Proximal is a San Francisco-based (with a Bangalore presence) research lab for coding data, building high-fidelity, long-horizon reinforcement learning environments grounded in real codebases to train and evaluate frontier coding agents. It emphasizes scalable, software-driven data engines over human contractors, and research into reward-hacking detection and 'fuzzy verifiers' that score code quality beyond functional correctness.
How we verified this
Confirmed this is the correct entity matching the directory note 'RL environments with fuzzy verifiers': Proximal (proximal.ai), a SF-based research lab building high-fidelity long-horizon RL environments for coding agents, with explicit 'fuzzy verifier' and reward-hacking framing on its official blog. Founders Justus Mattern, Calvin Chen, and Navid Pour confirmed via the Scribble Ventures announcement. Open-source FrontierSWE benchmark (GitHub Proximal-Labs/frontier-swe, 127 stars; frontierswe.com) verified. Most financial fields correctly left undisclosed: total_raised, valuation, revenue all unknown; round amount undisclosed and the 'Seed' stage label is inferred, so softened. Crunchbase was inaccessible (403). Investor/angel list (OpenAI, Anthropic, Thinking Machines, DeepMind, xAI, Meta Superintelligence, Cursor, Cognition) is corroborated by founder Calvin Chen's own site but remains self-reported and unverified by third parties, kept at 'reported'. Headcount: LinkedIn supports only the 11-50 band; the '~25' figure downgraded to estimated. No named customers exist publicly, so notable_customers correctly stays empty/unknown, no frontier-lab customer ties are verifiable. Main overreach corrected was in researcher_backgrounds: the specific Fetchr / $1.5M ARR / $9M sale and 'second founding engineer at Cursor' claims are NOT corroborated by the founders' own sites and were flagged rather than asserted. No SOC2 or certifications; security page 404s. Overall confidence: medium.