# trending-1005 evidence files

Everything behind tickfloor.com/research/trending-trading-techniques-tested and its five articles. Hypothetical historical backtests on cached data to
March 2025, not live results and not advice.

- prereg.md: the original pre-registration (5 October 2026), unchanged.
- results.json, charts.json, RESULTS.md: the original results as first published (random control v1), unchanged.
- prereg-amendment-2026-10-06.md: the dated amendment, committed before any corrected run. It describes the random-control
  defect and fix (control v2), three new cells, label corrections and what is still open.
- control-v2-null-check.json: the null check of the fixed control, run before the registered runs.
- results-amended-2026-10-06.json, charts-amended-2026-10-06.json, RESULTS-amended-2026-10-06.md: the control v2 numbers,
  superseded the same day.
- prereg-amendment-2026-10-06b.md: the second dated amendment, committed before any run under it. Control v2 placed each
  market's copies independently, which made the copies too easy to beat or lose to for rules that time many related
  markets at once; control v3 shifts the rule's own trades in time, the same shift for every market.
- random-control-sweep.json, random-control-sweep-per-asset.json: the calibration of control v3 on 600 coin-flip rules.
- results-amended-2026-10-06b.json, charts-amended-2026-10-06b.json, RESULTS-amended-2026-10-06b.md: the current numbers
  the pages show.
- cells.json, cells-amendment-2026-10-06.json: every cell with the sha256 of its rule file.
- rules/: each rule's exact source, named <cell id>.mjs. Its sha256 must match the cell files.
- engine/harness.mjs, engine/harness.test.mjs: the research harness at build time.
- SHA256SUMS: hashes of every file here.

Reproducing: the rules run on Tickfloor's research harness (research/hunt2) with cached Binance and US market data. The
harness and data cache are in Tickfloor's research repository; the cached bars are not redistributed here. With them,
`node --import tsx scripts/research/trending-1005-amend.mjs measure` re-measures every cell and fails loudly if any
number does not reproduce its ledger row.
