Three Drives
A symmetric harmonic of three drives separated by two retracements, read from the last seven pivots. Three rising drives → bearish
-1; three falling drives → bullish+1.
Quick reference
| Item | Value |
|---|---|
| Family | Harmonic Patterns |
| Input type | Candle (uses high, low) |
| Output type | f64 (+1 / -1 / 0) |
| Output range | {-1.0, 0.0, +1.0} |
| Default parameters | none (swing threshold 5%, baked) |
| Warmup period | 8 |
| Interpretation | Exhaustion after three symmetric drives |
Formula
last seven pivots, six alternating legs R1 D1 R2 D2 R3 D3:
D1 / R1 ∈ [1.13, 1.75] (each drive extends the leg before it)
D2 / R2 ∈ [1.13, 1.75]
D3 / R3 ∈ [1.13, 1.75]
D1 ≈ D2 ≈ D3 (within 20%) (the three drives are similar in size)
R1 ≈ R2 ≈ R3 (within 30%) (the retracements between them are similar)
direction: terminal pivot a swing high → -1 (drives up), a swing low → +1See crates/wickra-core/src/indicators/three_drives.rs.
Parameters
None. Swing threshold 0.05 is a baked-in family constant (pattern_swing.rs); the extension windows and symmetry tolerances are documented detector constants. ThreeDrives::new is infallible.
Inputs / Outputs
rust
const _: fn(&mut wickra::ThreeDrives, wickra::Candle) -> Option<f64> =
<wickra::ThreeDrives as wickra::Indicator>::update;- Python.
update((o,h,l,c,v,ts))→float(neverNone);batch(open, high, low, close)→array.array('d'). - Node.
update(open, high, low, close)→number;batch(open, high, low, close)→number[]. - WASM.
update(open, high, low, close)→number.
Warmup
warmup_period() == 8. Seven confirmed pivots are required — a structure that stops after two drives is incomplete, not a match, and the detector keeps returning None. Pinned by tests accessors_and_metadata and two_drives_alone_do_not_complete_the_pattern.
Edge cases
- Three rising drives report
-1(testbearish_three_drives_is_minus_one). - Three falling drives report
+1(testbullish_three_drives_is_plus_one). - Asymmetric / non-extending drives report
0.0(testasymmetric_drives_do_not_trigger). resetclears state (testreset_clears_state).- Streaming equals batch (test
batch_equals_streaming).
Examples
Rust
rust
use wickra::{Candle, Indicator, ThreeDrives};
fn main() -> Result<(), Box<dyn std::error::Error>> {
// Three rising drives (124, 128, 132), each extending a 10-point
// retracement by 14 → bearish exhaustion.
let bars = [
(119.88, 120.0, 119.88, 119.88),
(110.0, 118.8, 110.0, 110.0),
(111.1, 124.0, 111.1, 111.1),
(114.0, 122.76, 114.0, 114.0),
(115.14, 128.0, 115.14, 115.14),
(118.0, 126.72, 118.0, 118.0),
(119.18, 132.0, 119.18, 119.18),
(118.8, 130.68, 118.8, 118.8), // third drive confirms → bearish
];
let mut pat = ThreeDrives::new();
let mut last = 0.0;
for (ts, (o, h, l, c)) in bars.iter().enumerate() {
last = pat
.update(Candle::new(*o, *h, *l, *c, 1.0, ts as i64)?)
.unwrap_or(last);
}
println!("{last}"); // -1
Ok(())
}Python
python
import wickra as ta
bars = [
(119.88, 120.0, 119.88, 119.88, 1.0, 0),
(110.0, 118.8, 110.0, 110.0, 1.0, 1),
(111.1, 124.0, 111.1, 111.1, 1.0, 2),
(114.0, 122.76, 114.0, 114.0, 1.0, 3),
(115.14, 128.0, 115.14, 115.14, 1.0, 4),
(118.0, 126.72, 118.0, 118.0, 1.0, 5),
(119.18, 132.0, 119.18, 119.18, 1.0, 6),
(118.8, 130.68, 118.8, 118.8, 1.0, 7),
]
pat = ta.ThreeDrives()
print([pat.update(b) for b in bars][-1]) # -1.0Node
javascript
const wickra = require('wickra');
const bars = [
[119.88, 120.0, 119.88, 119.88], [110.0, 118.8, 110.0, 110.0],
[111.1, 124.0, 111.1, 111.1], [114.0, 122.76, 114.0, 114.0],
[115.14, 128.0, 115.14, 115.14], [118.0, 126.72, 118.0, 118.0],
[119.18, 132.0, 119.18, 119.18], [118.8, 130.68, 118.8, 118.8],
];
const pat = new wickra.ThreeDrives();
let last = 0;
for (const [o, h, l, c] of bars) last = pat.update(o, h, l, c);
console.log(last); // -1Streaming
python
pat = ta.ThreeDrives()
for o, h, l, c, v, ts in candle_feed:
signal = pat.update((o, h, l, c, v, ts))
if signal < 0:
pass # three drives up — exhaustion, reversal lower
elif signal > 0:
pass # three drives down — exhaustion, reversal higherInterpretation
- Symmetric exhaustion. Three measured, equal drives with equal pullbacks signal that the move is overextended; the reversal target is back toward the pattern's origin.
- Symmetry is the filter. The two
≈checks (equal drives, equal retracements) keep ragged, accidental three-leg moves out.
Common pitfalls
- Two-drive reading. This detector keys on the two visible drive legs of the five-pivot window; ensure the data actually shows the classic three-push structure before trusting the signal.
- Confirmation lag. Non-repainting but lags the third drive's extreme.
References
- Pesavento, L. Fibonacci Ratios with Pattern Recognition (1997).
See also
- Gartley, Crab.
- Indicators-Overview — full taxonomy.