
Reproduce one failure, separate trust verification from internal metadata preparation, and hand the evidence to the responsible operator.
Training participation is not the number of countries with a new health-data capability
Separate repeated attendance from applied learning and follow-up needs.
A health-AI access review must keep the groups that did not answer online
Define the intended population before counting the groups represented in a review.
A cross-border message needs a classification version, not only a valid file format
Approve jurisdiction-and-version combinations before reusing a customs data contract.
A thin digital record is not evidence that a small business has no sales
Treat missing business records as an evidence question before using them in a financial decision.
