01
Read
Start with the research paper and the Standard PDF.
Chris Royse / leapable.ai
Artifacts - paper, measurements, model card, Standard, systems
Open the proof files.
The paper, Standard, measurements, model card, system notes, and date trail sit here for inspection.
02
Measure
Then inspect the Case 3 measurement and the HuggingFace model card.
03
Trace
Use the date trail, citations, and source links to see when each claim became public.
Clean context beats a giant undifferentiated dump.
Watch on YouTube· 14:54
Focused data brains instead of one giant RAG database
This video explains the swappable-brain pattern: isolate corpora, preserve provenance, and search across bounded databases when needed. The artifacts page uses the same principle: source files stay inspectable instead of disappearing into a vague claim.
Watch videoOpen the full video on YouTubeNeurIPS 2026 Position Paper
NeurIPS 2026 Position Paper — Submission #64, OpenReview Forum ID `mpQXCwkQcq`.
Why this piece matters: the paper is public and independently reachable.
Inspect NeurIPS 2026 Position PaperCase 3 — measured voice-cloning result
0.961 mean WavLM-SECS (max 0.975, N=10, one speaker, English, within-WavLM-family). ECAPA-TDNN EXP-1 cross-encoder validation pending. Not a parity claim against any published TTS system.
Why this piece matters: one measured case is documented with scope guards.
Inspect Case 3 — measured voice-cloning resultContext Graph + ClipCannon
leapable.ai is the AI memory and marketplace surface. Context Graph: roughly 370K LOC Rust, 13 frozen embedders, 59 RocksDB column families, 75 MCP tools, and 78 pairwise cross-correlations. ClipCannon: roughly 67K LOC Python, 23-stage DAG, 58 MCP tools, and multimodal provenance records.
Why this piece matters: the primitive is backed by runnable infrastructure.
Inspect Context Graph + ClipCannonHuggingFace model card
HuggingFace model `cabdru/shakespeare-lora-gemma4` — open-source Case 1 with BSL license.
Why this piece matters: one model artifact is open for inspection.
Inspect HuggingFace model cardHave one context problem?
Send the audience, data type, target task, proof bar, and sharing limits. The first pass should stay bounded.