[cinar]

Summary

Adds internal/talibcompat, a test package that checks each indicator with a TA-Lib counterpart against TA-Lib output computed with the same parameters. TA-Lib serves as the reference implementation for indicator math.

• Generator. testdata/generate.py builds two seeded 300-bar OHLCV datasets and writes the inputs plus the TA-Lib reference outputs to CSV:
  • walk: a random walk whose prices practically never repeat;
  • ticks: prices rounded to a 0.5 tick, so equal highs, lows and closes occur and tie handling is exercised.
• Test. talibcompat_test.go reads the CSVs with the standard library and compares 72 cases (about 100 outputs) per dataset, with a relative tolerance of 1e-8.
• No new dependencies. Only the generator needs TA-Lib (pip install ta-lib==0.8.3 numpy). The Go module and CI stay dependency-free, and the CSVs are committed (about 630 KB).

How known differences are recorded

Each case states how it compares, so every intentional difference is explicit in one place:

Field	Meaning	Cases
scale	The library returns a ratio, TA-Lib a percentage (#533)	Roc, Trix, Kst, BollingerBandWidth, StochasticRsi
fromBar	EMA seeding differs, values converge after a warm-up (#535)	Macd (170), ChaikinOscillator (150)
skip	Known difference, linked to its issue	Aroon #520, Cci #521, Dpo #522, EhlersFisher #523, ConnorsRsi #524, Stc #525, Atr / ChandelierExit #526, SuperTrend #527, KeltnerChannel #528, Kvo #529, Cfo #530, Slope #531, InternalBarStrength #532, IchimokuCloud #534

Resolving one of those issues means removing its skip line. I checked that every skipped case currently fails when unskipped. The one exception is IBS on walk, which has no bar where high equals low.

Two passing cases, AtrWithRma and SuperTrendWithRma, show that the existing NewAtrWithMa / NewSuperTrendWithMa extension points already match TA-Lib exactly with Wilder's smoothing (relevant to #526 and #527).

Other changes

• taskfile.yml: adds ./internal/... to INDICATOR_BASE. Without it, task (and so CI) ran neither this package nor the existing internal/apiconsistency tests. Both pass.
• trend/README.md, momentum/README.md: line-number updates that task docs regenerates from earlier commits (Keep a 0 in the first MovingMin and MovingMax window #516, Report the completed TD Sequential countdown #517).

Test plan

• task (fmt, lint, test with -race, docs) passes locally
• 123 subtests pass, 30 skip (15 known differences × 2 datasets)
• Temporarily removing the skips makes every skipped case fail, except IBS on walk

🤖 Generated with Claude Code

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.75%. Comparing base (8e0d794) to head (35f6990).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #536      +/-   ##
==========================================
- Coverage   91.78%   91.75%   -0.03%     
==========================================
  Files         240      240              
  Lines        7835     7835              
==========================================
- Hits         7191     7189       -2     
- Misses        551      553       +2     
  Partials       93       93              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.