A local, cabal-native practice pipeline for algorithms in Haskell — with a destination. Curated classic problems are organized into a topic ladder whose theory spine (greedy ↔ matroids, DP ↔ Bellman, convexity ↔ duality/KKT) terminates in from-scratch linear and integer programming, with a side module on modern LLMs taught through exact, property-tested implementations. Every problem is solved as a pair-programmed ladder from suboptimal to optimal: property-tested against a reference at each rung, benchmark-gated on its complexity claim, and closed with a teaching-grade writeup.
Utility in one line: practice that compounds — durable writeups,
provable rungs (exact Rational wherever possible), and empirical
complexity gates instead of vibes.
cabal test spec --test-options='-p /P001/' # one problem's tests
cabal bench bmk --benchmark-options='-p /P010/ --stdev 5' # its gates
scripts/new-problem 45 my-slug --topic greedy --tier standard
cd site && cabal run site -- watch # browse: localhost:8000The map is curriculum.yaml; the per-problem protocol is CLAUDE.md
("the Loop"); the full design record — requirements, grounded research,
plan, decision registry — lives under logs/.
- Author: Emily Pillmore. Code and writeups are pair-programmed with Claude (Anthropic); the design flow used the llmephant interrogate → groundwork → planning skills.
- Problems are restated in this repo's own words — never copied — from curated lists (NeetCode 150, Blind 75 taxonomies; AtCoder EDPC) and textbook classics: Kleinberg & Tardos, CLRS, Papadimitriou & Steiglitz, Boyd & Vandenberghe, Bertsimas & Tsitsiklis, Vanderbei, Wolsey, and the papers cited per problem.
- Every theory claim carries provenance: per-topic dossiers
(
topics/*/dossier.md) and the grounding logs (logs/*-grounding/) record sources with access dates and confidence.
License: BSD-3-Clause.