sandbox surface
/sandbox/microstructure→ openagent runs · 0
— no agent run cites thisstrategies3 strategies2 sandbox1 agent
Strategy catalog.
Every strategy with a citable surface on the platform — deduped across the two sources of truth: sandbox decls(2, hand-wired demonstrations under /sandbox/*) and agent-proposed runs(1, ADR-0040 captures). For the multi-seed verified A/B across all 15 arms, see/strategies/compare. For pairwise feature/citation/run overlap of any two strategies, see /strategies/diff. For the framework that runs research workflows like these, see/projects/alpha-bench.
3 strategies · deduped
- sandboxdetail →
xs_momentum_demo
sandbox surface/sandbox/portfolio→ openagent runs · 0— no agent run cites this - sandbox surface— no sandbox declfeatures · 2 · ordered by leaderboard rankcites · 8 notes
show all 8 notes ▾
- noteAI research agents as platform citizens
- noteMLOps for quant research isn't MLOps for ML
- noteObservability for alpha pipelines: three dashboards, one rule
- noteRight-sizing the research data platform: five thresholds, in order
- noteThe citation graph as substrate
- noteThe deploy contract isn't a YAML file
- noteThe promotion gate: why bad data should be unreachable
- noteWalk-forward without leakage: a checklist that's saved me
how this catalog is wired
buildStrategyCatalog() insrc/lib/strategyCatalog.ts walks both sources of truth and folds them into one row per strategy slug. Dedup rule: sandbox wins the displayed origin tag, agent runs still contribute run ids + cited notes + features.
today:the card carries origin + sandbox link + agent-run pills (each pill shows the run's objective score) + features + cited notes — the citation-graph surface.future:wires live PnL share + IC realized vs predicted + drift through /v1/strategies/<slug>/health(/v1/strategies/compare/<arm>already lives; ADR-0059 sketches the rest).