buildOne report1 publisher ProvenanceGuard checks whether an agent's claim came from the tool it cited
Multiverse Computing published ProvenanceGuard on September 29, 2026, a layer that checks which MCP tool each of an agent's claims came from. It flags facts that are true somewhere in the pooled evidence but credited to the wrong tool.
Reality
- Evidence30
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence30
buildOne report1 publisher ProvenanceGuard flags MCP agent answers that credit a true fact to the wrong tool
Multiverse Computing's ProvenanceGuard caught 138 of 139 bad claims experts flagged in a medical MCP agent by checking which tool each fact came from. A source-blind check passes a true fact credited to the wrong tool, so the verifier needs traces that keep each output tagged with its source.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+15
- Incentives65
- Confidence55
buildOne report1 publisher Multiverse's block-pruning method widens its MMLU lead over block-influence pruning as compression grows, reaching 22.9 points at 50%
Multiverse Computing ranks candidate prunings by the energy of an Ising glass, so a single GPU scored about 29 billion combinations in two days. The paper reports the older per-block heuristics do just as well at lighter compression.
Reality
- Evidence50
- Adoption
- Not scored
- Hype gap+25
- Incentives75
- Confidence55
Bonsai 2 27B keeps 98 percent of Qwen3.8 27B's aggregate benchmark score, up from 95 percent for PrismML's first release, and TechCrunch reports the 5.9GB file is small enough for a PC and possibly a high-end phone.
Reality
- Evidence44
- Adoption38
- Hype gap+33
- Incentives71
- Confidence55
buildOne report1 publisher Escalating retries recover the 7,930 refusal prompts a single steering pass drops
Multiverse Computing's paper treats a deployment's refusal set as a subset of politics rather than the whole topic, which changes what the training corpus has to contain before any model is trained. The posted text breaks off before the results.
Reality
- Evidence42
- Adoption12
- Hype gap−20
- Incentives65
- Confidence58
Forty new unicorns in July, and 15 of them are under three years old
July was the biggest month for new unicorns in more than four years, and H1 alone had already beaten the whole of 2025. One company priced at $1.1 billion is less than a year old.
Reality
- Evidence62
- Adoption45
- Hype gap+18
- Incentives58
- Confidence55