KELLY sector cognition OS
Evidence bench
Watch research claims become falsifiable hypotheses, see contradiction pressure, and inspect the proof trail needed for a serious PhD-style defense.
KELLY treats research as a living evidence graph: claims need sources, contradictions need tracking, falsifiers must be named, and conclusions must survive replay.
Objects on the bench
Required receipts
Live KELLY routes
- Causalea public lens manifesthttps://www.causalea.com/api/public-lens-apps
- Causalea Glass snapshothttps://www.causalea.com/api/glass-lab/live-snapshot
- Advanced Processing Evidence Spinehttps://www.causalea.com/api/advanced-processing/evidence-spine
Evidence bench
Research Lab
Watch research claims become falsifiable hypotheses, see contradiction pressure, and inspect the proof trail needed for a serious PhD-style defense.
KELLY treats research as a living evidence graph: claims need sources, contradictions need tracking, falsifiers must be named, and conclusions must survive replay.
24-hour new-evidence scan
Which new paper, metadata update, or citation signal materially changes a watched claim?
Falsification watch
What evidence would most directly falsify or weaken a Causalea/KELLY claim?
PhD defense evidence trail
Which empirical results best support or weaken the KELLY/Causalea thesis over time?
This window shows what KELLY is doing in Research Lab: what it is reading, what it thinks might matter, what proof exists, and what is still waiting.
Read the answer in this lab first. Open the proof shelf or JSON packet for the evidence behind it. Check the route status and last-checked timestamp before trusting freshness. Look for the mind-changer: the fact that would make KELLY revise the answer. If evidence is waiting, treat the output as a question or watch item, not a settled answer.
Live evidence appears only when KELLY exposes proof-linked sector output. Empty states are intentional waiting states, not filler.
New approved evidence, a falsifier, settlement, replay drift, calibration movement, or memory-lineage revision.
This lab communicates with KELLY as an inspection window. It cannot write ground truth, memory, GraphState, provider state, or reality receipts.
Processing flow
Claims stay separate from evidence
KELLY answer lane
Ask about source posture, proof, falsifiers, forecast movement, settlement, calibration, or memory eligibility. The answer stays inside this lab and links back to Glass Lab where available.
Research evidence
Claims, sources, contradictions, and severe-test receipts
Checking KELLY proof surfaces
This lab renders live proof-linked items when KELLY exposes matching sector evidence.
Sources
Research inputs
- Crossref metadata
- arXiv
- Wikimedia analytics where relevant
- public research metadata
- KELLY contradiction and replay receipts
Proof standard
How users can tell
- Claims, sources, and conclusions are separate objects.
- Contradictions become research work, not embarrassment.
- Falsifiers are shown before strong claims are trusted.
- The PhD room must show what would disprove the invention.
Outputs
Research outputs
- evidence_packets
- contradiction_cards
- falsifier_candidates
- severe_test_receipts
- source_plan_updates
Storage posture
Postgres: central KELLY spine
Use canonical KELLY Neon/Postgres with science-scoped read models. This repo must not become an independent truth database.
Object storage
Immutable evidence shelf
Use R2/object storage for immutable paper snapshots, parsed artifacts, citation bundles, and evidence receipts.
Volume
No ground-truth volume
Optional local volume only for parser/model scratch cache. It must be rebuildable and never canonical.
Objects on the bench
Primary objects
- claim_object
- source_packet
- contradiction_card
- falsifier
Required receipts
Black-box recorder
- citation_receipt
- severe_test_receipt
- replication_posture
- closure_receipt
Authority
Write boundary
- KELLY is the only write authority for durable learning.
- Labs are read-only processors and inspection windows.
- Glass Lab remains the traceable evidence console.