Frontier model audits 20k-commit FrankenSim project brutally — and the author loves it
doodlestein · x · 2026-11-10
Developer doodlestein fed his physics-simulation project FrankenSim (just crossed 20k commits) to Anthropic's latest frontier model for a 'reality check,' and got a brutal audit (screenshot):
- The repo is 'three programs that do not touch each other': L3 suite at 7,821 pass / 95 fail, coupled-multiphysics source grew 40k lines in 7 days with 4M lines untouched, 7+ demo engines with only 2 graded close
- Seven ranked flaws: no composition kernel (0.5% of 1.2M lines is generic mechanism), no decoupled path, copy-paste growth (6,828 commits added in 7 days, mostly copies), demos not driven by the core, no performance work on physics, stale truth surfaces, duplicated/misplaced code
- Verdict: 'The fix: compose, don't enumerate,' with a five-step kernel/port/bridge/driver remediation plan
The author calls it the 'Galaxy-Brained tough love' he needed, marveling that individual devs now get frontier models the same day as the biggest corporate users.
More from coding & agent
- Anthropic's S-1: 'agent' appears 177 times, 'SaaS' zero — the AWS playbook — brucemacv · 2026-11-10
- CoreWeave Launches Claude Code Skill to Build GPU Clusters and Serve Models from the CLI — altryne · 2026-11-10
- Designer ships portfolio built entirely with Claude Code, declaring the Design Engineer era — evilrabbit_ · 2026-11-10
- Stripe's claimable sandboxes: frictionless test accounts for AI agents, 50k+ provisioned daily — dl_weekly · 2026-11-10
- Simon Willison Hides AI-Generated PR Descriptions Behind Collapsed details Tags — tdhopper · 2026-11-10
- Ex-film composer vibe codes his dream ambient DAW in a week with Claude Fable 5.2 — yawnxyz · 2026-11-10