Skip to content

The double bottom joins the nightly scan (#126) - #128

Merged
yanivil merged 1 commit into
mainfrom
feat/activate-double-bottom
Sep 17, 2026
Merged

yanivil merged 1 commit into
mainfrom
feat/activate-double-bottom

Conversation

@yanivil

@yanivil yanivil commented Sep 17, 2026

Copy link
Copy Markdown
Owner

Summary

The double bottom joins the nightly scan, on the evidence of #126.

Evidence (the eleven yearly point-in-time runs with all four patterns; the tuning page has the tables)

  • Tuned rules: 215 signals in eleven years, 58 % reaching the target, +0.13 R per signal, month-block interval [+0.00, +0.26], positive in 7 of 11 years, max drawdown 8 R. The reward:risk floor keeps the form with a higher second low.
  • Reward floor off: 752 signals at +0.10 R, interval [+0.01, +0.18], 9 of 11 years; the higher-second-low form +0.13 R [+0.04, +0.22] on 424.
  • Incremental: 7 of the 215 sit within 15 days of an inverse H&S signal on the same ticker. The other three patterns reproduced their known figures in the same runs.
  • Against the gate it passes narrowly: the thinnest edge adopted so far, about 20 signals a year at half the H&S's per-signal expectancy. Review after the first 30 live signals.

Changes

  • ACTIVE_PATTERNS gains "Double Bottom"; the report's age and listing limits and meta follow (8 bars, 6 sessions).
  • The page explains the pattern in plain words and quotes its eleven-year figure next to the others.
  • The replay's wait_bars and depth_atr features read the W's notes (L2 is its last anchor; depth is peak minus the lower low), and _drop_reason names "second low".
  • Tests (164): the mini universe gains a double bottom, so the end-to-end scan, the second and third runs, the report's age table and the backtest see four patterns; the random-walk false-positive bound of the scan moves from 5 % to 8 %, since the fourth detector fires on about 4 % of random walks (the other three on 2 %; the pattern's 10 % rise rule is what holds it there).
  • Docs: wiki 02 (status), wiki 03 ("The double bottom, tested", the constants' header, the ledger of adopted values), wiki 04, README (four patterns), changelog.

Closes #126.

Test plan

  • ruff clean, 164 tests pass locally
  • tests workflow and the backtest smoke run on this PR
  • tonight's scan: the report may list a Double Bottom row; the page explains it under "Why these stocks"

🤖 Generated with Claude Code

ACTIVE_PATTERNS gains the Double Bottom after the eleven yearly point-in-time
runs with all four patterns: 215 signals under the tuned rules, 58 %
reaching the target, +0.13 R per signal, interval [+0.00, +0.26], positive
in 7 of 11 years, 3 % overlap with the inverse H&S; 752 signals at +0.10 R
[+0.01, +0.18], 9 of 11 years, with the reward floor off. Thin but
incremental; review after 30 live signals. The page explains the pattern
and quotes its figure; the test universe gains a double bottom (counts,
age tables and ticker sets updated); the random-walk false-positive bound
of the scan moves from 5 % to 8 %; the replay's anchor and depth features
read the W's notes. Wiki 02 and 03 (the tested section, the ledger),
wiki 04, README, changelog.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@yanivil
yanivil merged commit 5c9dbe8 into main Sep 17, 2026
8 of 9 checks passed
@yanivil
yanivil deleted the feat/activate-double-bottom branch September 17, 2026 19:45
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Double bottom: eleven-year replay of a candidate pattern

1 participant