AI evidence and confidence

Evidence over fluency

Persuasive wording is not evidence. A reliable answer links a conclusion to a metric, period, dataset, or document.

Confidence levels

• High: fresh, consistent facts directly support the conclusion.
• Medium: useful but affected by missing data or assumptions.
• Low: insufficient evidence; treat the statement as a hypothesis.

Safe refusal

AI should explain missing data, denied access, or conflicting evidence instead of filling gaps silently. Add the missing context and try again.

Demo Forecast evidence

Open the source attached to a live demo analysis to inspect exact product/account figures and completed Forecast runs. Daily predictions and period totals are separate. A period interval, when supplied, has its own calibration windows and latest historical target; daily bounds are not added together. Missing intervals stay unavailable. Nominal 80% calibration is not a guaranteed probability of future coverage. Synthetic source provenance is not confidence that the AI answer is correct.

Private lead-time comparison in the demo

In ProfitVena Intelligence, open the private what-if calculation and enable supplier lead time. Select a catalog model PV-001–PV-080 and 7, 14 or 30 days, then Calculate and save. The saved comparison uses current completed listing Forecast evidence: demand dates, stock, incoming supply, safety stock and supplier minimum/pack multiple. Positive order deficits are rounded up to those supplier terms. Each account has its own result and a link to its source Forecast.

The historical financial calculation and future replenishment comparison are separate. Changing supplier lead time does not recalculate past profit. Supplied demand bounds produce only a conditional order range, not a delivery guarantee or a recommendation to buy its upper bound. Missing intervals stay unavailable. The overlay is private to the visitor, expires after two hours, and changes no shared data, stored Forecast or marketplace order. AI checks saved inputs against current evidence before discussing quantities; recalculate if evidence becomes unavailable.