Mission Control
Interface design study. Every figure on this screen is illustrative sample data.
Entities
48,712
+342 last 24h
Claims
312,488
+1,847 last 24h
Open controversies
47
+5 since Mon
Avg query latency
94ms
−12ms vs last week
Claims written per day
Last 25 days · all sources
healthy
graphiti
Books
142,816
Papers
98,204
YouTube
52,941
Social
18,527
Active ingestion runs
3 running · 12 queued
live
Extract · "Thinking Fast and Slow.pdf"
NER + relation extraction
Entity resolution against canonical store
Resolving "Kahneman, D." → Daniel Kahneman (entity_38219) · 0.97 confidence · embedding match
Graph write + contradiction scan
Vector index update
Graph Explorer
Subgraph: Andrej Karpathy · scaling sufficiency · depth 2
cypher
contested neighborhood
Entity
Concept
Claim (current)
Claim (superseded)
Source
claim
C₂ "scaling has limits"
Truth value
Strength
0.84
Confidence
0.91
Outgoing edges (3)
subject→ Andrej Karpathy
object→ scaling-sufficiency
supersedes→ C₁ "scaling is sufficient"
Incoming edges (2)
asserts← Lex podcast · 2024
controversy← Controversy_2891
Bi-temporal window
valid_from2024-02-10
valid_toopen
tx_from2024-02-12 09:14
last_seen2026-05-08
Why retrieved
Match on subject entity (Andrej Karpathy). Highest-confidence current claim in this neighborhood. Bi-temporal cursor is now; C₂'s validity window is open.
Contradictions Queue
47 open · 12 awaiting human review · 8 auto-resolvable
47 open
Optimal learning rate for transformer fine-tuning
"3e-5 is the right starting point for most fine-tuning workloads."
HuggingFace docs · 2024
str 0.78 / conf 0.84
"5e-5 to 1e-4 is appropriate depending on dataset scale."
Karpathy lecture · 2024
str 0.71 / conf 0.82
Did Hinton co-found OpenAI?
"Geoffrey Hinton was a co-founder of OpenAI."
Reddit thread · 2023
str 0.21 / conf 0.32
"Hinton was at Google Brain; OpenAI was founded by Altman, Musk, et al."
OpenAI official history · 2024
str 0.97 / conf 0.99
"RNNs are obsolete" vs. "RNNs are the right choice"
"RNNs are obsolete for long sequences."
NeurIPS tutorial · 2023
str 0.78 / conf 0.71
"RNN-style architectures (Mamba, RWKV) are the right choice for long contexts."
Mamba paper · 2024
str 0.84 / conf 0.79
Retrieval Lab
Hybrid retrieval comparison · vector × graph × BM25
α=0.4 / β=0.4 / γ=0.2
Vector · cosine
k=5 · 38ms
1. "Scaling laws have produced impressive results…" 0.892 · GPT-2 talk · 2019
2. "I no longer believe pure scaling is sufficient…" 0.871 · Lex podcast · 2024
3. "Structured world models are increasingly necessary…" 0.847 · YT lecture · 2024
4. "Compute and data are the dominant variables…" 0.812 · OpenAI blog · 2020
5. "The bitter lesson still applies but..." 0.798 · Twitter · 2022
Graph · traversal
3 hops · 12ms
1. Karpathy → BELIEVES → world-model-importance (current) str 0.87 conf 0.88
2. Karpathy → BELIEVES → scaling-sufficiency (superseded 2024) str 0.32 conf 0.78
3. Controversy_2891 → competing positions on scaling 2 positions · stable
4. Karpathy → AUTHOR_OF → "Software 2.0" → MENTIONS → world-models indirect
BM25 · lexical
k=5 · 18ms
1. "…karpathy noted that scaling alone has structural limits…" 14.8 · podcast transcript
2. "…karpathy's structured world models lecture clarifies…" 13.2 · review article
3. "…scaling vs. structure: karpathy responds…" 11.9 · blog comment
4. "…karpathy: scaling is necessary but not sufficient…" 10.4 · tweet
Merged result (hybrid rerank)
DELIVERED TO AGENT
Primary claim: Karpathy currently emphasizes that pure scaling has structural limits and structured world models are increasingly necessary. [str 0.87, conf 0.88]
Historical context: An earlier (2019) position emphasized scaling sufficiency; that claim's validity window was closed in Feb 2024 when superseded.
Active controversy: The broader field is divided on the exact role of structured priors. Controversy_2891 holds two stable positions.
Recommended hedging: present-with-disagreement · cite both the historical evolution and the active dispute.
Historical context: An earlier (2019) position emphasized scaling sufficiency; that claim's validity window was closed in Feb 2024 when superseded.
Active controversy: The broader field is divided on the exact role of structured priors. Controversy_2891 holds two stable positions.
Recommended hedging: present-with-disagreement · cite both the historical evolution and the active dispute.
Recent workflow runs
Last 12 · sorted by completion
12 active templates
| Run | Template | Status | Duration | Cost | Outputs | Owner |
|---|---|---|---|---|---|---|
| run_8a92f1 Ingest · Kahneman corpus |
book-ingestion-v3 | complete | 14m 32s | $2.84 | 2,841 entities · 4,127 relations · 0 contradictions | knowledge-eng |
| run_8a92e7 Meme variants · Hayek thread |
meme-generation-v2 | complete | 3m 14s | $0.42 | 12 variants · 4 approved | meme-shaman |
| run_8a92db Episodic update · client_847 |
session-ingest-v1 | running | 02:14… | $0.18 | none | system |
| run_8a92d4 Storyboard · launch hero |
storyboard-v4 | awaiting review | 22m 08s | $11.40 | 9 frames · 3 character refs | creative-dir |
| run_8a92c1 Retrieval eval · weekly |
retrieval-bench-v2 | complete | 8m 47s | $1.22 | hit@5 = 0.84 · MRR = 0.71 | system |
| run_8a929b Contradiction sweep |
contradiction-engine-v3 | complete | 1m 12s | $0.08 | 5 new flagged · 2 auto-resolved | system |
| run_8a9284 Voice · episode 14 narration |
voice-synthesis-v2 | failed | 1m 04s | $0.00 | elevenlabs · rate limit | creative-dir |
Source health
Top 8 by claim volume
| Source | Type | Credibility | Last ingest | Claims |
|---|---|---|---|---|
| Andrej Karpathy YouTube + Twitter + blog | creator | 2h ago | 2,841 | |
| Ben Goertzel papers + podcasts | researcher | 1d ago | 1,422 | |
| Dave Shapiro YouTube | creator | 4h ago | 3,108 | |
| arXiv · cs.LG paper feed | papers | 30m ago | 18,204 | |
| Lex Fridman podcast transcripts | podcast | 3h ago | 4,892 | |
| Kahneman corpus books · 8 titles | books | 14m ago | 2,841 |
Why this happened?
Recent agent explanations
provenance log
RETRIEVAL · 14:21:08
Why was C₂ surfaced over C₁?
Bi-temporal cursor = now. C₁'s validity window closed 2024-02-10 via SUPERSEDES edge from C₂. C₂ is the only currently-valid claim with matching predicate.
CONFIDENCE · 14:19:32
Why is confidence on C_4291 only 0.62?
2 of 5 supporting sources contradicted by 3 other credible sources. PLN propagation reduced confidence from 0.85 → 0.62.
REJECTION · 14:14:51
Why was meme variant 7 rejected?
Brand-fit score 0.31 (below 0.6 threshold). Tone mismatch flagged by humor-map classifier. Variant 4 substituted.