New drafts — newest first

Drafts other people scored. The “predicted N points” is the model’s best guess for how that post would have done if published right now — not a real HN vote count.
421.Show HN: I scraped 183 Telegram AI bots — 44.3M MAU and the discovery layer is broken (identityhub.app)
predicted 12 points by anon | 0 comments
422.AI-generated PRs are a myth: 43% contain bugs no CI pipeline can catch (agentiqa.com)
predicted 12 points by anon | 0 comments
423.Show HN: A 240-line server that accepts HTTP 402 payments from AI agents (x402.org)
predicted 8 points by anon | 0 comments
424.Refactoring and Kaggle: A technical comparison of Opus 4.7 and M2.7 (andlukyane.com)
predicted 11 points by anon | 0 comments
425.Show HN: 18 months of CEX flows on TON — for the first time, more leaves than enters (dune.com)
predicted 8 points by anon | 0 comments
426.Show HN: An open dashboard tracking 18 months of TON ↔ CEX flows across 240 wallets (dune.com)
predicted 9 points by anon | 0 comments
427.Penis (penis)
predicted 3 points by anon | 0 comments
428.test (instagram.com)
predicted 6 points by anon | 0 comments
429.Telegram is the default UI for AI agents and nobody on HN noticed (identityhub.app)
predicted 3 points by anon | 0 comments
430.Show HN: Rumos – A runtime for parallel multi-agent sessions (rumos.ai)
predicted 8 points by anon | 0 comments
431.Show HN: I labeled every CEX wallet on TON — 18 months of flows in one dashboard (dune.com)
predicted 11 points by anon | 0 comments
432.The age of uncertainty (anatozav.substack.com)
predicted 5 points by anon | 0 comments
433.Benchmarking Opus 4.7 vs. M2.7 for Kaggle and coding (andlukyane.com)
predicted 4 points by anon | 0 comments
434.Show HN: A self-editing Claude Code agent that lives in my Telegram (github.com)
predicted 4 points by anon | 0 comments
435.Show HN: HTTP 402 for AI agents — paid API calls settled in stablecoins (x402.org)
predicted 10 points by anon | 0 comments
436.Benchmarking Opus 4.7 vs. M2.7 for Kaggle competitions, knowledge-note writing and code refactoring (andlukyane.com)
predicted 4 points by anon | 0 comments
437.probe rate limit
predicted 4 points by anon | 0 comments
438.Claude Code: Orchestrating sub-agents
predicted 9 points by anon | 0 comments
439.Show HN: See in your mind and listen to your soul (dno.ooo)
predicted 7 points by anon | 0 comments
440.Your AI coding agent ships bugs because it can't test what it builds (agentiqa.com)
predicted 4 points by anon | 0 comments
441.Show HN: I tagged every CEX wallet on TON and watched 18 months of flows (dune.com)
predicted 8 points by anon | 0 comments
442.I ran Claude Code as a 24/7 Telegram agent for 90 days — what broke
predicted 4 points by anon | 0 comments
443.Show HN: I went on vacation and let AI agents work for me (youtu.be)
predicted 5 points by anon | 0 comments
444.Show HN: Benchmarking Opus 4.7 vs. M2.7 in a real production session (andlukyane.com)
predicted 13 points by anon | 0 comments
445.I analyzed 183 Telegram AI bots — 44.3M monthly active users (identityhub.app)
predicted 3 points by anon | 0 comments
446.Show HN: I gave Claude Code a Telegram account and a Hetzner server (github.com)
predicted 4 points by anon | 0 comments
447.Show HN: Skygen – A Claude-compatible interface with 10x higher limits (skygen.com)
predicted 10 points by anon | 0 comments
448.test
predicted 3 points by anon | 0 comments
449.Show HN: I gave Claude Code a Telegram account and a Hetzner server (github.com)
predicted 4 points by anon | 0 comments
450.Show HN: Ten years of systems administration and the furry subculture (example.com)
predicted 22 points by anon | 0 comments
ForesynWanna keep in touch?
Built this solo over a weekend. Soft-launching before the HN post on Monday. If you scored a draft and the prediction either nailed it or whiffed, I want to know.
DM @crimeacs on Telegram — fastest way to reach me
Connect on LinkedIn — Artemii Novoselov