Integrity infrastructure for Islamic knowledge
Know when trusted knowledge changes.
Istithbat detects source changes, tests their effect on AI answers, and keeps unreviewed knowledge out of production.
AI advises.Policy governs.Humans decide.


- «دون قوله: (ولم يستدر)» removed · ar.grade
- Evidence-scope drift · draft advisory, pending validation
- Matched regression pending validation
The source changed.Production didn’t.
Istithbat fingerprints every monitored record. When a field changes it shows the exact change, word for word, while production keeps serving the version a reviewer trusted.
Real HadeethEnc source record · Istithbat-controlled test mutation
Original record from HadeethEnc. The candidate mutation was created in the Istithbat sandbox and was not published by HadeethEnc.
View the original recordصحيح دون قوله: (ولم يستدر)
صحيح
See what the changecan reach.
Istithbat asks the protected app’s model the same questions twice, changing only the knowledge. Then it follows the record through every downstream asset.
Same modelSame settingsSame questionDifferent knowledge v13 ≠ v14matched-answer result pending validation Draft
Same model · Same settings · Only knowledge changed
Baseline answer
Candidate answer
- hadeethenc:10618ar.grade changed
- Evidence DatasetDataset · Exposed
- RAG ChunksChunk · Exposed
- Knowledge IndexIndex · Exposed
- Q&A APIAPI · Exposed
- Islamic Q&AApplication · Exposed · impact pending validation · protected
- Search APIAPI · Exposed · frozen copy
- Content ExplorerApplication · Exposed · frozen copy
Exposure comes from the dependency graph. Impact needs proof: a material regression against a protected app. Islamic Q&A stays exposed, not impacted, until the 10618 regression is validated.
Stale applies only to a frozen copy after a newer version is trusted while that copy still holds the previous one. Nothing has been promoted, so Search API’s frozen copy is exposed, not stale.
Unreviewed knowledgestops here.
A deterministic policy holds the candidate at the Trust Gateway. The protected app keeps reading the trusted version until a reviewer signs.
v14 is stopped at the Trust Gateway by POL-002; v13 passes through and is served to Islamic Q&A.
Latest v14 is not trusted v13; trusted v13 equals served v13.
- AI advises DraftEvidence-scope drift
Suggests quarantine. It can add caution; it cannot remove it.
Advisory. Not a ruling. - Policy containsPOL-002 · judgment change
Quarantine, retest, review. The same change gets the same outcome, whatever the AI says:
Policy outcome for this change under different AI inputs, computed by the policy engine AI suggests quarantine Draft POL-002 · quarantine AI suggests allow POL-002 · quarantine AI fails to answer POL-002 · quarantine - A human decidesReviewer decides
- Approve
- Reject
- Keep quarantined
- Escalate
Only a signed decision moves v14. Approval switches the gateway in one transaction, and every step is appended to the record.
Real sources.Measured scope.
Both connectors read the official HadeethEnc and QuranEnc APIs. Thousands of records were validated offline; Production deliberately serves a small, trusted baseline.
Full-corpus validation · offline
- 3,574
- HadeethEnc Arabic records validated
- One mark per record
- 2,328
- English translations validated
- One Arabic · English pair per translation
- 114/6,236
- Quran surahs / ayat validated
- One bar per surah · height = its ayat
- 188
- Automated tests passing
- One mark per passing test
Validated artifacts remain separate from the intentionally small live Production baseline.
2 independent passes over the official APIs; every raw, canonical, record and field hash matched. Quran figures are for english_saheeh (Noor International Center, v1.1.2). Tests: 7 optional live-database tests skipped; typecheck and Production build pass. Validated artifacts stay unassessed and unserved.
Current live baseline
- HadeethEnc
- 3 monitored records · Arabic and English
- QuranEnc
- 11 monitored ayat · english_saheeh
Production intentionally uses a small trusted baseline. Full-corpus coverage was validated offline before live ingestion is expanded.
AI advises.Policy governs.Humans decide.
- Analysis and regression verdicts are advisory. They can add caution; they never lower a policy floor.
- 5 deterministic rules set the floor for every change, by field role.
- Only a signed reviewer decision moves a candidate into production.