How ShotIQ Works: The Physics & Math Behind Your Numbers
Golf apps love to hand you conclusions. ShotIQ's design principle is the opposite: every conclusion should be traceable to arithmetic on your actual shots — and where the data can't support a conclusion, the app says so instead of guessing. This page explains the machinery: what happens to a session between upload and advice.
On this page
First, the data has to be trusted: the import gateway
Every upload — CSV, screenshot, PDF, any device — passes through one intelligence gateway (ShotIQ's Universal Session Intelligence Engine) before a single number reaches your bag. It works in four honest stages:
- Understand. Detect the source (TrackMan, Rapsodo, GSPro, Garmin and a dozen more each have recognizable fingerprints — column dialects like GSPro's "Efficiency" for smash factor are translated automatically) and detect the intent: a gapping session reads differently from a skills challenge, which reads differently from a simulated round. Club-switch patterns are one tell: switching clubs on more than a third of consecutive shots looks like a combine, not block practice.
- Normalize. Units are converted (meters to yards, km/h and m/s to mph), original values preserved, and unknown columns kept rather than discarded.
- Validate. Physics gates catch what shouldn't pass: smash factors beyond physical limits (above ~1.62 is flagged — real drivers cap near 1.50), ball speed below what the club speed could produce, carry exceeding total, duplicate rows from doubled exports. Suspect shots are flagged, never silently deleted.
- Clarify. The one unforgivable import error is merging different clubs into one fake "dispersion." A file with no club labels, many shots, and a huge carry spread blocks the import and asks you — one tap to resolve — rather than proceeding confidently on garbage.
Each import carries an honest confidence score. Low confidence doesn't hide; it shows.
Your bag: pooled, not cherry-picked
Once trusted, shots pool into per-club statistics — the foundation nearly everything else reads:
- Carry is the mean of all real recorded carries for that club, and carry consistency is their standard deviation (the ± number you see). Not your best day. All of it.
- Side spread — the width of your pattern — is computed as the root-mean-square of side-carry values, which naturally weights bigger misses more (a 20-yard miss hurts more than two 10-yard misses, and the math agrees). The mean side captures your directional bias: the pattern's center, not just its width.
- Dispersion ellipses you see on shot maps are drawn from exactly these numbers — depth from carry deviation, width from side spread, centered on your bias. The ellipse isn't decoration; it is your data.
Recency weighting. When one carry number must represent a club (the caddie, gapping), sessions are weighted by age: full weight inside two weeks, then 0.6× to a month, 0.3× to two months, 0.15× beyond. Your swing changes; your numbers follow at a sensible pace, without one great session three months ago haunting your club selection.
Honesty gates everywhere. Sensor fields that weren't recorded are never invented — a device without face/path data simply doesn't produce face/path analysis. Zeroes on unsigned metrics are treated as missing, while a genuine 0.0° face angle (square!) is kept — sign conventions matter, and ShotIQ's is consistent app-wide: positive = right, positive attack = up.
Gapping: geometry with a quality bar
Adjacent clubs are compared by carry: gaps of 8–20 yards are healthy, under 8 is an overlap (two clubs doing one job), over 20 is a hole in your coverage. Crucially, the analysis only compares truly adjacent clubs you've actually logged — a 60-yard span between your driver and 7-iron is unlogged clubs, not a "fitting gap," and the app is explicitly built not to make that rookie-analyst error.
From numbers to advice: one source of truth
Every "what should I work on?" surface — the home coaching headline, the Game Plan, practice prescriptions, the Learn curriculum — reads a single computed game state, so the app never tells you two different stories. That state ranks improvement paths by expected strokes, weighing pattern width (dispersion), strike quality (smash versus your club's ceiling), delivery faults (like a negative driver attack angle), and short-game signals — each traceable to the statistics above.
Layered on top:
- Golf DNA names your patterns from the data: your shape (from mean side and spin axis), your typical miss, your most reliable club (tightest relative pattern with enough shots), and a course-readiness score.
- Transfer Score compares GPS-verified on-course carries against your range carries per club — the fraction of practice distance that actually survives on grass, which is the honest referee of both your practice quality and your planning numbers.
- The caddie and Quick Caddie pick clubs from your recency-weighted carries, then show your real dispersion ellipse over the target — with adjustments for lie, wind and elevation applied to the physics, not vibes. Plans are built on reliable carry (what you usually do), never best-case.
- Setup Lab scans for address-rooted signatures — start-line bias with little curve (aim), low iron launch (ball position), negative driver attack (tee height), erratic smash (routine) — and ranks fixes by strokes. Every rule is gated on sufficient real data; nothing fires on a hunch.
- Club Potential estimates the yards and strokes on the table per club by comparing your deliveries against your own demonstrated bests and club-appropriate objectives — never turf advice for a driver, never distance-chasing for a wedge.
Rounds, targets and everything else
Simulated rounds (from any launch monitor's course mode) flow through a universal Round → Hole → Shot model, so scoring analysis works identically whatever software produced it. Target-practice sessions score proximity separately in Target Lab — target games never pollute your stock-shot bag averages, because a three-quarter wedge at a 60-yard flag is not data about your full wedge. Manual shots logged on-course become first-class sessions too, feeding the same statistics with their sensor fields honestly zeroed rather than faked.
The AI layer: grounded, not free-styling
ShotIQ's AI coach doesn't improvise golf theory. It's fed your computed statistics plus the same knowledge engine that powers the concepts library — so its explanations cite the numbers the app calculated and the concepts this library documents. When the AI analyzes a session, it compares against your previous same-club sessions and your history; when it recommends, it recommends from the same single game state as every other surface. And club-aware rules keep its advice physically sensible — you will never be told to "brush the turf" with a driver.
Frequently asked questions
Does ShotIQ ever estimate numbers my launch monitor didn't measure?
No. Metrics your device didn't record are omitted and listed as not recorded — never guessed. The one transparent exception: when a device reports curve as side-spin instead of spin-axis tilt, ShotIQ converts between those two representations of the same physics, and labels estimates as estimates.
Why does my ShotIQ carry differ from my launch monitor app's number?
Your monitor's app typically shows single-session averages; ShotIQ shows a pooled, recency-weighted picture across sessions — deliberately. One session is an audition; the pooled number is your game. Elevation and device calibration offsets, if you've set them, also apply.
How many shots before the numbers mean something?
Statistical honesty scales with sample size: minimums are enforced per analysis (Setup Lab rules need several shots minimum; consistency numbers need more than one), and confidence grows visibly as clubs accumulate shots. Ten shots per club is a fair start; the app tells you where data is thin rather than bluffing through it.
Can I trust analysis from a budget launch monitor?
For the metrics it genuinely measures, yes — and ShotIQ's validation catches the rows that defy physics regardless of price tag. The capability gating means a budget unit gets a smaller, honest analysis rather than a padded, invented one.