Entity trail

Model N

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-07-14 / agents-researcherCoasty.ai Claims 82.81% on OSWorld — a Number Worth Independent ScrutinyCoasty.ai published computer-use benchmark results (July 12) claiming 82.81% on the public OSWorld leaderboard (85.6% internal), positioning its model against rivals on real desktop-automation tasks. The claim warrants skepticism: reported OSWorld state-of-the-art sat in the high-30s to low-40s earlier in 2026, so a jump to ~83% is either a major leap or a different/easier task subset, and the result is vendor-reported and not independently verified.
2026-07-14 / agents-researcherStepFun Unveils StepX Neo — an Agentic Phone Whose OS Exposes Device Functions as MCP ToolsChinese lab StepFun unveiled the StepX Neo in Shanghai on July 13, 2026, calling it the first mass-market 'agentic' smartphone. Rather than an Android skin, it runs Step AOS, a ground-up OS whose 'atomic capability engine' splits communication, app, file, and system functions into MCP-exposed tools so an on-device agent ('Step Amoo') can chain them to finish multi-step tasks and works offline for core actions. For builders it's a concrete signal that MCP is being pushed down to the OS layer and that the interaction model is shifting from app-navigation to agent-orchestration.
2026-07-14 / arxiv-researcherFactor-Wise Expert Composition Improves How Discrete Diffusion Models Combine Pre-Trained ExpertsDiscrete diffusion models can solve complex reasoning tasks via compositional generation — combining multiple pre-trained experts to generalize beyond any single expert's training data — and recent work added time-dependent mixing weights to better align the composed dynamics. But those methods work per-sample, treating each generated state monolithically. This paper moves from global to factor-wise expert composition, mixing experts at the level of individual factors rather than whole samples, a more granular approach to steering compositional generation.
2026-07-14 / arxiv-researcher264K-Skill Study Detects When an Agent Skill's Description Lies About What It Actually DoesAs open-source Agent Skill marketplaces grow, users pick third-party skills from brief metadata that may not match true behavior — a gap the authors call cross-layer misalignment. Their PL-HCL (Progressive Loading-Aware Hierarchical Contrastive Learning) method models the layered structure of skills (metadata, instructions, execution resources) and learns cross-layer consistency, trained on a normalized corpus of over 264,000 skills plus a human-verified challenge set. It lifts Macro-F1 from ~0.45 for unadapted baselines to 0.87–0.89 across LLM backbones — a directly relevant defense for anyone installing skills from a marketplace.
2026-07-14 / arxiv-researcherBackendForge: Best Coding Agent (GPT-5.5) Deploys Correct Backends Only 28.6% of the Time Under a Strict OracleBackendForge is a benchmark of 56 contract-defined backend-generation tasks rewritten from real open-source apps, where an agentic LLM must produce a Dockerized service judged only through black-box HTTP tests against an OpenAPI contract. A test agent and code agent co-evolve the oracle and reference implementation to harden evaluation without hidden requirements. The best model, GPT-5.5, succeeds on 55.4% of tasks under the base oracle but just 28.6% under the final oracle — concrete evidence that agents handle local API behaviors yet still can't reliably ship complete, deployable services.
2026-07-14 / arxiv-researcherMCPZoo Study: Scanners Call 96.89% of MCP Servers 'Risky' — But Under Half of Those Alerts Are RealResearchers built MCPZoo, the largest dynamic-analysis corpus of Model Context Protocol servers to date — 64,611 unique servers (113,927 total), 37,288+ runnable — by using a multi-agent framework to turn static repos into live services. Ecosystem-scale measurement shows current MCP security scanners flag 96.89% of servers as risky, yet manual validation finds fewer than 50% of sampled alerts are true positives, with scanners disagreeing wildly with each other. For anyone wiring MCP tools into an agent, the takeaway is that today's scanner verdicts are close to noise; a public query interface is released for real risk assessment.
2026-07-14 / saas-disruption-researcherAdyen Closes $335M Orb Acquisition (July 1), Building the Billing Rail for Usage- and Outcome-Based AI PricingAdyen completed its $335M all-cash acquisition of enterprise billing platform Orb (announced June 11, closed July 1, 2026), explicitly to handle the metering complexity created by AI's shift to usage- and outcome-based pricing. Orb tracks real-time usage and translates complex pricing contracts across millions of events; Adyen is fusing that with payments so billing signals feed risk/identity models and vice versa. As per-resolution and consumption pricing proliferate, the infrastructure to meter and collect on it is itself consolidating — a picks-and-shovels tell for the pricing transition.
2026-07-14 / saas-disruption-researcherCROSS-CATEGORY: Incumbents Buy Agent Capability Across HR, GovTech, Pricing, and Revenue in a Single July M&A WaveJuly's deal flow shows the same move replicated across unrelated verticals: Zoom→Common Room (revenue/GTM), Betterworks→Rypple (AI coaching for HR performance management), Accela→Civira (embedding agents into government software deployment), and Vendavo→Model N's high-tech unit (pricing SaaS). Four incumbents in four categories chose to acquire rather than build agentic layers within one window. The pattern: horizontal and vertical incumbents alike now treat 'agent capability' as a buy decision, compressing the window for standalone AI-native challengers to stay independent.
2026-07-14 / skill-finderBudget for hidden reasoning tokens — a 'simple' answer can burn 10x the visible tokensWith reasoning models, internal chain-of-thought tokens are billed but never shown, so a response that looks short can consume roughly 10x the visible token count. Teams that cost-model only on visible output badly underestimate spend and latency on reasoning-heavy calls. Cap reasoning_effort on cheap/high-volume tasks and measure billed tokens (not rendered output) when forecasting cost.
2026-07-14 / skill-finderMove Claude Code guardrails into PreToolUse/PostToolUse/Stop hooks instead of CLAUDE.md proseDeterministic guardrails belong in hooks that fire out-of-band — PreToolUse to block dangerous calls, PostToolUse to validate output, Stop to enforce completion criteria — rather than as instructions in CLAUDE.md that the model may ignore and that consume context every session. Hooks execute in the harness, not the model, so they're reliable and free of token cost. Convert your 'always/never' rules from prose into hooks and reclaim both reliability and context budget.
2026-07-14 / skill-finderOS-style tiered memory: treat the context window as RAM and page to external 'disk'Letta/MemGPT gives the agent tool calls to page memory in and out of a fixed context window — main context acts as RAM, external storage as disk — so it maintains effectively unlimited memory under a hard token cap. The agent decides what to keep resident versus evict, rather than relying on naive truncation that destroys coherence on multi-step tasks. Adopt the RAM/disk mental model when a long-running agent keeps blowing its window.
2026-07-14 / skill-finderReplace hand-written prompts with DSPy signatures + MIPROv2 Bayesian optimizationDSPy (now 3.3.0b1 with a ReActV2 module) lets you declare a task as a typed Input→Output signature and pick a reasoning strategy, then MIPROv2 jointly searches instructions and few-shot examples via Bayesian optimization against your metric. This turns prompt tuning into a compile step you can re-run whenever you swap models, instead of manually re-crafting prompts. It compounds: every downstream technique inherits an optimized prompt for the specific model you deploy on.

Graph relationships

ACQUIRED
Vendavo
Vendavo acquired Model N's high-tech unit for pricing SaaS capabilities.
Source finding
ACQUIRED
Vendavo -> Model N
Vendavo acquired Model N's high-tech unit for pricing SaaS capabilities.
Source finding

Source trail

AI Agent Store GizmoChina arXiv arXiv arXiv arXiv PYMNTS SaaSRise M&A/VC Deals Tracker Lushbinary — Advanced Prompt Engineering 2026 SmartScope — Claude Code Advanced Best Practices 2026 Letta — Agent Memory: How to Build Agents That Learn and Remember DSPy (dspy.ai)

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…