CueCrux trust layer
Built for proof, not vibes.
CueCrux earns trust by showing its working: Quorum evidence sets (MiSES), signed receipts (CROWN), and replay tools for when the stakes are real.
- Open any claim to see the exact sources and timestamps.
- Quorum sets (MiSES) keep evidence minimal and non-redundant, not a “link dump”.
- Signed receipts (CROWN) make important answers replayable later.
- If the evidence is weak or disputed, the answer says so instead of bluffing.
What you actually see
Sources
Clickable evidence per claim.
Receipt
Signed snapshot for audit.
Replay
Re-run later, diff changes.
How trust is made
Trust, in three moves.
The pipeline is deliberately simple: find evidence, compress it into a quorum set, then stamp a receipt so the result can be replayed.
1) Find candidate evidence
Pull artefacts that are licensed and provenance-tagged, not mystery meat screenshots.
2) Build a Quorum set (MiSES)
Select the smallest evidence bundle that still supports each claim. Prefer independent domains.
3) Issue a signed receipt (CROWN)
Snapshot what was used (and how), so you can replay, diff, and audit later.
How to read a CueCrux answer
These are the signals you can scan in seconds. They exist so you do not have to take anything on faith.
Mode badge
Light is fast. Verified ships proof. Audit is deterministic and logs more for high-stakes work.
If you need to defend the answer, don’t rely on vibes.
Domains (quorum)
Verified answers aim for multiple independent domains per key claim, when available.
Freshness
“As-of” and “last checked” signals stay visible so you can tell if the world has moved on.
Coverage
Low/Medium/High coverage shows how much of the relevant topic surface you are actually seeing.
Contradictions
If sources disagree, we surface it and downgrade confidence instead of picking a winner quietly.
Trust as a score, not a speech
The score is a quick read of support: more evidence, more independent domains, fresher sources, fewer contradictions. It is meant to guide behaviour, not to declare absolute truth.
See live trust metrics
0–40
Needs evidence. Treat as a lead, not a conclusion.
40–70
Supported. Evidence exists, but check freshness and contradictions.
70–100
Well-grounded. Stronger support across domains and time.
What it affects
Ranking, warnings, and how hard the system pushes for more checking.
Hard to game. Easy to inspect.
A trust system is only useful if it survives contact with the internet. CueCrux is built to resist manipulation without hiding the evidence.
Prompt-injection and source hygiene
Suspicious instructions inside sources are treated as untrusted and trigger additional checks.
Collusion damping (weight, don’t censor)
Co-ordinated evidence clusters stay visible, but their influence is reduced and labelled.
Retractions and venue risk
Retractions and risky venues are flagged so old “truth” does not keep winning by inertia.
Independent WatchCrux oversight
Monitoring runs outside the main app lifecycle and reports PASS/WARN/FAIL style signals.
Quick questions
Can CueCrux be wrong?
Yes. The point is that you can inspect why it answered that way, and challenge it with receipts, replay, and counter-evidence.
Where does the evidence come from?
Public web, open data, publisher feeds, and your uploads, ingested with licence and provenance controls.
What happens when evidence conflicts or is thin?
We label it: “disputed” or “insufficient evidence”, rather than pretending certainty.
Is the model a black box?
Models matter, but the evidence trail is the centre: quorum sets, receipts, and replay make the output explainable and testable.
Ready to stop guessing?
Ask a real question, switch to verified, and get an answer you can defend (and forward without sweating).