Trust, shown — not self-attested

How every citation is verified

An answer is only as trustworthy as the source under it. Before any cited answer is shown, a dedicated citation-verification stage checks each reference against the indexed corpus and links it back to the original article.

1. 1The model proposes an answer with the specific articles it relied on.
2. 2The citation-verification stage confirms each cited article actually exists in the corpus and supports the claim — unverifiable claims are flagged, not hidden.
3. 3You can trace every citation to the original Arabic source text and read it yourself.

Confidence is a band — never a bare percentage

A single number like “87% confident” looks precise and means nothing. We refuse it. Every answer carries a confidence BAND with the drivers behind it and the sources it rests on, so you can judge the reasoning instead of trusting a digit.

We explicitly reject the false-precision “5–95%” score some tools display.

• High — directly grounded in in-force, cited articles with no conflicting source.
• Medium — supported, but with assumptions or gaps you should check.
• Low — thin or conflicting sourcing; treat as a lead, verify before relying on it.

The sources, with real freshness dates

This list is read live from the corpus. Each row is a source family, its record count, and the last time we refreshed it from the official origin — straight from the freshness ledger, no estimates.

Freshness data is updating — check back shortly.

Where your data lives — our honest posture

AI inference and your data are handled in regional data centres, with no unnecessary cross-region routing.

We never train on your data

Your matters, documents and questions are never used to train any AI model. The models we call are inference-only; we send the minimum context needed to answer, and nothing you submit is ever used to train a model. (Anonymous query text may be kept briefly for quality and abuse review, then deleted — see our Privacy Policy.)