Science in practice · Scoring & metrics

The science behind the score.

A good swing number isn't a verdict handed down from a black box — it's a measurement you can trace back to the movement that produced it. This is how PinPoint Studio reads a swing, the biomechanics it scores against, and the maths that keeps a single number honest.

In development & validation

This page explains the approach. The wrist analyzer and the full-body kinematics pipeline both produce real output today — 52 of the catalogue's 95 metrics — and are being checked against a hand-annotated corpus of real swings before any of it is released. The whole-swing score is the piece still being seated on real data. Nothing here is a finished, dependable product yet — that's deliberate. See how we earn the numbers →

How a shot is read

From frames to a score

To map a swing in milliseconds, each shot runs through a sequence of stages — every one building on the last, from raw pixels up to a coachable number.

  • 1

    Pose & object detection

    The analysis pass runs a whole-body pose model over the frozen swing — 133 landmarks: body, feet, face and 21 per hand — while the club and ball are tracked separately. (The lighter skeleton you see drawn live during capture is a different, faster model, and no measurement is taken from it.)

  • 2

    Fusion & smoothing

    The whole-body track is filtered by an offline smoother and fused with the IMUs' high-rate motion data on one shared clock — resolving the jitter and gaps a single frame can't. A second, down-the-line camera will later add the depth axis.

  • 3

    Skeleton, scores & faults

    Events are anchored on impact, metrics are computed against coaching bands, and faults are ranked by how many points they cost — each with a plain-English drill.

PROCESSING STAGES Skeleton, scores & faults 3D fusion & smoothing Pose & object detection Each shot flows bottom → top in milliseconds

Illustrative — the analysis pipeline, stage by stage.

Works with the kit you have

Full numbers from a single camera

The system adapts to whatever hardware is in the bay — a single calibrated camera with motion sensors is a first-class setup, not a fallback. More kit sharpens the picture.

  • 1

    One face-on camera

    A single face-on view already carries most of the catalogue — posture, the head, stance and balance, rotation and X-factor, club delivery and tempo. Adding wrist sensors is what makes the wrist angles instrument-grade rather than inferred.

  • 2

    A second, down-the-line camera

    Resolves depth the single view can't, unlocking the depth-axis measures — club path, launch direction and a true 3D swing plane.

  • 3

    A sensor on the club

    A light sensor on the shaft gives direct club-head speed and shaft lean without needing the second camera.

Thorax IMU Pelvis IMU Wrist IMU Face-on camera

Illustrative — sensor placement and a face-on camera.

Power generation

The power coil: X-factor

Speed in golf isn't arm strength — it's elastic tension stored in the core, which we measure in degrees of separation between the shoulders and hips.

  • 1

    The backswing coil

    The difference in rotation between shoulders and hips at the top. A strong coil separates the two by roughly 40–50°, stretching the core.

  • 2

    The downswing stretch

    As the downswing starts the hips turn first while the shoulders stay coiled, briefly increasing separation — a stretch that acts like a released sling.

  • 3

    How to use it

    A low coil usually means the hips are drifting or over-turning — work on hip stability. A weak stretch means starting the downswing from the ground up, not the upper body.

Hips 45° Shoulders 90° X-factor: 45° Target line 0°

Illustrative — separation of shoulders and hips, viewed top-down.

Swing timing · still to come

The kinematic sequence

Like cracking a whip, speed travels through the body in a specific order. Each segment peaks, then brakes to pass its energy to the next. This one is not measured yet — the phase ladder and the tracks it needs are both in place, but the angular-velocity series the sequence is read from are still to be built.

  • 1

    The proper order

    Energy peaks bottom to top: hips → chest → lead arm → club. Each segment acts as a pivot, handing speed forward.

  • 2

    Peak, then brake

    To transfer speed a segment has to decelerate rapidly. When the hips brake, that energy surges into the chest; when the chest brakes, it snaps into the arms.

  • 3

    Out of order

    If the shoulders or arms peak before the hips, the swing is "casting" — speed bleeds away, the path changes and slices follow.

Impact 1. Hips 2. Chest 3. Arm 4. Club Angular speed through the downswing

Illustrative — peaks fire proximal to distal, in order.

The first analyzer · in validation

Lead wrist: flexion & extension

The clubface direction is governed by the lead wrist — the one movement read straight from the IMUs and the first analyzer producing real output today.

  • 1

    Flexion (bowed, palm down)

    Bowing the lead wrist rotates the face closed. For a slicer, more flexion helps square or close it.

  • 2

    Extension (cupped, knuckles up)

    Cupping the wrist rotates the face open. Excessive cupping is the most common cause of weak, high slices.

  • 3

    Radial / ulnar hinge

    Cocking the wrist up stores the lag through the downswing; releasing it down snaps the club square at impact.

Forearm Bowed — face closed Cupped — face open Wrist joint

Illustrative — lead-wrist flexion vs extension and the face.

Stability & posture

Sway, thrust & lift

Consistency means turning around a stable spine angle — not swaying side to side or thrusting the hips toward the ball. Two of the three are measured face-on today; the third is the clearest example on this page of a metric a single camera physically cannot deliver.

  • 1

    Sway (side to side)

    A small shift away from the target in the backswing, then a strong drive back toward the target coming down. Too much either way costs strike.

  • 2

    Thrust (early extension)

    Hips moving toward the ball crowd the arms and force you to stand up out of posture — a big source of blocks and shanks. It moves along the axis pointing at a face-on lens, so it takes a down-the-line camera: no amount of cleverness recovers a movement the projection collapsed to nothing.

  • 3

    Lift (up and down)

    Hold your height through the backswing, then push off the ground into impact — controlled vertical lift is a source of power.

Ideal rotation centre Sway target Thrust / early extension Hips, viewed from above

Illustrative — a good drive vs a thrust toward the ball.

The scoring maths · being seated

One number, no hiding

For a full swing we don't average the metrics — a plain average lets a serious fault hide behind everything you do well. The method is a weighted geometric mean, so the weakest link sets the tone. The whole-swing score is the one piece deliberately not shipped: the maths is written and the wrist score already runs on it, but a score is only as honest as the corridors underneath it, and the full-swing bands are still being seated on measured swings.

  • 1

    Why averages mislead

    Five perfect metrics (100) and one critical fault (20) average to a flattering 86 — a score that hides the very thing holding you back.

  • 2

    The geometric-mean penalty

    Multiplying the scores and taking the root lets one low number pull the whole score down. The same swing becomes a realistic 40.

  • 3

    The coaching idea

    You can't paper over a critical mechanical fault by perfecting something else. The score stays honest about what to fix first.

Simple average — misleading 86 Weighted geometric mean — honest 40 score = (m₁^w₁ × … × mₙ^wₙ) ^ (1/Σw) One metric near zero drags the whole score down.

Illustrative — the same swing, two ways of scoring it.

Style diagnostics

There's no one right wrist

Unlike posture, wrist shape has no single correct answer — great players differ. So instead of scoring you against one ideal, we diagnose which style you're closest to.

  • 1

    Three archetypes

    Your wrist angles at the top and at impact are matched to three patterns: bowed (strong/closed), neutral, or cupped (open).

  • 2

    A resemblance for each (0–100)

    You get an independent match score for every archetype. Your highest one names your pattern; a high score means you're consistent within that style.

  • 3

    Blended

    If your top two scores are within 10 points, you're flagged "blended" — your wrist action is hybrid, and steadying onto one style is usually the goal.

BOWED (strong / closed face) 86% NEUTRAL 40% CUPPED (open face) 8% Pattern: Bowed — highly consistent Blended: no — one clear, repeatable style.

Illustrative — resemblance scores name the swing's style.

Honest uncertainty

A score with a margin

Cameras and sensors live in the real world — shadows, noise and timing jitter are facts of it. Rather than pretend to a false precision, the wrist resemblance score carries an honest range around it, propagated from a real error budget: sensor noise, the wrist cross-talk, and a timing term that grows when the phase ladder is poorly seated.

  • 1

    "82 (±9)", not just "82"

    The result is an interval — say 73 to 91, centred on 82 — that reflects the real margin of timing and sensor noise.

  • 2

    Data quality sets the width

    Good light, snug straps and two calibrated cameras give a tight band (±3). Poor light or loose straps widen it (±12).

  • 3

    Noise never costs you

    Lower confidence only widens the interval — it never lowers the score. A bad environment shouldn't dent your history. This replaced an earlier design in which confidence multiplied into the number itself — the surest route to being precise, confident and wrong.

82 (±9) 73 91 0 100 Band width = timing jitter + sensor noise

Illustrative — the score sits inside an honest interval.

Putting it to work

Fix one thing, then re-test

The point of all this measurement is a clear next move — not a wall of numbers. The aim is a simple, prioritised loop you can actually practise.

  • 1

    Know your style first

    Check your wrist resemblance. If you're a bowed-wrist player, don't chase neutral or cupped tips — consistency within your own style is the goal.

  • 2

    Attack the top-ranked fault

    Faults are ranked by points lost, so the top one is whatever is holding the score back most. Work on that single thing.

  • 3

    Drill, then re-capture

    Every ranked fault comes with a plain-English drill. Do it, capture a fresh swing, and watch that one metric drift the right way — and the interval tighten as it gets repeatable.

The takeaway: improvement isn't guesswork about "feel" — it's measured movement. Change one thing at a time, let the numbers confirm it, and trust the physics over the urge to overhaul everything at once.

4. Re-test 3. Drill 2. Top fault 1. Your style

Illustrative — the practice loop, one fault at a time.

Reference · The full catalogue

Every metric, in plain terms

The walk-through above covered the ideas; this is the reference. The app's catalogue holds 95 metrics, and these are the ones worth knowing by name, grouped by what they tell you. Each is read at the swing's key positions (the P-position ladder, address to finish) and, where it is scored, held against a reference band at the phase that matters most.

A note on honesty: 52 of the 95 produce real output today — the wrist set, rotation, posture, the head, stance and balance, club delivery and tempo among them. Another 26 are readings a connected launch monitor supplies, and 17 are still to come: the depth-axis measures a face-on camera physically cannot see, and ball flight. The typical ranges are reference targets drawn from the published biomechanics literature and tour datasets, and stay provisional until our own validation completes. Each card flags what it needs:

IMU — a worn wrist sensor delivers it 1 camera — a single face-on view, today not yet — needs a second camera, a club sensor or a launch monitor

Rotation & power — the engine

Pelvis rotation

pelvisRotation
1 camera

How far the hips turn away from the ball and back through. The base of the coil and the first link in the sequence.

Typical ~45° at the top · Route face-on camera today (an estimate); a second camera or torso IMUs sharpen it

Thorax rotation

thoraxRotation
1 camera

Shoulder turn — the larger of the two rotations, and the one that stretches the core against the hips.

Typical ~90° at the top · Route as pelvis rotation

X-factor

xFactor
1 camera

Shoulders minus hips — the separation that stores elastic energy in the trunk. The single most-cited power measure in the literature.

Typical ~40–45° at the top

X-factor stretch

xFactorStretch
1 camera

The extra separation gained just after the downswing starts, as the hips fire while the shoulders stay coiled — the stretch-shortening that releases like a sling.

Typical a few degrees beyond the top value

Shoulder plane angle

shoulderPlaneAngle
1 camera

How steep or flat the shoulders sit at the top. Positive means the trail shoulder is the higher one. As much a consequence of build and setup as of what the golfer did, so it's read beside address posture.

Reads per-frame curve · the value at the top is the one that matters

Kinematic sequence

kinematicSequence
not yet

The order and timing of peak rotational speed up the chain — hips, chest, lead arm, club. Out of order is the classic "casting" pattern.

Typical peaks rise ~480 → 605 → 1310 deg/s · ~50 ms apart · Needs angular-velocity series and pelvis + thorax sensors

Posture, tilt & the head

Secondary axis tilt

secondaryAxisTilt
1 camera

Lean of the spine away from the target at impact — the hip-to-shoulder line measured from vertical in the face-on plane.

Typical ~20–25° at impact (from ~6–8° at address)

Spine side bend

spineSideBend
1 camera

Shoulder line against hip line. Too little goes with a steep, over-the-top delivery; too much throws the low point behind the ball.

Typical thorax ~32° at driver impact, against ~10° at the pelvis

Head sway

headSway
1 camera

Lateral travel of the head through the swing. A little back-and-through is normal; only excess is a fault. Read beside pelvis sway to tell a head move from a whole-body slide.

Reads per-frame curve in cm · scaled by the inter-ear ruler

Head lift

headLift
1 camera

Rise or dip against address height. An early rise through the downswing is the visible half of standing up out of posture; a dip is a drop into the shot.

Reads per-frame curve in cm · read with pelvis lift

Head tilt

headTilt
1 camera

Change in eye-line angle from address. Small and stable is normal; a large or abrupt change tends to accompany a loss of posture.

Reads per-frame curve · completes the head picture with sway and lift

Spine forward bend

spineForwardBend
not yet

How much the golfer is bent over from the hips, and whether that angle is held. It lives in the sagittal plane — present in a face-on image but foreshortened to noise by the projection.

Typical ~30–40° at address · Needs a down-the-line view

Setup, stance & balance

These arrived with whole-body pose. The feet, heels and toes are landmarks in their own right now, which gives a ground line at shoe level rather than at the ankle — and with the ball in frame as a 42.67 mm ruler, several of them convert to real units.

Stance width

stanceWidth
1 camera

Heel-to-heel span, expressed against the golfer's own shoulder width rather than in millimetres — so a tall player and a short one are judged on the same scale.

Typical 100 % is exactly shoulder-width · wider for long clubs, narrower for wedges

Ball position

ballPosition
1 camera

Where the ball sits in the stance, as a fraction of stance width from the lead heel. Read against the club in hand, never against one ideal.

Typical driver near 0 % (off the lead heel) · wedge ~50 % (mid-stance)

Foot flare

leadFootFlare / trailFootFlare
1 camera

How far each toe points out at address. More lead-foot flare makes it easier for the lead hip to clear through the strike — a setup lever for a golfer who struggles to finish the turn.

Reads heel→big-toe direction, per foot, at address

Toe line

toeLineAngle
1 camera

Stance alignment — open, square or closed. Measured face-on, so it reads the apparent line; a down-the-line view would resolve true target-line alignment.

Reads a single address measurement

Lead heel lift

leadHeelLift
1 camera

How far the lead heel comes off the ground around the top. A little is common and can free the turn; keeping it down is a legitimate stylistic choice.

Reads per-frame curve in cm · scaled by the ball-diameter ruler

Balance at the finish

comOverLeadFoot
1 camera

How far the body's centre sits from the lead ankle at the finish — still hanging back, or fallen through it. A balanced finish is the low end.

Honest limit a geometric proxy for balance, not a measurement of it — that needs pressure data

Pelvis & lateral movement

Pelvis sway

pelvisSway
1 camera

Side-to-side travel of the hips along the target line: slightly away from the lead side going back, then a drive toward it coming down, returning near zero by impact.

Typical a large negative peak going back is sway — it costs turn and centredness · measured against stance width

Pelvis lift

pelvisLift
1 camera

The pelvis centre rising or sinking as a whole. A small controlled push up into impact adds speed; an early or excessive lift pulls the club off its path.

Read with pelvis thrust and spine forward bend · measured against stance width

Hip line tilt

hipLineTilt
1 camera

How far the trail hip sits above the lead hip at the top. Deliberately an absolute angle rather than a change from address, because the figure it's read against is absolute — everyone has some.

Reads at the top, in the image plane

Plumb bob

plumbBobDistance
1 camera

Where the hips sit over the stance, dropped as a vertical line and measured in inches against the ball. Read first at address against the club's own corridor, then read the travel: the hips move lead-side through every good downswing.

Reads P1 through P7 · needs the ball for its ruler

Thorax lateral drift

thoraxLateralDrift
1 camera

The chest's counterpart to pelvis sway, same sign convention — the two are read together, because the same absolute movement means something different at each end.

Reads displacement from address · the early downswing is the telling part

Lead knee drift

leadKneeDrift
1 camera

The lead knee's sideways travel minus its own hip's — a difference, deliberately, since pelvic rotation carries the hip across anyway. A large value at the top is the observation.

Reads per-frame curve · measured against stance width

Pelvis thrust

pelvisThrust
not yet

Movement of the hips toward the ball — the classic "early extension" that crowds the arms and forces you upright. It lives on the depth axis a face-on camera is blind to.

Needs a down-the-line camera

Lead wrist & forearm

Bow / cup

leadWristFlexExt
IMU

Flexion (bowed, palm down) versus extension (cupped) of the lead wrist — the main control of clubface angle, since the back of the lead hand mirrors the face.

Typical 15–30° more flexed at impact than address · cupping at the top is a fault

Hinge (set / lag)

leadWristRadUln
IMU

Radial (cocking up) versus ulnar deviation — how the wrist sets the hinge, holds the lag, then releases it. The lateness of the release into impact is the marker of good timing.

Typical large set held to ~P6, releasing late · early drop = casting

Forearm roll

forearmPronation
IMU

Axial rotation of the lead forearm — the "roll" that squares the face through impact.

Typical rotates toward square through impact · holding off = block / weak fade

Forearm rotation from address

forearmRotation
IMU

The same axis stated as a change from address, which is the honest way to read it: the absolute value is meaningless across sessions, because it depends on whatever posture address happened to be. Large, late pronation into impact is flipping.

Read the P1→P7 change, never a single value · a HackMotion supplies a measured twin

Lead elbow

leadArmFlexion
IMU

How straight the lead arm stays, from the shoulder, elbow and wrist geometry. A bent, lifting lead elbow late is the "chicken wing".

Typical near-straight (~165–180°) through impact

Measured twins

hm.leadWristFlexExt · hm.leadWristRadUln · hm.forearmRotation
HackMotion

When a HackMotion wG3 is worn, its three readings are recorded under their own keys beside ours rather than in place of them — so which instrument graded a swing stays recoverable after the fact.

Why separate keys the comparison is the point; a single blended number would destroy it

Arms, connection & the trail side

Lead arm connection

leadUpperArmToChest
1 camera

The gap between the lead upper arm and the chest, widest somewhere between address and the top. Some separation is normal — zero would mean the arm is pinned, which is its own fault.

Reads the widest gap · measured against shoulder width

Swing width at the top

leadHandWidth
1 camera

How far the hands stay from the body at the top, expressed against the golfer's own arm length — a fraction of the width actually available to them rather than an absolute distance.

Reads lower is narrower · at the top

Trail elbow height

trailElbowHeight
1 camera

How far the trail elbow has risen above the shoulder line at the top — the "flying elbow" made into a number, normalised by shoulder width so it means the same for any golfer.

Reads at the top · higher is further above the line

Lead arm to torso

leadArmToTorso
1 camera

The arm-to-body angle through the strike. A rising angle after impact means the arm is separating from the body rather than extending down the line.

Reads the change from impact into the follow-through

Trail wrist (the "tray")

trailWristFlexExt
1 camera

The trail wrist's extension that "holds the tray" supporting the club. Flattening it too early is the same throw/flip event the lead wrist shows — used to corroborate the face reading and raise confidence.

Typical extension held to ~P6, releasing late · read from the whole-body hand landmarks

Elbow alignment

elbowAlignment
1 camera

Trail elbow against lead elbow, at address and at impact. The change between the two shows how the arms fold, rotate and re-deliver — the trail elbow tucking on the way down, for instance.

Positive the trail elbow sits above the lead one

Club delivery & speed

You might wonder why these sit here at all when a launch monitor already reads the ball and club. The overlap is deliberate. A launch monitor tells you the outcome; PinPoint tells you what the body did to cause it — and club delivery is where those two worlds meet. That's not hypothetical: a connected monitor's readings — over Open Connect or from a GCQuad — are recorded beside our own estimates on every swing, never replacing them. Where the shared numbers agree, each set vouches for the other: you get provenance for every reading, and a clean line from a fault in the ball flight back to the move that produced it.

Clubhead speed

clubheadSpeed
1 camera

Speed of the clubhead at impact, differentiated from its path in the face-on image and scaled by the ball-diameter ruler. A second camera, or a sensor on the shaft, would restore the depth component of its velocity.

Typical driver ~113 mph · 7-iron ~89 mph · drawn as a review curve, not scored

Hand speed

handSpeed
1 camera

Speed of the grip. In a good release its peak comes slightly before impact — the hands slowing is what lets the clubhead sling past. Hands still accelerating at impact means a late release or a stalled body.

Read the peak and where it falls relative to impact

Lag angle

lagAngle
1 camera

The angle between the lead arm and the shaft through the downswing. Casting widens it early and throws away speed; holding it too long leaves the face open.

Read how deep into the downswing the angle is retained

Shaft lean

impactShaftLean
1 camera

How far the shaft leans forward at impact. Hands ahead of the ball is typical and desirable for irons; the driver is played with the shaft near vertical or leaning back. Too little on an iron is an early release, with thin and fat tendencies.

Reads at impact, from the face-on club track

Shaft angle at the top

shaftAngleVsHorizontal
1 camera

Swing length, as an angle. Zero is parallel to the ground, positive is past parallel. Strongly club-dependent and partly a matter of flexibility and style — so the corridor is wide and a reading outside it is a conversation, not a verdict.

Reads the grip-to-head vector at the top

Attack angle

attackAngle
1 camera

Whether the clubhead is moving down or up at impact — the vertical angle of its travel, read from the tracked club in the face-on plane.

Typical driver ~−1° (many good drives are positive) · 7-iron ~−4.5°

Low point

lowPointAhead
1 camera

How far target-side of the ball the clubhead bottoms out its arc, signed in inches. Positive is a low point ahead of the ball — the descending, ball-then-turf strike an iron wants; negative sits behind it, wrong for an iron but exactly right hitting up on a driver. A delivery number a launch monitor doesn't give you.

Honest limit read off the synthesized club arc, since impact is where the head is fastest and most blurred — about ±2 in of spread on one swing, but unbiased, so several swings average to something trustworthy. A projected clubhead is refused outright: its arc vertex would be the grip's, not the club's.

Transition plane delta

transitionPlaneDelta
experimental

Whether the club steepened or shallowed between backswing and downswing — over-the-top, quantified. Positive is steeper.

Status deliberately normless: its corridor is a placeholder set wider than anything yet observed, so it records and accumulates without grading anyone until it has more than one golfer behind it

Club path

clubPath
not yet

The horizontal direction the clubhead travels through impact relative to the target line — in-to-out (+) or out-to-in (−). The discriminating axis is the one pointing at the camera, which is exactly what a face-on view cannot see.

Typical near 0 ± a few degrees · Needs a down-the-line camera

Swing plane

swingPlane
not yet

The tilt and direction of the best-fit plane the clubhead travels on through the downswing — a genuinely three-dimensional object.

Needs a down-the-line camera and the club track

Ball flight & the launch monitor

The measured set

lm.… (26 readings)
launch monitor

Ball speed, spin, launch, face angle, face-to-path, strike location and the rest, read out of a connected monitor — a Foresight GCQuad through FSX2020, or any device speaking GSPro Open Connect (Garmin R10, Rapsodo MLM2PRO, SkyTrak+, PiTrac). They are live readings, not promises — and they carry an lm. prefix precisely so that a measured number can never be mistaken for one of our estimates.

Six of them have a twin we estimate ourselves — that pair is the validation surface

Compound miss

compoundMiss
launch monitor

Start line and curvature combined into one signed reading: positive started and curved right, negative both left. A pull-fade reads exactly zero, not a small number — the halves cancelling is what makes it a different shot, and saying so is the point.

Reads each half against the point its own word starts to apply

Ball speed & launch

ballSpeed · launchAngle · launchDirection
not yet

Our own optical estimates of what the monitor measures. They need a ball-flight tracker; what we have today is an at-spot presence detector, which is a different instrument.

Why keep them owning the device is only useful if there is an estimate to check against it — so both have to exist

Tempo & the scores

Backswing time

tempoBackswing
any source

How long the backswing takes, from address to the top — read straight from the swing's phase events, so it needs no particular device.

Typical ~0.75–0.85 s

Tempo ratio

tempoRatio
any source

Backswing time divided by downswing time — the rhythm of the swing, and a number that's remarkably consistent among good players.

Typical ~3:1

Wrist score

wristScore
IMU

The headline for the wrist analyzer — higher is closer to the reference. Read it, then drill into the individual wrist metrics to see what moved it; most often it is a cupped lead wrist at the top.

Needs the lead-forearm and lead-hand sensors

Wrist resemblance

wristResemblance
IMU

How closely the wrist action matches each of three archetypes, scored independently. The headline is the best match; the trio tells you how decisively. Within a few points and it is flagged "blended".

Use it to read a player's natural pattern before coaching toward or away from it

Swing score

swingScore
not yet

The whole-swing counterpart to the wrist score: one 0–100 rating of how the swing adheres to the reference. The scorer and its full-swing bands are the piece still being seated on real data.

Status the last big gap, and deliberately not shipped until the corridors behind it are earned
Science-led, kept open

Measured, not guessed.

Every method here is drawn from the published literature and held against real, hand-annotated swings before it ships. It's an open, inspectable platform — and still a prototype under validation: the first beta shipped in September 2026, with 1.0 targeted for late in the year.