REVIEW 4 major objections 5 minor 12 references
Remember when? Deciphering Ediacaran-Cambrian Metazoan behaviour and temporal memory using fossil movement paths
T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash
Pith's one-line read Fossil trail turns show early Cambrian grazers already used time-tuned, memory-like behaviour while Ediacaran trails show no temporal correlation.
desk verdict Clever quantitative idea applied to fossil trails, but the distance-to-time mapping is load-bearing and unverified, so the memory conclusion overreaches. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The key object is the autocorrelation function of turning angles along a digitized fossil path. Trails are traced and divided into segments whose length is a multiple of the trail width; under a constant average velocity, each segment is treated as roughly one unit of time. The autocorrelation at lag h asks whether a turn predicts a turn h units later, and partial autocorrelation removes the influence of the lags in between. The span of an initial sampling-induced positive correlation is used to estimate the distance of one 'action step'—a single locomotory cycle—and any correlation that persists beyond that span is read as evidence for internally driven, time-tuned behaviour.
What would settle it
Track a living grazing or burrowing animal whose speed is known to vary, discretize its path under the same constant-average-velocity assumption, and compute the autocorrelation of turning angles. If a mechanically generated or memory-free path yields positive-then-negative autocorrelation like the Cambrian trails, the inference collapses. A complementary check is to simulate paths with and without time-tuned turning rules and ask whether the method reliably separates them at fossil sample sizes.
Extended reading notes
Core claim
The central discovery is a statistical signature in fossil movement paths. After discretizing trails into equal segments and computing turning angles, the autocorrelation function shows that in Helminthoidichnites tenuis and Parapsammichnites pretzeliformis from the Ediacaran, turns are essentially independent of turns taken one to several steps earlier, apart from a brief sampling-induced positive correlation and a tentative short-range anticorrelation. In Psammichnites cf. saltensis and Psammichnites gigas circularis from the early Cambrian, turns are positively autocorrelated at short lags, and P. cf. saltensis shows a prolonged anticorrelation extending out to roughly 28 trail-widths. Th
Load-bearing premise
The load-bearing premise is that distance along the trail is a clock: the analysis assumes each tracemaker moved at roughly constant average speed, so equal distances stand for equal time intervals, and it further assumes that any correlation persisting beyond one locomotory cycle is an internally generated behavioural signal rather than a mechanical, geometric, or substrate artefact.
Editorial extensions
If this is right
- Autocorrelation analysis of fossil trails can serve as a new proxy for navigation capacity, extending the record of sensory evolution beyond body fossils and qualitative trail descriptions.
- Similar-looking trails can hide different temporal strategies: the two 'looping' ichnospecies, one Ediacaran and one Cambrian, have different autocorrelation patterns, so morphological similarity alone is not a reliable guide to behaviour.
- The decay length of the sampling-induced autocorrelation gives an estimate of the action-step distance, potentially a general way to infer locomotory cycle lengths from fossil trails.
- Time-tuned behaviour in the early Cambrian would support the idea that increasing environmental information complexity—not only ecological or anatomical change—was a distinct evolutionary force in the Cambrian radiation.
Reading between the lines
- Beyond the paper: applying the same analysis to later Phanerozoic trails could map when time-tuned behaviours spread across environments and lineages.
- Beyond the paper: modern tracking data on variable-speed grazers could test whether the constant-velocity assumption is safe; if it is not, the spatial-to-temporal mapping would need a velocity model.
- Beyond the paper: coupling autocorrelation with periodicity detection could separate clock-driven rhythms from memory-like search strategies, a distinction the paper notes is unresolved.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript applies autocorrelation and partial autocorrelation analyses to turning angles extracted from four sets of fossil grazing trails: two Ediacaran ichnospecies (Helminthoidichnites tenuis, Parapsammichnites pretzeliformis) and two early Cambrian ichnospecies (Psammichnites cf. saltensis, Psammichnites gigas circularis). Trails are digitized, divided into equal-length segments scaled by trail width, and the authors assume that each segment corresponds to an approximately constant unit of time. They report that the Ediacaran trajectories show no temporal autocorrelation beyond short lags, whereas both Cambrian trajectories show positive and/or negative autocorrelation over longer lags. This is interpreted as evidence that time-tuned behaviours and possible biological memory appeared by the early Cambrian, supporting the Cambrian Information Revolution hypothesis. The paper also proposes that the length of sampling-induced positive autocorrelation can be used to infer an organism's 'action step'.
Significance. If the central inference were valid, this would be a valuable quantitative contribution to debates about behavioural and cognitive evolution across the Ediacaran–Cambrian transition. The study is conceptually novel in applying autocorrelation methods, common in movement ecology, to fossil trails, and it makes explicit, falsifiable predictions about Ediacaran versus Cambrian behaviour. The authors are also transparent about some interpretative caveats, such as the distinction between statistical and biological memory. However, the central claim that the observed spatial autocorrelation reflects temporal, time-tuned behaviour depends on an unverified and likely unverifiable assumption of constant movement speed, and the analysis lacks formal hypothesis tests and null-model controls. These issues are load-bearing: without them, the paper documents spatial autocorrelation patterns in trace fossils, but not the temporal memory or time-tuned navigation claimed in the title and abstract.
major comments (4)
- [§4.3 Methods, 'For our specimens we assumed...'] The distance-to-time mapping is the core assumption of the paper, and it is unverified. The authors state that they 'assumed that an average velocity provided a reasonable approximation of the velocity distributions of each tracemaker' and that 'Each equidistant segment can be interpreted to represent the passage of an approximately constant unit of time.' Trace fossils preserve only the path, not the pace. If the tracemaker varied its speed while forming the trail—pausing, slowing at turns, or reacting to substrate—equal spatial segments would correspond to unequal time intervals. The autocorrelation of turning angles computed from distance-based lags would then mix spatial and temporal structure, and the patterns interpreted as 'temporal autocorrelation' and 'time-tuned behaviour' could arise from speed variation without any memory or timing mechanism. No sensitivity analysis, simulati
- [§4.4 Results and Figure 4.5] The 'action step' is inferred from the very autocorrelation curves that are subsequently interpreted. The text states that the distance at which sampling-induced autocorrelation ends 'could indicate the typical distance travelled in a single action by the organism,' and the turning angles are then re-calculated and re-analysed using these inferred action steps. This is circular: the unit used to define action-step lags is derived from the same data that are then used to claim autocorrelation at certain action-step lags. The reported 'similar trend' in the action-step re-analysis is therefore expected by construction and does not independently validate the existence of action-scale temporal memory. An independent definition of an action step, or a demonstration that the results are robust to alternative choices of this parameter, is needed.
- [§4.4 Results and §4.5 Discussion] The manuscript lacks formal hypothesis tests, confidence intervals, and null-model controls. The analysis reports mean autocorrelation coefficients, standard errors of the mean, and boxplot summaries, and then makes claims such as 'greater evidence in favour of significant anticorrelation' (§4.4) without any significance test. A small standard error of the mean across specimens does not test whether the mean autocorrelation differs from zero or from the expected value under a null model. No null model is presented for the autocorrelation function of a random walk with the same path length, discretization, and measurement protocol. Without such a null, the observed Cambrian–Ediacaran difference cannot be distinguished from artifacts of segment length, trail geometry, or sample size. The number of specimens per ichnospecies is also not stated in the Methods or Results, so the statistical p
- [§4.2 Introduction, paragraph beginning 'The presence of temporal autocorrelation...'] The paper's interpretive axiom—that autocorrelation at time lags larger than the action step is internally driven and evidences time-tuned behaviour or biological memory—is asserted rather than established. The authors dismiss non-periodic external factors such as nutrient distribution or sediment consistency as unlikely to produce 'mathematically repeating locomotory patterns,' but the analysis actually measures spatial autocorrelation of turning angles, not periodicity. A meandering or looping trail produced by local taxis, by following a chemical gradient, or by a simple mechanical interaction with the substrate can generate positive and negative autocorrelations at multiple lags without any internal timer or memory. The paper's own discussion acknowledges that external memory systems (e.g., slime mould markers) can produce complex trajectories, yet it does not explain why such mechan
minor comments (5)
- [§4.3 Methods, ichnospecies list] There is a typo: 'Parapsammichnits pretzleformis' should be 'Parapsammichnites pretzeliformis'.
- [Figure 4.2 caption] The caption lists two panels labelled '(B)': one for the trajectory and one for the scatterplots. The second should presumably be '(D)' or another letter, and the panel labels in the figure should be checked.
- [Supplementary Information references] The text refers to 'Supplementary Information, Figures 4.10 & 4.11' and 'Figures 4.11 & 4.12' for partial autocorrelation functions, but the supplementary figures are numbered 4.16–4.19. These cross-references should be corrected.
- [§4.3 Methods, 'Chapter 3'] The discretization method is repeatedly referenced as 'Chapter 3' (a thesis chapter) rather than described in sufficient detail. Since this method is central, the key algorithmic steps, including image-to-curve conversion and segment resampling, should be summarized in the paper or appendices, and the referenced chapter should be made available.
- [Data availability] The text states that mean r and standard error data are available in Supplementary Data, but no data table or repository link is included in the manuscript. Given the quantitative nature of the claims, the raw autocorrelation values, specimen counts, and code should be provided in a permanent repository.
Circularity Check
Action-step normalization and an unpublished 'Chapter 3' method are load-bearing, but the core Ediacaran-Cambrian autocorrelation contrast is not itself manufactured.
-
fitted input called prediction
[Results, Section 4.4, paragraph after Figure 4.5; Figures 4.5 and 4.6 captions]
"The distance this autocorrelation ends at, therefore, could indicate the typical distance travelled in a single action by the organism. Turning angles can be subsequently re-calculated and re-analysed by these 'action steps' (Figure 4.5) and demonstrate a similar trend to that observed in the lag distance plots (Figure 4.3 & 4.4)."
The 'action step' is not an independent measurement; it is read off the same autocorrelation functions that the paper then interprets. The distance where the short-lag positive autocorrelation is judged to be a sampling artifact becomes the unit used for all subsequent statements ('after 2 action steps', 'past 3 action steps'). The boundary between artifact and 'real' autocorrelation is therefore estimated from the very curves under interpretation, making the action-step-based account a rescaling of a fitted value rather than an independent test. The raw lag-distance autocorrelations still show the Ediacaran-Cambrian contrast, so this is a partial circularity in the interpretive unit, not the sole source of the main conclusion.
-
self citation load bearing
[Introduction and Methods, Sections 4.2 and 4.3; repeated 'Chapter 3' references]
"These images were subsequently imported into MATLAB and discretized according to the methodology outlined in Chapter 3. This methodology converts images of trace fossil paths to 2D curves that can then be subdivided into equidistant segments along the fossil trajectory according to an inferred velocity distribution."
The discretization procedure is the first step of the entire derivation chain: it generates the point series, the segment lengths, and the 'inferred time' per segment on which every autocorrelation and partial-autocorrelation result depends. The manuscript does not describe or validate this procedure; it refers only to 'Chapter 3', an unpublished thesis chapter not available in the reference list. This is a load-bearing, self-referential methodological citation rather than an independent, checkable source. It does not make the Ediacaran-Cambrian contrast tautological, because the raw traced paths could in principle support the same contrast, but it is a central unverified input.
full rationale
The main empirical contrast is not circular: Ediacaran and Cambrian paths are traced, discretized, and subjected to standard autocorrelation and partial-autocorrelation analyses, and the raw lag-distance plots show the Ediacaran curves decaying toward zero while the Cambrian Psammichnites curves show positive/negative structure. That contrast does not reduce to the fitted action step. Two issues raise the score to 4. First, the action step is inferred from the same autocorrelation curves as the lag where the short-distance sampling artifact ends, then used as the unit for interpretation and to separate artifact from internally driven autocorrelation; this is a fitted parameter used to reinterpret the same data, but the raw contrast survives without it. Second, the path-discretization method is load-bearing and is referenced only as 'Chapter 3', an unpublished self-citation, so the first processing step that creates the time-equidistant points is not independently documented. The constant-velocity assumption is an auxiliary premise rather than circularity: if false, the distance-to-time translation fails, but that is a correctness/robustness concern, not an equivalence between input and conclusion. Similarly, interpreting autocorrelation as time-tuned behaviour is an inference, not a definitional tautology. Overall, there is some self-citation and a fitted unit that is partially self-defined, but the central Ediacaran-Cambrian autocorrelation claim retains independent raw content.
Assumptions & free parameters
free parameters (5)
- segment distance multiplier s =
0.1 and 1
- action step for Helminthoidichnites tenuis =
1.4 w
- action step for Parapsammichnites pretzeliformis =
0.9 w
- action step for Psammichnites cf. saltensis =
1.4 w
- action step for Psammichnites gigas circularis =
0.7 w
assumptions (5)
- domain assumption Each tracemaker moved at roughly constant average velocity, so equal spatial segments correspond to equal time intervals.
- domain assumption Linear Pearson correlation is a valid way to measure autocorrelation of turning angles.
- ad hoc to paper Autocorrelation at lags larger than the action step is internally driven and indicates time-tuned behaviour or biological memory.
- ad hoc to paper Non-periodic external factors, such as nutrient distribution or sediment consistency, cannot produce the observed autocorrelation.
- domain assumption Traced fossil paths faithfully represent the original movement trajectories.
Cite this review
Pith. "Pith review of Remember when? Deciphering Ediacaran-Cambrian Metazoan behaviour and temporal memory using fossil movement paths." pith.science (2026). https://pith.science/paper/4TQNL23R
@misc{pith2026250902940,
author = {Pith},
title = {Pith review of: Remember when? Deciphering Ediacaran-Cambrian Metazoan behaviour and temporal memory using fossil movement paths},
year = {2026},
howpublished = {\url{https://pith.science/paper/4TQNL23R}},
note = {Machine review of arXiv:2509.02940}
}
read the original abstract
Evaluating the timing and trajectory of sensory system innovations is crucial for understanding the increase in phylogenetic, behavioural, and ecological diversity during the Ediacaran-Cambrian transition. Elucidation of sensory adaptations has relied on either body-fossil evidence based on anatomical features or qualitative descriptions of trace-fossil morphology, leaving a gap in the record of sensory system innovations between the development of basic sensory capacities and that of more advanced sensory organs and brains. Here, we examine fossil movement trajectories of Ediacaran and Cambrian grazers for the presence of autocorrelation. Our analysis reveals a lack of temporal correlation in the studied Ediacaran trajectories and its presence in both analysed Cambrian trajectories, indicating time-tuned behaviours were in place by the early Cambrian. These results support the Cambrian Information Revolution hypothesis and indicates that increases in cognitive complexity and behavioural strategies were yet another important evolutionary innovation that occurred during the Ediacaran Cambrian transition.
Figures
Figures from the paper (9 more)
Reference graph
Works this paper leans on
-
[1]
93 REMEMBER WHEN? DECIPHERING EDIACARAN-CAMBRIAN METAZOAN BEHAVIOUR AND MEMORY USING FOSSIL MOVEMENT PATHS Brittany A. Laing.1,2, M. Gabriela Mángano1, Luis A. Buatois1,Glenn A. Brock2, Romain Gougeon1, Zoe Vestrum3, Luke C. Strotz4, & Lyndon Koens5 4.1 ABSTRACT Evaluating the timing and trajectory of sensory system innovations is crucial for understandin...
work page 2008
-
[3]
have provided insights on the sensory capabilities of some Ediacaran organisms. In turn, all extant organisms which are capable of directed self- propelled movement possess, at minimum, a basic sensory capacity (e.g. chemoreception, odor- gated rheotaxis) (Hildebrand, 1995). It is likely, then, that the makers of early trace fossils, unequivocally present...
work page 1995
-
[5]
for discretizing fossil movement paths to analyse Ediacaran and Cambrian trajectories for temporal trends. Specifically, we will be examining trajectories for the presence of temporally correlated turning angles to reveal any presence of temporally repeating patterns. To do so, we ran autocorrelation and partial autocorrelation analyses to examine if turn...
work page 2004
-
[6]
This methodology converts images of trace fossil paths to 2D curves that can then be subdivided into equidistant segments along the fossil trajectory according to an inferred velocity distribution. Figure 4.1. Sample movement path (thick grey line), subdivided into equidistant segments, with measures used in this study indicated: p is the point data which...
work page 1977
-
[7]
To investigate temporal trends in the fossil trajectories we sought to investigate the relationship between pairs of turning angles spaced increasingly far apart (i.e. “time lags”, h). To do so, we calculated the autocorrelation of the turning angles of a path, where the turning angles (θi) are compared with a lagged copy of themselves (θi+h) at increasin...
work page 2004
-
[8]
discretized data overlain. Dashed box shown in C. (C) Key variables used in the autocorrelation calculation (point data, p, with associated turning angles, θ). Lag distances used in B are illustrated by dotted gray lines. (B) Scatterplots of turning angles (θi) vs. their time-lagged counterparts (θi+h) for lag distances of 3, 5, 10, and 25 w (s = 1). r = ...
work page 1994
-
[9]
The lagged turning angles therefore are, on average, uncorrelated past this interval. This indicates that turns spaced greater than 3.4 or 2.4 trail widths apart are unlikely to be correlated for Helminthoidichnites tenuis and P. pretzeliformis, respectively. In terms of inferred action-steps, both Helminthoidichnites tenuis and P. pretzeliformis trajecto...
work page 2012
-
[538]
https://doi.org/10.1030/s41559-019-0821-6 Zhao, F., Bottjer, D. J., Hu, S., Yin, Z., & Zhu, M. (2013). Complexity and diversity of eyes in Early Cambrian ecosystems. Scientific Reports, 3(2751), 1–6. https://doi.org/10.1038/srep02751 Zollner, P. A., & Lima, S. L. (1999). Search strategies for landscape-level interpatch movements. Ecology, 80(3), 1019–1030...
Show all 12 references
-
[1909]
https://doi.org/10.1038/s41467-018-04311-8 Gougeon, R. C. (2023). The Chapel Island Formation of Newfoundland (Canada) Revisited: Integrating Ichnologic and Sedimentologic Datasets to Unravel Early Metazoan Evolution [Docotoral thesis, University of Saskatchewan]. Harvest. htt...
2023
-
[2008]
It describes the formation of a movement path as a function of four factors: (1) the organism’s intrinsic motivation to move (i.e
offers a way to examine the formation of these trajectories methodologically. It describes the formation of a movement path as a function of four factors: (1) the organism’s intrinsic motivation to move (i.e. internal state), (2) the organism’s ability to sense and respond to ...
1967
-
[2018]
and trace-fossil analysis (Carbone & Narbonne, 2014; Gehling & Droser,
2014
-
[2020]
action step
and trajectories of P. pretzeliformis from the Spitskop Member (Urusis Formation) of Namibia with bilaterian sediment bulldozers (Buatois et al., 2018). It is probable then, that these tracemakers possessed neurons and neuromodulators capable of triggering and producing time-t...
2018
Reviewed August 5, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.