Every place across the corpus where two documents point different ways. The individual paper pages carry these inline, labelled Superseded, Unreconciled and Open question.
About this page. Nothing here comes from re-analysing data. This is what shows up when one reader takes a whole published record in sequence — which is a thing no reviewer, examiner or funder is ever asked to do. Most of what follows is a research programme working correctly: a group published, learned something, and published better. That is filed separately from the cases where two documents in the current record still point different ways.
On attribution. This catalogue is organised around one publication list because that is the only tractable boundary for a corpus this dispersed — not because it represents a single research agenda. Roughly half of these papers are student-led, with Felton in a supervisory or third-author role, and two of the doctoral projects here never cite each other. Where this page notes a divergence, that is a property of the literature, not of any author’s reasoning.
The three labels used throughout:
| Label | Meaning |
|---|---|
| Superseded | A later study covers the same question with a larger or more controlled design. |
| Unreconciled | Two documents in the current record point different ways, and no published work reconciles them. |
| Open question | No study in the record resolves it. |
Where the record differs from established coaching orthodoxy
These are not internal disagreements. They are the places where this body of work tested a previously accepted account and the finding ran the other way.
1. The mixed action does not predict lumbar bone stress injury
Alway, Felton et al. (2021), MSSE — 50 elite male bowlers, prospective, MRI-confirmed.
Shoulder counter-rotation: 43 ± 14° (injured) vs 40 ± 20° (uninjured), not significant. Pelvis–shoulder separation at BFC: 21 ± 15° vs 13 ± 24°, not significant. The injured group averaged above the 30° “mixed action” threshold from Portus et al. (2004) that has driven UK and Australian coaching since the 1990s — and it did not separate them from the bowlers who stayed healthy.
The paper’s own conclusion: “Coach education should incorporate these findings and move away from using far removed derivatives to inform practice which are not consistent with predicting LBSI injury.”
What replaces it: rear hip angle at back foot contact and lumbopelvic angle at front foot contact — 88% classification accuracy.
2. Excessive contralateral trunk side flexion runs the opposite way to the received account
Bayne et al. (2016) and Ranson et al. (2008) linked excessive contralateral thoracolumbar side flexion to injury. Alway et al. found injured bowlers were significantly less contralaterally side flexed at the thoracolumbar joint at release (163 ± 4° vs 160 ± 3°, large effect). Both groups reach similar total lean; the injured group sources more of it one joint lower, at the lumbopelvic junction adjacent to where the fractures occur (p = 0.09, d = 0.57 — the paper’s weakest number, and inferred).
3. Ground reaction force did not distinguish injured from uninjured bowlers
No GRF measure differed between injured and uninjured — peak vertical ~6.9 vs 6.8 bodyweights, plus horizontal, loading rates and impulses, all non-significant. Coaching a “softer” front foot landing is not supported as an injury intervention. Nor did pace: release speed 35.1 vs 35.8 m/s, non-significant.
4. Injured adolescents had more bone, not less
Keylock, Felton et al. (2022) — counter-intuitive and worth sitting with. Alongside it: every prospective injury occurred at 17 or 18 (80 per 100 bowlers/year at 18; zero at 14–16), driven by chronological age (g = 1.396) rather than skeletal age (g = 0.274). And peak 7-day workload discriminated (229 vs 165 balls) where season totals did not — it’s the spike, not the volume.
5. Front-leg bracing is unsupported wherever it was actually tested outside fast bowling
Sanders tested the braced-knee cue directly in finger spin bowling — front knee extension against spin rate, r = 0.156, p = 0.477 — and found nothing; the average elite finger spinner’s knee flexes rather than braces. That part is sound. The batting evidence is weaker than it looks: McErlain-Naylor et al. (2021) tested a male-vs-female group difference in lead knee kinematics, not knee bracing against bat speed or carry distance, so a null sex difference doesn’t speak to whether the front knee predicts hitting performance — and even descriptively the female lead-knee change is close to flat (−0.1 ± 9.7°), not a clear flex. The cue’s provenance is also usually mis-stated: Sanders traces the spin-bowling brace principle to Brayshaw (1978) and Woolmer & Noakes (2008), both spin-coaching texts — it looks like an independent cue in the spin literature, not one borrowed from fast bowling.
6. Finger spin and wrist spin invert
From the unpublished elite wrist-spin study in the Sanders (2019) thesis: pelvis and shoulder orientation at release carry opposite signs between the two (wrist spinners: less shoulder rotation, shoulders short of chest-on at release, r = −0.837 and −0.875). The two spin types load the same parameters in opposite directions.
7. Trunk flexion is a male-only predictor
A pillar of the male ball-speed model. In women: r = −0.19, p = 0.57, despite a 33° spread across the sample. With only 11 bowlers the 95% CI runs from −0.71 to 0.46 — too wide to rule out a moderate relationship in either direction. What is established is that trunk flexion did not emerge as a predictor in this female sample.
8. “Run in faster” fails upstream, within a narrow elite band
Pre-bound run-up speed predicts arrival speed at back foot contact at r = 0.844 but ball speed at r = 0.262 (n.s.). Within this elite, narrow-band sample (release speed 35.4 ± 1.33 m/s), run-up speed did not predict ball speed — but that is not the same as run-up speed ceasing to matter: the null sits inside a restricted range, and Bull himself cites Worthington’s finding that run-up velocity at back foot contact is the strongest predictor of release speed — a different measurement point from the pre-bound velocity that produced this null. In women, run-up speed correlates at r = 0.75 — but collapses to r = 0.41 (n.s.) once height is controlled.
Superseded — where later work replaced earlier work
In each case a later, larger or better-controlled study covers the same question as an earlier one.
9. The strength conclusion between the 2024 conference note and the 2025 journal paper
The two studies tested different things. The 2024 ISBS paper ran a single condition — isometric strength increased 5% — giving 40.7 → 41.1 m/s, and concluded isometric strength does not limit technique in elite bowlers. The 2025 J Sports Sci paper ran three conditions (original; +5% lower body → 41.3 m/s; +5% lower body and shoulder → 41.5 m/s) with one-way ANOVAs and post-hoc tests, and found no significant joint kinematic differences, though the increased-strength techniques exhibited non-significant trends its own authors describe as “contrary to expectations” — less knee extension, reduced trunk flexion, greater shoulder extension — closing with the advice that “caution is advised when considering using strength interventions to alter the front foot contact-phase technique.”
These are not competing estimates of one quantity. The 2024 conference paper re-optimised the joint torque activation parameters only, inheriting the optimal landing position from the 2023 study. The 2025 journal paper re-optimised all 112 parameters in every condition, including the initial joint angles, angular velocities and trunk orientation. Different optimisation scopes give different ceilings.
The 2025 paper is the journal version, with three conditions and the wider optimisation.
10. The individual-specific claim, refined by the group studies
2017 argued optima are individual, so general technical models mislead. The two 2023 papers found genuine commonalities across elite bowlers. Resolution: the direction of the optimum is shared; the attainability is individual. Worth stating carefully rather than as a flat reversal.
11. The 22%, the 9.8% and the 13.5% are three different quantities
These figures are frequently read as a single estimate that shrank. They are not. Verified against the source PDFs:
- 2015 conference — 22%. An optimisation that included the initial body configuration. The same paper reports 10% for the initial optimisation before configuration was optimised.
- 2020 journal — 9.8% (38.8 vs 35.3 m/s). Optimising activation timings against a matched simulation. This paper states explicitly that varying the initial position of the bowling arm was outside its scope, so it could not have reported the 22% quantity.
- 2023 journal — 13.5%, ten bowlers, optimising initial position and movement pattern.
The 22% and the 13.5% are the comparable pair: both optimise the initial configuration. The 13.5% is the peer-reviewed, ten-bowler version; the 22% is a single-bowler conference result.
12. The planar-model story was materially revised
2016 called the method “suitable” while reporting no plain-planar baseline. 2019 supplied the baseline and showed the unmodified assumption was substantially inadequate (18% force error, 23% vertical). The later work does not reverse the earlier conclusion, but it materially revises the impression it leaves: planarity was a real problem, not a benign simplification. Separately, 2019 calls 3D “non-viable” and planar the solution, while the 2021 review reframes 2D as a temporary stopgap.
13. Power hitting: the separation plane between conference and journal
Conference: frontal plane. Journal: transverse plane. Rear elbow extension went from “very strong” to non-significant (p = 0.086). The journal analysis covaries out height and mass; the males were +12 cm and +12 kg on the females.
14. Rear hip flexion timing
The 2020 single-bowler optimisation produced earlier rear hip flexion; the 2023 ten-bowler study produced delayed rear hip flexion. Same model, same lab. The 2023 paper cites the 2020 mechanism and proposes that delaying rear hip flexion delays upper trunk flexion until after front foot contact, for a more efficient transfer of momentum through the kinetic chain. Both papers state the function of rear hip flexion is uncertain.
15. The same 11 bowlers, two sets of results
Lyons 2022 poster vs Lyons 2023 journal. Run-up speed: non-significant on the poster; the journal’s zero-order correlation is r = 0.75, p = 0.01, collapsing to r = 0.41, p = 0.28 once height is controlled. That is a genuine poster-vs-journal difference, and it isn’t resolved: the poster does not say which correlation it was reporting.
Front knee at FFC is a smaller story than it first looks. The journal’s own bivariate correlation is r = 0.49, p = 0.13 — non-significant, which agrees with the poster. It only becomes significant, r = 0.68, p = 0.04, once height is controlled — a correlation the poster never ran. So the front-knee change is a bivariate-vs-partial distinction inside the journal, not a poster-vs-journal disagreement.
Same cohort. The journal is the peer-reviewed version, adding a full correlation table, partial correlations controlling for height and arm length, and formal stepwise regression.
16. The thesis and the 2025 paper differ on what strength buys
The thesis and the 2015/2017 conference papers argued strength lets the front leg stay straighter, from a single-bowler design. The 2025 paper — ten bowlers, three strength conditions, ANOVA with post-hoc tests — found the opposite trend. The later, larger, more controlled design supersedes the earlier one on this question.
Unreconciled — two citable documents, no stated resolution
In each case both documents stand in the current record and no published work reconciles them.
17. Does impact shock cause the lumbar injury?
Every Lamb paper opens by framing impact shock as the risk mechanism — and the 2023 ISB paper concludes the elastic wave is “unlikely to contribute to lumbar stress fracture risk.” The premise of the workload- monitoring line and the conclusion of one of its own papers point different ways, and no paper in the set reconciles them.
18. Two accounts of what the 2020 simulation model can be used for
All three Lamb papers name Felton, Yeadon & King (2020) as performance-focused and argue rigid pin-jointed models are unsuitable for estimating internal loading. That is the explicit motivation for the compliance model. Both accounts come from the same group. The 2020 model was built and validated against kinematics and ball speed; the compliance papers address internal loading, which it was not built to estimate.
19. Model compliance parameters
The 2019 cricket model allowed 4.5–10 cm of wobbling-mass movement, and separately allowed the front foot itself to displace up to 6 cm vertically / 9 cm horizontally before a penalty applied. The 2021 review cites measured soft-tissue displacement of 1.4 cm and measured heel/shoe deformation of 11.5–12.7 mm. The model’s compliance limits sit several-fold above the measured values, in exactly the parameters that generate the force numbers. The review states this itself rather than it being an outside observation, which is the right way for it to have surfaced.
20. Female arm timing: earlier or later?
2015 conference: faster women had earlier circumduction (measured at FFC). Lyons 2023: a more delayed arm is the top predictor (measured at BR, r = 0.95, 89% of variance in 11 bowlers, 0.224 m/s per degree). The two were measured at different instants, which may account for the difference; no published work reconciles them.
21. Female front knee: straighter or more flexed?
2015: more flexed at BR in the faster women. Lyons 2023: front knee at FFC is not a significant bivariate predictor (r = 0.49, p = 0.13); only once height is controlled does it become significant, and in the straighter direction (r = 0.68, p = 0.04). That weakens rather than removes the disagreement with 2015 — the two studies differ in the instant measured (BR vs FFC), and Lyons’ claim now also rests on a partial correlation that 2015 has no equivalent of. The 2019 paper shows the sexes don’t differ on the mean at all — only on the spread (SD 24.9° vs 18.8°), which is itself interesting: women are more variable here, not different on average.
Items 20–21 above, plus the Lyons poster-vs-journal item in the Superseded section (§15), all sit in the female fast-bowling line: three studies, small samples, and no reconciliation published.
22. Attenuation beyond L5
Lamb finds shock attenuation continues L5 → L1 → C7 (p < 0.01); McErlain-Naylor, King & Allen (2021) found none beyond L5 in drop landings. McErlain-Naylor co-authors both, and the two use different tasks.
23. Hip internal rotation
Across three studies in the injury cluster, hip internal rotation findings do not align. No resolution is offered in any of them.
Open questions — genuinely unresolved
Nothing in the published record settles these. They are the frontier, not a failing.
24. Peak ground reaction force in the 2023 papers
The 2023 journal abstract states the optimised techniques resulted “in lower peak ground reaction forces and loading rates”. In its own Table 1 the loading-rate differences reach significance (horizontal braking p = 0.008, vertical p = 0.022) but the peak-force differences do not (horizontal braking p = 0.394, vertical p = 0.093).
Fig. 1 is the SPM1D continuous waveform analysis of the same force curves, and the paper’s own Results section reports what it found there: “no significantly different periods between the ground reaction time histories were observed.” So neither the discrete peaks in Table 1 nor the continuous analysis in Fig. 1 found a significant peak-force difference — only the loading rates did. The abstract’s “lower peak ground reaction forces and loading rates” is supported for loading rates and not for peak forces.
25. The front leg: speed versus shock
The central unresolved trade-off in fast bowling. A braced, straight front leg produces more ball speed (Felton, Yeadon & King 2020). A compliant front leg attenuates shock (all three Lamb papers; causally confirmed in simulation — stiffen the leg 100× and shank acceleration goes 76 → 197 g). Same joint, same 100 ms of the action, opposite advice.
For scale: at a 10× stiffening, the rise at L5 is only +5%. At 100×, the shank rises +158% while L5 rises +43% — stiffening hurts the leg’s own protection far more than it loads the back. The ISCSB 2023 conference paper stated that the simulation’s spring-stiffness optimisation — the piece that could price the exchange rate — would be presented at the conference; that optimisation does not appear in any accessible published source. It is presumably embargoed inside Lamb’s PhD thesis, which is not accessible until 1 October 2027.
Worth naming the reason this gap exists: the performance line and the injury line are separately funded, separately reviewed and published in different journals, so no single paper has ever been required to hold both. That is a structural feature of how the field is organised rather than an oversight by any author — and it is the strongest argument for reading a corpus whole.
26. Elbow hyperextension: free speed, or an injury pathway?
2016 values it at ~0.2% ball speed per degree (≈5%, ~5 mph, at 20°) and ICC-legal. 2025 finds increased strength drove hyperextension to every bowler’s ceiling and warns of posterior elbow impingement and bone stress injury. The same mechanism, load-bearing for performance and for risk, with no study pricing the exchange.
27. Trunk rotation both builds bone and marks the injured
The same trunk rotation that best predicts high lumbar bone mineral density is also significantly greater in the injured group. Adaptation and damage may be the same signal at different doses.
28. Thoracic flexion does two jobs
More flexion → more ball speed (Worthington 2013) and → shorter length (Manawadu). Coach it for pace and length may creep short. Raised in the Manawadu thesis and not yet investigated.
29. Spin rate versus legality
Spratford’s advice for reducing elbow extension — be more side-on — is exactly what lowers spin rate. Sanders raises this directly. A bowler cleaning up a suspect action should expect to lose revolutions.
30. Optimal technique — does it raise or lower front-foot load?
Depends on whether the landing position was also optimised. Not yet reconciled within the simulation cluster.
31. Robustness of the optima
The 2021 review sets the standard that optima should be robust to perturbation and notes uptake of it “has been sporadic”. The cricket optimisations are single-best-performance optima, so the standard the review sets has not yet been applied to them. Related: the review credits in-vivo dynamometry with giving “assurance” that torques are realistic, while Parkinson (2022) shows single-angle protocols can be 30–96% wrong. The cricket work used multi-angle protocols and is unaffected; the review’s assurance should not be transferred to studies that used single-angle testing.
32. Objective function determines the answer
In gymnastics, minimising joint torque diverged from what the elite athlete actually did; maximising success under variability converged on them. Always ask: optimal for what?
33. Bound take-off angle has no single right answer
Bull et al. (2026): a lower take-off angle gives more speed at BFC and a better front-leg plant angle, and no bound variable predicts ball speed at all. The authors resolve the relationship as non-linear, so there is no angle to prescribe across a squad.
34. Sex differences in batting may be adaptation, not fault
All 15 male batters extended the lead elbow (30 ± 12°); 8 of 15 female batters flexed it. 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. And the 78%-of-bat-speed model is male-only: whether those variables predict distance within female batters has not been tested.