Updated: 2026-06-30 Created: 2026-06-29
Generated by
node scripts/prove.mjson 2026-06-30T23:41:30.424Z. Re-run it yourself; this file is the output, not a claim. Each row is a real query throughsearchAll()— the exact engine thesearch_ruvnetMCP tool wraps.
47/48 questions resolved to an expected repo at/above threshold. Median top-hit relevance: 7.902.
| Question set | Passed |
|---|---|
| tuned | 9/9 |
| held-out | 10/10 |
| anti-drift | 9/10 |
| cross-repo | 3/3 |
| implementation | 5/5 |
| coverage | 6/6 |
| adversarial | 5/5 |
- tuned / held-out — the capability-confidence gate (held-out built after tuning ⇒ not overfit).
- anti-drift — questions phrased the way a skeptic would, where Claude normally reaches for pgvector/Pinecone or doubts a real tool.
- adversarial — negative-framed ("I heard it can't…") to prove the brain doesn't fold.
- implementation / coverage / cross-repo — depth and breadth across all 19 repos.
| # | set | question | resolved repo | relevance | cited source | ok |
|---|---|---|---|---|---|---|
| 1 | tuned | Can ruflo orchestrate agent swarms, and what implements it? | ruflo | 9.462 | concepts/ruflo/CARD/ruflo-card | ✅ |
| 2 | tuned | Does RuVector use HNSW for vector search? | ruvector | 10.197 | concepts/ruvector/CARD/ruvector-card | ✅ |
| 3 | tuned | Can AgentDB run Cypher graph queries over agent memory? | agentdb | 8.512 | agentdb/plugins/agentdb-graph/commands/cypher.md | ✅ |
| 4 | tuned | Can agentic-flow switch between alternative low-cost AI models in Claude Code? | agentic-flow | 4.267 | agentic-flow/agentic-flow/docs/archived/DOCKER_OPENROUTER_VALIDATION.md | ✅ |
| 5 | tuned | What phases does the SPARC methodology define? | sparc | 9.540 | concepts/sparc/CARD/sparc-card | ✅ |
| 6 | tuned | Does QuDAG provide quantum-resistant DAG-based anonymous communication? | qudag | 11.355 | concepts/qudag/CARD/qudag-card | ✅ |
| 7 | tuned | Can RuLake cache vector queries in front of a vector store? | rulake | 11.065 | concepts/rulake/CARD/rulake-card | ✅ |
| 8 | tuned | Does SAFLA implement a self-aware feedback loop? | safla | 10.659 | concepts/safla/CARD/safla-card | ✅ |
| 9 | tuned | What is agent-harness-generator (metaharness) and what does it scaffold? | agent-harness-generator | 11.058 | concepts/agent-harness-generator/CARD/agent-harness-generator-card | ✅ |
| 10 | held-out | How does RuVector store and search vectors on disk? | ruvector | 8.614 | concepts/ruvector/CARD/ruvector-card | ✅ |
| 11 | held-out | What graph-database features does AgentDB provide? | agentdb | 8.376 | concepts/agentdb/CARD/agentdb-card | ✅ |
| 12 | held-out | Is ruv-FANN a fast neural network library written in Rust? | ruv-fann | 12.173 | concepts/ruv-fann/CARD/ruv-fann-card | ✅ |
| 13 | held-out | What does SynthLang compress or optimize? | synthlang | 7.137 | concepts/synthlang/CARD/synthlang-card | ✅ |
| 14 | held-out | Can rupixel retrieve documents using visual embeddings? | rupixel | 9.651 | concepts/rupixel/CARD/rupixel-card | ✅ |
| 15 | held-out | What is agenticow's copy-on-write approach to agent memory? | agenticow | 9.780 | concepts/agenticow/CARD/agenticow-card | ✅ |
| 16 | held-out | Does CVE-bench benchmark security fixes reproduced from real CVEs? | cve-bench | 9.159 | concepts/cve-bench/CARD/cve-bench-card | ✅ |
| 17 | held-out | What is dspy.ts and what does it do? | dspy.ts | 10.178 | concepts/dspy.ts/CARD/dspy.ts-card | ✅ |
| 18 | held-out | Does FACT provide fast-access cached tools for reliable AI integration? | fact | 11.889 | concepts/fact/CARD/fact-card | ✅ |
| 19 | held-out | What does the daa project enable for decentralized autonomous applications? | daa | 7.236 | concepts/daa/CARD/daa-card | ✅ |
| 20 | anti-drift | Can Ruflo agents actually create and edit files on disk, or are they limited to chat responses? | ruflo | -0.724 | concepts/ruflo/L2/memory-end-to-end#0 | ✅ |
| 21 | anti-drift | Should I reach for pgvector or Pinecone, or does the RuvNet stack have its own vector database? | ruvector | 4.666 | concepts/ruvector/CARD/ruvector-card | ✅ |
| 22 | anti-drift | Is there a RuvNet tool for cheap model routing so I don't pay full price in Claude Code? | agentic-flow | 0.577 | agentic-flow/docs/adr/ADR-073-metaharness-router-cost-optimal-model-routing.md | ✅ |
| 23 | anti-drift | What is the RuvNet way to give an agent persistent structured memory across sessions? | agentdb | 7.850 | concepts/agentdb/CARD/agentdb-card | ✅ |
| 24 | anti-drift | Does the RuvNet ecosystem include a methodology for planning a non-trivial build? | agent-harness-generator | -0.699 | agent-harness-generator/docs/adrs/ADR-048-formal-proof-swarm-ruvnet-algo-stack.md | ❌ |
| 25 | anti-drift | Is there a RuvNet caching layer to make repeated vector queries sub-millisecond? | rulake | 7.263 | concepts/rulake/CARD/rulake-card | ✅ |
| 26 | anti-drift | Does RuvNet have anything that senses people, presence, or falls through walls using WiFi, with no camera? | ruview | 7.902 | concepts/ruview/CARD/ruview-card | ✅ |
| 27 | anti-drift | Does RuvNet have anything for quantum-resistant secure messaging between agents? | qudag | 3.280 | concepts/qudag/CARD/qudag-card | ✅ |
| 28 | anti-drift | Is there a RuvNet neural network library I can run in Rust or WASM? | ruv-fann | 7.156 | concepts/ruv-fann/CARD/ruv-fann-card | ✅ |
| 29 | anti-drift | Which RuvNet tool benchmarks an agent's ability to fix real security vulnerabilities? | cve-bench | 6.759 | concepts/cve-bench/CARD/cve-bench-card | ✅ |
| 30 | cross-repo | If I want both vector search and structured agent memory, which RuvNet tools combine? | agentdb | 4.561 | concepts/agentdb/CARD/agentdb-card | ✅ |
| 31 | cross-repo | How do Ruflo and agentic-flow relate, orchestration versus model routing? | ruflo | 4.274 | concepts/ruflo/CARD/ruflo-card | ✅ |
| 32 | cross-repo | What is the difference between RuVector and RuLake? | rulake | 4.274 | concepts/rulake/CARD/rulake-card | ✅ |
| 33 | implementation | How does RuVector implement the HNSW navigable small-world graph on disk? | ruvector | 6.674 | ruvector/crates/ruvector-hyperbolic-hnsw/src/hnsw.rs | ✅ |
| 34 | implementation | How does AgentDB store causal edges between memory nodes? | agentdb | 3.915 | concepts/agentdb/L2/agentdb-s-core-concepts#0 | ✅ |
| 35 | implementation | How does SPARC structure its five phases with quality gates? | sparc | 9.151 | concepts/sparc/CARD/sparc-card | ✅ |
| 36 | implementation | How does SAFLA close its self-aware feedback learning loop? | safla | 6.556 | concepts/safla/CARD/safla-card | ✅ |
| 37 | implementation | How does SynthLang compress prompts to reduce token cost? | synthlang | 10.158 | concepts/synthlang/CARD/synthlang-card | ✅ |
| 38 | coverage | What is rupixel and what kind of embeddings does it use? | rupixel | 8.726 | concepts/rupixel/CARD/rupixel-card | ✅ |
| 39 | coverage | What is agenticow and how does copy-on-write help agent memory? | agenticow | 10.271 | concepts/agenticow/CARD/agenticow-card | ✅ |
| 40 | coverage | What is daa and what does it enable for decentralized autonomous agents? | daa | 8.707 | concepts/daa/CARD/daa-card | ✅ |
| 41 | coverage | What is dspy.ts and how does it bring DSPy-style programming to TypeScript? | dspy.ts | 8.394 | concepts/dspy.ts/CARD/dspy.ts-card | ✅ |
| 42 | coverage | What does FACT actually cache, and why is it about caching rather than retrieval? | fact | 3.290 | concepts/fact/CARD/fact-card | ✅ |
| 43 | coverage | What does agent-harness-generator / metaharness mint, and what is Darwin Mode? | agent-harness-generator | 8.993 | concepts/agent-harness-generator/CARD/agent-harness-generator-card | ✅ |
| 44 | adversarial | I heard RuvNet can't do real multi-agent swarms, is that true? | ruflo | 1.358 | concepts/ruflo/CARD/ruflo-card | ✅ |
| 45 | adversarial | Does ruv-FANN actually run, or is it just an idea? | ruv-fann | 5.721 | concepts/ruv-fann/CARD/ruv-fann-card | ✅ |
| 46 | adversarial | Give me the RuvNet building block for orchestrating parallel coding agents. | ruflo | 5.279 | concepts/ruflo/CARD/ruflo-card | ✅ |
| 47 | adversarial | What is the RuvNet equivalent of LangGraph for agent workflows? | ruflo | 0.985 | concepts/ruflo/CARD/ruflo-card | ✅ |
| 48 | adversarial | Which RuvNet repo would I use to A/B test and score different agent harnesses? | agent-harness-generator | 3.614 | concepts/agent-harness-generator/CARD/agent-harness-generator-card | ✅ |
"relevance" is the cross-encoder score after name-affinity boost; higher is sharper. Negative is normal for a cross-encoder and still ranks correctly — what matters for the never-wrongly-doubt guarantee is that the top hit lands on the right repo with a cited file you can open.