Every video-checkable cue in Felton’s corpus, organised by where you put the camera and which frame you scrub to. Each cue names the paper behind it so you can check the evidence yourself.

Read the two warnings at the bottom before you coach any of this.

Filming setup

You need three angles. Most cues live at one specific angle and are invisible from the others.

Slow motion is essential. The cues below are separated by 5–10° and by single frames.

The four frames that matter in fast bowling. Everything is timed to these:

1. Bound take-off — last contact before the delivery stride
2. BFC — back foot contact
3. FFC — front foot contact
4. BR — ball release

A useful frame-count baseline: elite men spend ≈192 ms from BFC to FFC and ≈103 ms from FFC to BR. Elite women ≈172 / 128 ms — same total, distributed differently.

FAST BOWLING

Frame 1 — Bound take-off (side-on)

Anterior pelvic tilt at take-off — the best upstream cue in the corpus

• Look for: the tilt of the belt line as the bowler leaves the ground into the bound.
• Fault: pelvis tipped forward, backside out.
• Why it matters: anterior pelvic tilt at bound take-off predicts a flexed rear hip at back foot contact (r = 0.716, p < 0.001) — and a flexed rear hip at BFC is the single strongest predictor of lumbar stress fracture in the whole corpus. One visible frame, upstream of the fault, fixable with lumbopelvic control work rather than a technical cue.
• Source: Bull et al. (2026), 03/; injury link from Alway et al. (2021), 02/.

Bound take-off angle — check, don’t prescribe

• Reference values: 11 ± 3°; bound length ≈122 ± 18% of standing height.
• The catch: a lower take-off angle gives more speed at BFC and a better front-leg plant angle; a higher angle gives a straighter front knee at release. No angle optimises both, and no bound variable predicts ball speed at all. Any squad-wide bound instruction is wrong for someone.
• Source: Bull et al. (2026), 03/.

Frame 2 — Back foot contact

Back leg collapse (SIDE-ON) — the strongest injury cue there is

• Look for: the angle at the hip — trunk down through the thigh — as the back foot lands.
• Good: tall and extended over the back foot, riding over it. Uninjured group 156° at the hip and 156° at the knee; the lowest-risk bowler in the study was at 170°.
• Fault: sat down into the back foot — hip and knee both folded. Injured group 146° / 146°, a full 10° more bent at each. The worst bowler was at 123°.
• Why: ten degrees, clearly visible on a tablet. Moving from 156° to 146° raises the odds of lumbar bone stress injury from 1.0 to 4.8. This parameter alone correctly classifies 76% of bowlers.
• Important caution: the paper does not say “tell them to straighten the back leg.” It argues the flexed position is a consequence — the bowler flexes to reach the joint range where more torque is available, because they need it to control the pelvis. Treat a collapsed back leg as a signal to investigate strength, control and run-up, not as a cue to shout.
• Source: Alway, Felton et al. (2021), 02/.

Trunk side flexion and rotation at BFC (FRONT-ON)

• Injured bowlers arrive side-flexed towards the bowling arm and rotated away from it (182° / 177°); uninjured bowlers are the mirror image (179° / 182°). Medium-to-large effects.
• Source: Alway, Felton et al. (2021), 02/.

Frame 3 — Front foot contact

Pelvis and low back (SIDE-ON) — the other half of the 88% model

• Look for: whether the pelvis tips forward and the small of the back hollows as the front foot loads.
• Good: pelvis relatively upright (175°), lumbopelvic junction not driven into extension (172°).
• Fault: belt line tips forward, backside out, and the low back arches sharply just above the waistband while the front thigh comes up toward the chest. Injured: pelvis 170°, front hip 130° (vs 137°), lumbopelvic 176°.
• Why: each 1° of extra lumbopelvic extension multiplies injury odds by 1.25. Combined with a flexed rear hip at BFC, the odds ratio goes from 4.8 to 11.5, and the pair classifies 88% of bowlers.
• Possible mechanism: the bowler may be buying ball speed with lumbar extension — the position sets up a bigger trunk flexion into release. If so, this is a genuine performance-injury trade, not just a fault.
• Source: Alway, Felton et al. (2021), 02/.

Front knee at landing (SIDE-ON) — performance, not injury

• Good: straighter — elite female squad averaged 167.0 ± 2.9°; correlates with ball speed in women at r = 0.68 (height controlled).
• Note: front knee angle at FFC did not differ between injured and uninjured bowlers (163° vs 163°). This is a pace cue, not a back-safety cue.
• Source: Lyons, Felton & McCabe (2023), 03/; Alway et al. (2021), 02/.

Time budget (ANY ANGLE — just count frames)

• BFC→FFC ≈192 ms and FFC→BR ≈103 ms in elite men; ≈172/128 ms in elite women. Large effect size between sexes on the split, with the same total. A cheap, robust measurement that needs no angles at all.
• Source: Felton et al. (2019), 03/.

Frame 4 — Ball release

Arm delay (SIDE-ON) — the strongest single pace correlate found in any sample

• Look for: how late the bowling arm is still coming through at release.
• Good: delayed. In 11 elite female bowlers this alone explained 89% of the variance in ball speed (r = 0.95, p < 0.001) — worth 0.224 m/s per degree, roughly 0.8 km/h per degree.
• Film it: side-on, square to the plane of motion, at the release frame.
• Health warning: this is 11 bowlers, and the 2015 conference paper on female bowlers found the opposite (faster women had earlier arm action — but measured at FFC, not release). See the Contradictions page, §19.
• Source: Lyons, Felton & McCabe (2023), 03/.

Where the sideways lean comes from (FRONT-ON) — subtle, and counter-orthodox

• Look for: when the bowler leans away from the bowling arm to get the hand high, which part of the trunk bends.
• Good: the lean is sourced high, as a smooth curve through the mid-back (thoracolumbar). Uninjured bowlers had more contralateral lean at that joint (160 ± 3°).
• Fault: a stiff mid-back with a sharp kink just above the belt instead. Injured: 163 ± 4° at the thoracolumbar joint (less lean there), 174 ± 5° at the lumbopelvic junction (more lean there).
• Why: both groups reach the same total lean. The injured group does more of it immediately adjacent to where the fractures occur.
• Confidence: the lumbopelvic half of this reached only p = 0.09, d = 0.57. It fits the injury location and it is plausible, but it is inferred. Treat as a watch-item, not a diagnosis.
• Source: Alway, Felton et al. (2021), 02/.

Ipsilateral pelvic drop (FRONT-ON, tracked BFC→BR)

• Good: pelvis stays level; uninjured max 190 ± 6°. Fault: bowling-arm-side hip sags; injured 195 ± 8°. Medium effect; part of the pelvic-control picture but it did not survive into the final model.
• Source: Alway, Felton et al. (2021), 02/.

Elbow hyperextension (FRONT-ON or behind-the-arm)

• Worth roughly 0.2% of ball speed per degree past 1° — about 5%, or ~5 mph, at 20°. Legal under ICC rules; Felton sits on the ICC panel that adjudicates this.
• But: his 2025 work found that increasing strength drove hyperextension to every bowler’s ceiling, and warns of posterior elbow impingement and bone stress injury. Screen for it, then monitor it — especially after a strength block.
• Source: Felton & King (2016), Felton et al. (2025), 01/.

Length control — “he keeps dropping short”

The headline: bowlers change length by changing the ball’s upward tilt at release, not by changing pace. Release angle: yorker ~2°, stock ~5°, bouncer ~13° (η² = 0.78). Horizontal ball speed did not differ between them (p = 0.07); the entire difference is vertical (1.2 / 3.0 / 7.4 m/s, η² = 0.77).

Check in this order:

1. Wrist at release (side-on). Wrist angle alone explained 75.3% of yorker release-angle variance (86.6% with hand orientation). Yorkers carry a more extended/cocked wrist from back foot contact onward, use ~6° more travel, and are fastest at release (~16 vs ~12 rad/s).
2. Finger/ball line in the final frame. ~58° yorker, ~52° stock, ~42° bouncer. The thesis’s own nominated coachable-by-feel cue: stay behind and under the ball to the last instant.
3. Upper-back posture — this sets release height (55–60% explained), not angle. Don’t confuse the two.
4. Check nothing else changed. Front knee, front and back hip, back knee, lumbar angles, stride length and COM velocity showed zero difference at any moment across lengths. Length variation is a last-20%, upper-body event. If you are re-coaching the run-up to fix a length problem, you are in the wrong place.

Before you blame the bowler: yorker success was 24.8%, mean miss 3.8 ± 3.3 m, and a 0.005 s timing shift moves a yorker 5.68 m. The best bowler averaged 0.73 m, the worst 6.76 m; six of 21 averaged over 5 m. That spread is a selection question as much as a coaching one.

Cost of a variation: none detectable. Yorker vs stock ball speed p = 0.710; bouncers were faster. Front-foot loading identical on every measure (p = 0.238). But that is a null on n = 21 — absence of evidence.

Source: Manawadu et al. (2022), Manawadu et al. (2023), and Manawadu (2023) thesis, 04/.

Workload monitoring

• An ankle IMU is a good proxy for impact exposure — FFC ~150 g, follow-through ~110 g, BFC only ~56 g.
• It is a poor proxy for lumbar load: 80–91% of the shock is gone by L5, and the surviving fraction varies bowler to bowler.
• Count indoor and artificial-surface deliveries in workload. Kinematics on artificial pitches are indistinguishable from grass, so those deliveries load the bowler the same way.
• Peak 7-day workload discriminated injury (229 vs 165 balls) where season totals did not. Watch the spike.
• Age is the risk factor. Every prospective adolescent injury occurred at 17 or 18 — 80 per 100 bowlers/year at 18, and zero at 14–16 — driven by chronological age (g = 1.396), not skeletal maturity (g = 0.274).
• Source: Lamb et al. (2022), Lamb et al. (2023), 04/; Alway et al. (2024), 03/; Keylock et al. (2022), 02/.

Off-field screening (no video needed)

• Front-shoulder internal rotation (75 ± 8°) is the most informative single measure — restriction predicts three separate injury-profile technique faults, including lumbopelvic extension at FFC.
• Bowling-shoulder external rotation (125 ± 11°) — restriction predicts a flexed rear hip at BFC.
• Ankle dorsiflexion (109 / 103 mm).
• Sit-and-reach and straight-leg-raise predict nothing. Stop using them for this.
• Source: Felton, McCaig & King (2023), 02/.

Artificial vs grass nets — does your winter footage transfer?

Yes. Ball speed 30.2 vs 30.4 m/s (p = 0.438), and no significant kinematic difference at any joint at any point across the whole delivery (whole-waveform analysis). The only difference is foot slide (+1.7 cm at BFC, +2.1 cm at FFC), largely attributable to footwear.

Caveat: n = 8, and the authors call it underpowered. A null on 8 bowlers rules out large effects only.

Source: Alway, Felton et al. (2024), 03/.

COACHING FEMALE CRICKETERS

The men’s model does not transfer. This is the clearest cross-cutting message in the corpus.

• Trunk flexion — a pillar of the male ball-speed model — predicts nothing in women (r = −0.19, p = 0.57), across a 33° spread. Not a small effect; no effect.
• Run-up speed is the one variable that works in both, but in women it collapses from r = 0.75 to r = 0.41 (n.s.) once height is controlled. Individualise by height.
• Height and arm length predict ball speed (r = 0.76 / 0.61). Leg length and body mass predict nothing.
• Men and women don’t differ in mean front knee angle — only in spread (SD 24.9° vs 18.8°).
• In batting: all 15 male batters extended the lead elbow through the downswing (30 ± 12°); 8 of 15 female batters flexed it, closer to a checked drive. The authors explicitly decline to call this a fault — it may be a rational response to shorter boundaries, different bat moment of inertia, or strength.
• The 78%-of-bat-speed model is male-only. Whether those variables predict distance within female batters has never been tested.

The honest position: the female fast bowling literature here is three small studies that contradict each other on arm timing and front knee (see the Contradictions page, §19–23), including two opposite results from the same 11 bowlers. Coach from observation and individual response, not from these numbers.

SPIN BOWLING

What video can and cannot see — read this first

Sanders placed four finger markers specifically to measure finger action and had to discard them: 18 cameras at 300 Hz could not track them. So a phone cannot resolve finger action, wrist flexion, upper-arm internal rotation, or legality. Full stop.

The consolation: the variables that did predict spin — pelvis line, shoulder line, hip–shoulder separation, foot orientations — are exactly what a camera sees best. But they cap out at about 43% of spin-rate variance. The other ~57% is in a hand you cannot film.

Practical implication: buy a radar, not a better camera. And track spin rate and release speed together — drift and dip come from Magnus force, which needs both.

Coachable targets from the unpublished Hawk-Eye analysis

From 69,552 deliveries across 60 Test matches, in the Sanders thesis and published nowhere else:

• Length 4–5 m; line within 0.15 m of middle stump.
• Release speed up to 56–57 mph, but no further — line and length scatter break down sharply above that (r = 0.934).
• Turn it away from the batter — 24% better bowling average.
• Flight height did nothing across all 69,552 balls.

Finger spin vs wrist spin

They invert. Pelvis and shoulder orientation at release carry opposite signs (wrist spinners: less shoulder rotation, shoulders short of chest-on at release, r = −0.837 / −0.875). Do not transfer finger-spin coaching to a leg-spinner.

Off-field

Passive hip and shoulder range of motion correlates with spin rate — a screening and training target that needs no camera at all.

One trade-off: the standard advice for reducing elbow extension in a suspect action (be more side-on) is exactly what lowers spin rate. A bowler cleaning up an action should expect to lose revolutions.

Source: Sanders, Felton & King (2018), Sanders et al. (2019), Sanders (2019) thesis, 05/.

BATTING / POWER HITTING

Distance is set by launch angle and launch speed. Launch angle is bat angle at impact (R² = 0.83). Launch speed is impact location (48% of variance) then bat speed (68% for the pair).

Impact location beats bat speed

• The sweet spot is 17.5 cm from the toe of the bat, on the midline.
• Within 2 cm across the face and 4.5 cm up it, you lose <6% ball speed and <10° of direction. That is the size of the target.
• The tell without a marked bat: the bat visibly twisting in the hands at contact, and the ball leaving in a different direction from where the face was pointing.
• The instruction: do not chase bat speed at the expense of timing. The paper says so outright.

Three cues that explain 78% of between-player bat speed (male batters)

1. Hip–shoulder separation at the top of the backswing — 28%, the largest single contributor. Film overhead or front-on; this is transverse-plane rotation and is nearly invisible side-on. Good: shoulders turned well past the hips, hips already opening while the chest stays closed. Fault: hips and shoulders turning as one block.
2. Lead elbow extension through the downswing — front arm visibly straightening. (Male finding — see the female-cricketers section.)
3. Wrist uncocking late — held, then released. Fault: casting early.

The ranking (torso > elbow > wrist) is the signature of a proximal-to-distal sequence. Sequencing check: the swing should start at the torso, not the hands.

Bowling-machine footage lies to you

Against a machine, pelvis–thorax separation at the top of the backswing is at its lowest, and wrist action at its highest — the later the visual cues arrive, the more distally-dominant the technique becomes. Do not diagnose “no torso rotation” from bowling-machine footage. Judge separation against a real bowler.

Source: McErlain-Naylor, Peploe, Felton & King (2022) BASES article, 08/; McErlain-Naylor et al. (2021), 05/.

WHAT TO STOP COACHING

Each of these was tested and failed.

TWO WARNINGS

1. Almost none of this is causal

These are studies of bowlers who already had certain techniques and certain outcomes. Barely anyone was coached to change a variable and re-measured. The evidence is strong enough to stop screening for the mixed action. It is not strong enough to claim that straightening a back leg prevents fractures.

The injury work compares 39 injured to 11 uninjured bowlers, all male, mean age 18.9, English elite pathway. The female work rests on samples of 11 to 15. The length-control work is 21 bowlers. The artificial pitch study is 8.

And the entire simulation programme in 01/ — every ball-speed prediction in it — has never been tested in the field. The follow-up study was proposed in 2017 and has not appeared.

2. What your camera can and cannot tell you

From the modelling work in 06/, error magnitudes for a planar side-on model:

Fast bowling is the worst case for single-camera analysis, precisely because it is a side-on action with real out-of-plane hip and shoulder motion that one camera cannot recover. Angles and timings: yes. Forces: no.

So: measure what the camera is good at — angles at four frames, and the frame counts between them. That is where every reliable cue in this document lives.