QuantProp. Analyze a trade file ↗
Home / Research archive / Coinback research archive
coinback / Experiment result

E-24. RSI 24h long-only independent val AggTrade check (2026-07-15)

69 independent val signals replayed with real AggTrade fills: all markouts negative, so the RSI 24h maker family was rejected.

Original record: 2026-07-15 · Migrated: 2026-10-27 · Translated from Korean: 2026-10-27
A historical record, not rerun in this migration.

The figures, hypotheses and verdicts of the time are preserved. A past positive result or candidate mention does not mean validated profitability today or a live-trading recommendation. Validation periods that were already used are not treated as new independent validation.

This is an English translation of a preserved Korean original. It describes conditions at the time of writing, not current provider specifications, costs or regulations. Phrases in the original about blocking or avoiding losses interpret simulated experiments and are not realized account results. Statistical trial counts differ from the number of independent strategies or CSV rows.

E-24. RSI 24h long-only independent val AggTrade check (2026-07-15)

  • Fixed rules: majority macro, RSI reclaim, long-only, offset0, latency250ms, TTL20m, maker entry/taker exit, 24h. All 69 non-overlapping signals in val.
  • F1 68 fill(98.55%), -0.19084%/signal. F2 65 fill(94.20%), -0.20065%/signal. Every 1/5/30/60-second median markout was also negative.
  • Verdict D-030: family rejected. The train long selection did not reproduce in the independent period, and Shadow candidate pass=false. The 269 official high-resolution signal records and failure codes must not be deleted.

Source and migration evidence

Source: backtest/docs/experiment-log.md#E-24

Original SHA-256: 069c4aaaf53a20930ca8d94ba6365b4ce2b6bc560171d662a4c179ea55a6857c

Raw data and account information are not part of the public package. The original results do not replace statistical or scientific revalidation. The English text is a translation; numbers, tables and charts are carried over unchanged.

Linked records

Evidence and follow-up records

archive / hub

Coinback research archive

E-01 through E-25 and the surrounding research record, including execution-cost, funding, maker replay and liquidation-data diagnostics.

READ THE EVIDENCE
archive / hub

The record behind the research.

Stockback and Coinback research records in English, translated from the Korean originals and preserved with source hashes and explicit historical context.

READ THE EVIDENCE
methodology / hub

Evidence has a chain of custody.

Every result should tell you what was known, what was assumed, and what remains untested.

READ THE EVIDENCE
policy / policy

The assumptions belong beside the result.

Research, education and tools are not personalized investment advice.

READ THE EVIDENCE
archive / article

E-23. RSI 24h maker official AggTrade expansion to 200 signals (2026-07-15)

Expanding a Binance AggTrade limit-order replay to 200 signals: 6/9 scenarios passed the pilot, long/short asymmetry carried to independent val.

READ THE EVIDENCE
archive / article

E-25. Measuring detection power of candle proxies vs real liquidation clusters (2026-07-16)

Measuring a candle-based cascade proxy against 53,398 real COIN-M liquidations: 69–93% precision but 81–95% of cascades missed.

READ THE EVIDENCE