Skip to content
← Build journal
Research

Aim as a tolerance: the 001 alignment study

Alignment systems get marketed with a lot of confidence and not much method. We want to be specific, in public, and to let you check our claims against our method — including the numbers we'd rather not have found.

The number that matters

To hole a twelve-foot putt on its intended line, the face needs to be within roughly half a degree of square at impact. That's the tolerance the whole chain has to hit — read, aim, setup, stroke. Published aim studies suggest many golfers use up a large share of that budget before the stroke even begins: at address, misaim of a degree or more is common.

That's the opportunity 001 is built around, and it splits into two separate questions that are easy to tangle together: what does the shape of the head do to a golfer's aim, with no alignment aid present at all — and what does a printed sightline add on top of that shape? We run those as two tracks, in that order, so one effect can't be mistaken for the other.

Track one: geometry, tested bare

Ten mallet heads, 3D-printed from a single parametric family, varying only in four dimensions: heel-toe face width, flange (rear) depth, wing forward-sweep angle, and topline thickness. No sightline, no alignment mark, no dot on any of them — the heads were geometrically bare, so any effect on aim could only come from shape. Seven golfers of mixed handicap aimed every head, repeatedly, in randomized order, while an overhead machine-vision gauge measured face angle at address to roughly 0.02° resolution. 389 aims recorded in total, analyzed with a mixed-effects model that separates each golfer's personal aiming tendency from the effect of the head itself.

Figure 1 — static aim by test head Mean aim angle by test head, geometry screening study Ten mallet heads, mean aim angle at address in degrees open, pooled across golfers, sorted from squarest to most open: R10 1.44 degrees, R06 1.52, R08 1.54, R09 1.55, R04 1.58, R05 1.60, R07 1.91, R03 2.11, R01 2.13, R02 2.16. R09 and R10 are a repeated center-point build; the rest are corners of a two-to-the-four-minus-one fractional-factorial design. R10 (CP) 1.44° R06 1.52° R08 1.54° R09 (CP) 1.55° R04 1.58° R05 1.60° R07 1.91° R03 2.11° R01 2.13° R02 2.16° 0° 0.5° 1.0° 1.5° 2.0° 2.5° aim angle at address, open (right of target)
Mean aim angle per test head, pooled across golfers, geometry screening study (AS-GEO). Every head aimed open on average — consistent with the right-of-target bias documented for right-handed golfers — but roughly 0.7° separates the best-aiming head (R10) from the worst (R02). Dashed outlines mark the two repeated center-point builds (R09, R10); the rest are corners of the design.

What moved the needle

One factor dominated. Increasing wing forward-sweep angle from 20° to 30° squared the average golfer's aim by 0.38° (p = 5.3×10⁻⁵) — a result that held up under every check we threw at it, including refitting the model with individual golfer sessions removed. Face width showed a smaller, less certain effect. Flange depth and topline thickness moved aim by less than a tenth of a degree — statistically indistinguishable from nothing.

Figure 2 — effect of each geometric factor on aim bias Fitted effect of each geometric factor on aim bias, across its tested range Wing forward-sweep angle: 0.38 degrees, p equals 5.3 times ten to the negative fifth, significant and robust. Face width: 0.23 degrees, p equals 0.014, tentative — not robust to excluding a single golfer. Flange length: under a tenth of a degree, not significant. Topline thickness: under a tenth of a degree, not significant. Wing forward-sweep 0.38° · p = 5.3×10⁻⁵ Face width 0.23° · p = 0.014 (tentative) Flange length < 0.1° · not significant Topline thickness < 0.1° · not significant 0° 0.1° 0.2° 0.3° 0.4°
Fitted effect of each geometric factor on aim bias, across its full tested range, from the mixed-effects model on 389 aims. Wing sweep is the only effect that stayed significant, same-signed, and top-ranked no matter which golfer sessions we excluded to stress-test it. Face width shrank toward zero once we dropped a single session, so we're treating it as tentative rather than settled. Flange length and topline thickness never separated from noise — a genuine negative result, not an absence of looking.

That last point is worth sitting with: flange depth and topline thickness are both dimensions putter marketing leans on, and in this sample neither one moved where a golfer aimed. Geometry also had no measurable effect on aim consistency — how tightly a golfer's aims clustered around their own average looked like a property of the golfer, not the head. Bias and dispersion are different problems, and they answered to different levers.

Why does sweep dominate? The interpretation most consistent with the perception research this study was designed around: forward-swept wings place two long, converging edges in the golfer's peripheral view, pointing down the target line — a frame the visual system's alignment acuity can lock onto before any sightline is added. But a screening study measures the effect, not the mechanism. Other accounts — familiarity with a silhouette closer to the putter a player already games, for instance — weren't separated here, and probing the mechanism belongs to the sightline track. The 0.38° is a finding; the why is still an interpretation.

What this informs

These results now inform 001's current CAD direction and starting geometry — a forward-swept wing profile in the range the data supports, with the center-point build (25° sweep) aiming as well as the best corner we tested, so the guidance is a band, not a single extreme. Flange depth and topline thickness are, for now, set by mass, inertia, and acoustic goals rather than alignment. None of this is a finalized production body: it remains subject to prototype evaluation, design-for-manufacture, sightline testing, and further validation before anything is locked in. It's also why the sightline study below hasn't started yet: we wanted the starting geometry it runs on to already reflect this data, not a placeholder we'd have to redo the study on later.

Seven golfers is a screening sample, not a population estimate, and two of those sessions carried disclosed measurement caveats we controlled for by refitting the model without them — the headline ranking held. We're confident in which dimension matters and roughly by how much; we're not yet publishing population-level miss probabilities from a sample this size.

Track two: alignment, ahead

Additional alignment development remains underway, building on the geometry-track results above. We're not publishing a study design or promising an outcome ahead of the work — OURO uses an iterative research, design, prototype, and measurement process to guide development, and alignment follows the same rule as everything else on this page: no number until it's measured. Results will land here and on the 001 page — means, spreads, and method, not adjectives — exactly as they did above.