References¶

Rule R6 (CLAUDE.md): every algorithmic decision cites a paper or a documented OSS implementation. This page is the consolidated index.

Scoring and calibration¶

Brier, G. W. (1950). Verification of forecasts expressed in terms of probability. Monthly Weather Review, 78(1), 1–3. DOI

Murphy, A. H. (1973). A new vector partition of the probability score. Journal of Applied Meteorology, 12(4), 595–600. DOI Basis for the reliability / resolution / uncertainty decomposition used by decompose_brier and decompose_brier_sql.

Naeini, M. P., Cooper, G., & Hauskrecht, M. (2015). Obtaining well calibrated probabilities using Bayesian binning. AAAI. arXiv ECE binning convention.

Kumar, A., Liang, P. S., & Ma, T. (2019). Verified uncertainty calibration. NeurIPS. arXiv Discusses ECE's bias and bin-sensitivity. K-Fish uses 10 bins for legacy parity; acknowledged as a coarse metric.

Conformal / quantile calibration¶

Vovk, V., Petej, I., & Fedorova, V. (2015). Large-scale probabilistic predictors with and without guarantees of validity. NeurIPS. arXiv Venn-Abers predictors and the inductive (IVAP) variant we use via venn-abers package.

Vovk, V. (2022). Conformal e-prediction. arXiv Background on conformal prediction theory.

Romano, Y., Patterson, E., & Candès, E. (2019). Conformalized quantile regression. NeurIPS. arXiv Mondrian-style stratified conformal — K-Fish stratifies by (category, true_class).

Cauchois, M., Gupta, S., & Duchi, J. (2021). Knowing what you know: valid and validated confidence sets in multiclass and multilabel prediction. JMLR. arXiv

Wisdom of crowds and extremization¶

Galton, F. (1907). Vox populi. Nature, 75, 450–451. The original crowd-median result.

Tetlock, P. E. (2005). Expert political judgment: How good is it? How can we know? Princeton University Press. Foundational. K-Fish personas draw from Tetlock's "fox vs hedgehog" typology.

Baron, J., Mellers, B. A., Tetlock, P. E., Stone, E., & Ungar, L. H. (2014). Two reasons to make aggregated probability forecasts more extreme. Decision Analysis, 11(2), 133–145. DOI Basis for the asymmetric extremization formula in agents.aggregator.asymmetric_extremize.

Atanasov, P., Rescober, P., Stone, E., Swift, S., Servan-Schreiber, E., Tetlock, P., Ungar, L., & Mellers, B. (2017). Distilling the wisdom of crowds: Prediction markets vs. prediction polls. Management Science, 63(3), 691–706.

Schoenegger, P., Tuminello, P., Karger, E., & Tetlock, P. (2024). Wisdom of the silicon crowd: LLM ensemble prediction capabilities rival human crowd accuracy. Science Advances. arXiv Direct basis for the 9-persona LLM-ensemble design.

Korean NLP¶

Lee, M.-h. Kiwi: Korean morphological analyzer. GitHub / kiwipiepy The tokenizer K-Fish uses for FTS indexing.

Similarity hashing¶

Charikar, M. (2002). Similarity estimation techniques from rounding algorithms. STOC. ACM SimHash origin. K-Fish uses the 64-bit variant with Hamming ≤ 4 over a 48h window.

Manku, G. S., Jain, A., & Sarma, A. D. (2007). Detecting near-duplicates for web crawling. WWW. ACM SimHash for large-scale dedup — the pattern we implement.

Translation¶

Papago NMT. Papago developer docs. The Korean-to-English machine translation API.

LLM clients¶

Anthropic Claude. API docs. Primary forecaster (ADR-0005).

OpenAI GPT. API docs. Independent evaluator (ADR-0005).

Data platform¶

Raasveldt, M., & Mühleisen, H. (2019). DuckDB: An embeddable analytical database. SIGMOD. ACM The warehouse substrate.

DuckDB Labs. The ASOF JOIN documentation. docs

Telegram / trading stack¶

aiogram 3. docs. Async Telegram Bot API framework. ADR-0003 rationale.

Hyperliquid. docs. Per-user agent wallet pattern (ADR-0004).

Observability¶

Langfuse. docs. Self-hosted LLM tracing (ADR-0007).

Software architecture references¶

Nygard, M. T. (2018). Release It! Design and Deploy Production-Ready Software (2nd ed.). Pragmatic Bookshelf. Patterns for failure-tolerant pipelines; K-Fish nightly uses bulkheads per-step.

ArchitectureDecisionRecord. MADR 4 spec. GitHub. The ADR format used in docs/decisions/.

Legal (see separate Legal section)¶

가상자산이용자보호법 (VAUPA) — Act on the Protection of Users of Virtual Assets. Effective 2024-07-19.
특정 금융거래정보의 보고 및 이용 등에 관한 법률 (특금법) — the KoFIU VASP registration regime.
형법 제246조, 제247조 — Criminal Code, gambling and operating gambling places.

Full statute citations and commentary in the private runbooks/kr-legal-brief.md; public-safe summary at Korean Legal Status.