v0.3.0 — algorithm discovery + discoverability revamp
v0.3.0 — the multi-language solving engine gets a verified local algorithm synthesizer, plus a discoverability revamp.
New: local algorithm synthesizer (src/synth)
- Grammar-based scanner search + strategy templates (
vote,seen,fib,circular-kadane) rediscovers and verifies algorithms from curated I/O examples. - Smoke run: 7 targets · 7 verified · 0 rejected — max_subarray rediscovered; best-time-buy-sell, jump game, contains-duplicate, majority-element, climbing-stairs surfaced as new-to-catalog.
- Every candidate is fuzz-verified (60+ inputs) against an independent oracle before reporting.
uv run python -m synth discover [--smoke]orpython engine/runner.py discover; outputs tocatalog/discoveries/.
Repertoire (from 0.2.0)
- 10 algorithms + 6 data structures solved, tested, and benchmarked in Java · C++17 · Rust · Python.
- Single
catalog/problems.jsonsource of truth; byte-identical vector sets;--checkkeeps them in sync. - Pattern-discovery framework (
algo_discovery, 10 hypotheses, 65 tests).
Discoverability & packaging
- 18 GitHub topics, keyword-rich description, README with hero banner, mermaid diagram, benchmarks, and dev guide.
- GitHub Pages landing site (
docs/) deployed to https://dsk-dev-ai.github.io/algorithm-discovery-engine/ - GitHub Sponsors funding config (NextGenAI Labs) + custom social-preview banner.
- PyPI metadata: 18 keywords, version 0.3.0.
Checks: ruff · mypy (strict) · 118 pytest · runner test/bench/discover — all green on CI (3.10–3.13).