Science and Research Lab · Answered directly by KELLY · public writes blocked
Science and Research Lab
KELLY's evidence, citation, contradiction, replication, and falsification organ.
Turns external research into source-bound evidence packets, contradiction pressure, falsifiers, and research receipts that other KELLY organs can inspect.
Samples Glass Lab, Advanced Processing Evidence Spine, and the lab-specific runtime routes every 15 seconds.
Glass panels
0/2
Snapshot
0/?
Routes
linked
Refresh
15s
This lab uses anchored Glass Lab proof routes only.
Science & Research Workbench
Evidence digestion lens — support/contradiction only, not ground truth.
- • Contradiction radar is inspect-only until durable evidence rows attach.
- • Demo/fallback packets must stay labeled — never treated as live neon.
- • Public users cannot write prime memory from this surface.
- • Gauges follow GET event-bus status (process-local Map), not multi-replica truth.
- • Claims list may serve process-local Map rows when non-empty; else demo_fallback — never multi-replica.
- • CLI Map seed scripts write a separate process Map and do not hydrate this engine HTTP list.
- • Claim→hypothesis links are test-plan edges only; contradiction radar is inspect-only — not settlement.
- • Claim strength is a scoped inspect lens — not settlement grade or receipt.
Event bus status
Loading event-bus status…
Map stores status
Loading map-stores status…
Organs glass chain
Loading organs chain…
process_is_not_truth · never uses terminal “blocked” · Glass mission control only
APES organ rollups
Loading organ rollups…
instrument rollups · labeled_unproven · never demo-as-live · process_is_not_truth
Glass Lab visibility
Loading visibility pack…
four panels · inspect-only · never terminal “blocked” · process_is_not_truth
Cognition Organs V1
Loading organs V1 completion…
ladder=n/a · neon_any=no · instrumented=no
envelope · neon-first labeled · no direct writes · process_is_not_truth
Operator command ladder
Loading operator ladder…
never auto-deploy · process_is_not_truth · missing_evidence not “blocked”
Program status
Loading program status…
process_is_not_truth · never terminal “blocked” · public lens only
Organ honesty summary
- ✓ Map stores process-local
- ✓ CLI seed ≠ engine HTTP
- ✓ Claim→hyp links not truth
- ✓ Radar inspect-only
- ✓ Severe-test inspect-only
- ✓ Claim strength not settlement
- ✓ memory_candidate ≠ written
- ✓ Falsifiers not proof of kill
- ✓ Closure pending ≠ promotion
- ✓ lineage_view not write
- ✓ birth_chain not settlement
Standing law checklist — never multi_replica_ready; never public write; process is not truth.
Cognition workbench
This is the sector-specific operating view for Science/Research: gauges, receipt queue, closure timeline, topology, and empty-state behavior. Live route health remains in the real-time panel above.
Claim map
support / contradict
Every research answer separates supporting evidence, contradiction pressure, and missing replication.
Citation state
source-bound
Papers, APIs, and research packets must carry source metadata before they can influence a claim.
Severe test
falsifier-first
The lab should ask what would weaken the claim before treating support as useful.
Handoff
to hypothesis
Closed or contested evidence becomes a hypothesis input, not a hidden truth write.
Receipt queue
The queue is a receipt path, not a claim of hidden live data.
Source intake
collectApproved paper, citation, API, or research metadata is captured with provenance.
Contradiction card
compareSupport and contradiction are recorded as separate evidence lanes.
Falsifier packet
testA severe-test or refutation candidate is attached before promotion.
Research closure
handoffEvidence closes as support, weak support, contradiction, or open question.
Lineage and closure timeline
Claim
capturedState the research claim and boundary.
Source
indexedAttach DOI, API, citation, or provider metadata.
Conflict
searchedLook for contradiction and replication posture.
Falsifier
plannedName the evidence that would weaken the claim.
Receipt
readyEmit source-plan and severe-test receipts.
Cognitive topology
Claim
Plain claim with scope and uncertainty.
Sources
Citation and provider evidence lanes.
Contradictions
Conflicting evidence and replication limits.
Falsifier
Severe test that can weaken the claim.
Hypothesis
Downstream question or source plan.
Viewport empty-state contract
No research packet selected
If live research panels are empty, this lab must show a waiting state and route the operator to source intake rather than inventing support.
Open research queueLive inputs
Active objects
Receipts
Closure status
Automation loop
Runs as a continuous research triage organ from Glass Lab and standalone lab runtime snapshots.
step 1
Collect source metadata from approved research lanes.
step 2
Separate support, contradiction, replication posture, and uncertainty.
step 3
Emit falsifier candidates and severe-test receipts.
step 4
Hand eligible evidence packets to Hypothesis Lab and Glass Lab evidence spine.
Glass Lab source
These are the full proof windows this lab samples into the one-page view.
Storage posture
Postgres
Use canonical KELLY Neon/Postgres with science-scoped read models for research_sources, evidence_packets, citation_edges, contradiction_cards, and severe_test_receipts.
Object store
Use R2/object storage prefixes for paper snapshots, parsed artifacts, citation bundles, and immutable evidence receipts.
Volume
Optional app-local cache volume only for parser/model scratch files. It must never be the source of truth.
Runtime routes
Lab functions stay simple on this page; deeper receipts open in the exact runtime or Glass Lab window.