So a new paravian was described recently- Yuezhuornis yangyuanensis (Xu et al., 2026; note this is Xu Liming, not Xu Xing from the Gigantoraptor post, so I'm not just critiquing the same person again). From the same formation as Baminornis, Zhengheornis and the supposed euornithine furcula IVPP V33260, the shtick of these authors has been that everything in this Jurassic locality is remarkably advanced. Baminornis had a pygostyle (actually just the synsacrum) and narrow coracoid neck (just traced incorrectly), Zhengheornis was a tiny adult with a short unfused tail (no actual evidence of synsacral or tibiotarsal fusion, and thus age) and a birdlike ischium with no proximodorsal or obturator processes (it near certainly had one or the other), and the furcula is somehow just like Yanornis' despite being 25 million years earlier (the irregular outline makes me doubt this is even a furcula), etc...
And now we have Yuezhuornis, another supposed avialan closer to Aves than Archaeopteryx. Without even reading the full paper and glancing at the figures my thought was immediately that this looked a lot more microraptorian than even Archaeopteryx (which my matrix consistently but weakly places as a deinonychosaur). Long serrated teeth, short coracoid, prominent proximoventral heel on pedal digit II, etc.. BUT I haven't scored it yet in my matrix, so I'm not drawing any conclusions yet. What I AM drawing is a cranial reconstruction of Yuezhuornis to help me score it, but that comes later.
 |
| Extended Data Figure 1 from Xu et al. 2026- "Skeletal reconstruction of Yuezhuornis yangyuanensis. Elements not preserved in IVPP V34427 are coloured in grey. Scale bar, 50 mm." |
But first, Xu et al.'s skeletal reconstruction as shown in their Extended Data Figure 1 is just terrible.
- Quadrate not articulated with the squamosal or quadratojugal.
- Mandible medial to quadrate.
- The skull and neck give a new definition to 'shrink-wrapped', with basically no flesh outside of the bones.
- Pectoral girdle anterior edge at the anterior edge of the eighth cervical vertebra.
- Proximal caudals twice the size of the last dorsal.
- Proximal caudals articulate too far ventrally at the ischial level instead of at the ilial level where the sacrum would be.
- Caudal neural spines and tip of the tail extending outside the body outline.
- No carpus, with the antebrachium articulating with the posterior/lateral edge of metacarpal III.
- Seven huge sternal ribs articulating with nothing.
- Medial gastralia extending below what would be the body outline.
- Posterior thigh not even twice as deep as the femur.
- Right leg too large and misshapen with e.g. giant distal tibiotarsal musculature.
- Fibula thick enough to cover all but the cnemial process of the tibia in lateral view.
- Metatarsus in posterior view proximally, then digit II is lateral to metatarsal III and digits III and IV.
- At least pedal phalanges III-1, III-2 and IV-3 are in dorsal/ventral view, and III-1/III-2 and IV-2/IV-3 overlap their ends.
These aren't issues with Yuezhuornis' anatomy itself, but basic anatomy of any theropod or even any skeleton. But e.g. pedal phalanx II-2 is a classic deinonychosaur shape, so someone who knew paravian anatomy was involved somewhere along the process, so it's baffling. I reported most of this on the Dinosaur Mailing Group and Brad McFeeters asked "How probable is it that the skeletal reconstruction was AI-generated?" I tried and couldn't get ChatGPT to generate an even semi-realistic paravian skeleton (see my DMG post for those results). It's sorta like the infamous 'too many fingers' AI issue with illustrating humans. ChatGPT can give the feel of a paravian GSP-style skeletal reconstruction, but anyone remotely familiar with anatomy immediately sees things don't make sense. Totally different from the Yuezhuornis recon, where it has the right elements but they are placed incorrectly. As I noted there, in the ChatGPT attempts at Yuezhuornis "the proximal caudals are positioned correctly in the dorsoventral axis, there is a carpus between the metacarpals and antebrachium, the second toe stays medial, none of the pedal phalanges overlap at their joints, and the body outline is pretty good!" So ChatGPT knows basic anatomy better than the Yuezhuornis illustrator in some ways at least.
 | ||
| The result when I put Xu et al.'s reconstruction into ChatGPT and asked it "Can you keep all the bones the same in this skeletal reconstruction, but change the black flesh outline to be in the probably correct area?" So it's a modification of Xu et al. (2026). |
Amusingly, when I put Xu et al.'s reconstruction into ChatGPT and asked it "Can you keep all the bones the same in this skeletal reconstruction, but change the black flesh outline to be in the probably correct area?", it not only gave a mostly correct flesh outline, but even gave it a carpus, smaller sternal ribs, no overlapping pedal phalanges, better ribcage and gastralia, correctly sized right leg, and moved the pectoral girdle a little bit back, despite me explicitly telling it to keep the bones the same. As I said there, "Some parts were too bad for even ChatGPT to ignore, I guess."
 |
The best result I got after trying to correct the anatomical errors of Xu et al.'s (2026) reconstruction via ChatGPT. So this is ultimately also a modification of Xu et al. (2026), but DO NOT TRUST IT! (see below) |
I went on to use ChatGPT to try to fix things and ended up with a very aesthetically pleasing reconstruction. But as I say there "It's pretty easy to, say, give it a realistic feathery covering, exoparia, or make the right leg a silhouette, but I could not get back to the actual manus, keep the pes the same while moving the sickle claw to the background, move the pectoral girdle and forelimb back while keeping them with the same anatomy, or give it the correct fibula. It's not software where you can just say 'move this structure right X pixels' and it doesn't understand anatomy enough to e.g. make the fibula very narrow and not reach the tarsus. It also can just randomly change things, saying 'the image generator may have changed some skeletal details' even when you explicitly tell it not to, so is unreliable. For all I know the proportions could have been changed a bit by that. But IMO it beats Xu et al.'s reconstruction by a mile..."
So I would strongly advise against using ChatGPT for any part of scientific figures. That's not what it's good at. In any case, I'll need a cranial reconstruction if I'm going to score Yuezhuornis, and call me crazy for not trusting the skeletal. So I take Xu et al.'s own cranial element outlines from Figure 2c, and besides the jugal and quadratojugal, it's just WRONG. I'm no artist, but I CAN trace. Take a look-
 |
| Top- Cranial elements in situ from Figure 2c of Xu et al. (2026). Middle- Cranial reconstruction from Extended Data Figure 1 of Xu et al. (2026). Bottom- my tracing of the elements in Xu et al.'s (2026) Figure 2c to be realistically articulated. Scaled to same jugal height. |
The point is- Xu et al. didn't even try to follow their own interpretations. Jugal- great. Quadratojugal- a little too small. But maxilla? That's completely different. The reconstruction pretends it's Archaeopteryx, but the "real thing" has a much longer posterior process, more rounded antorbital fenestra edge, convex anterodorsal edge, supposed promaxillary fenestra (probably the maxillary fenestra, based on dromaeosaurids) elongate and dorsally placed, and ten teeth preserved with space for at least fourteen. The reconstruction gives it seven teeth.
The postorbital is the one element I did reinterpret, because Xu et al.'s orientation would give the orbit a weird acute posterodorsal corner not seen in any maniraptoromorph. But the reconstruction doesn't bother with that at all, because they make it a third of the size and a completely randomly different triradiate shape. The lacrimal isn't so bad because who knows what they're hiding under other bones, but the antorbital fenestra edge is definitely too acute. The premaxillary teeth should be more slender and procumbant. The dentary should be a bit upcurved, while Xu et al. just seem to have copy/pasted the mandible of Archaeopteryx. There are trickle down effects too, like with that convex anterodorsal maxillary margin, there's no way the external naris is going to be so huge without a big oviraptorid-type premaxilla subnarial process or a weirdly ventrally concave naris even assuming the gray snout line isn't correct.Importantly though- it's not just a poor reconstruction, it's misleadingly wrong in almost entirely the same direction. The changes make Yuezhuornis seem birdier. Shorter maxilla with less teeth, larger external naris, shorter antorbital fenestra, Archaeopteryx-like maxillary/promaxillary fenestra, smaller postorbital, smaller quadratojugal, non-upcurved dentary, and the things they add (without indicating they're made up) like the short nasal and posterior-facing parietal make it look even more so. The artist knew what the jugal looked like, so it wasn't just a case of being told to create some Archaeopteryx-like skull with no data....
Also note how misleading the amount of white "known" anatomy is. Sure, it's ONLY the squamosal, quadrate and sclerotic ossicles we don't have.... :| And er.. lateral nasal?
In any case, since THIS is so wrong, I was curious if at least the proportions of the skeletal reconstruction were accurate despite being bad in so many other ways. So using the femur as our element to scale to-
element real length recon length scaled to femur (80.74 mm) recon/real
Humerus 69.56 66.7 0.96
Radius 32.01* 57.7 1.80
Ulna 64.88 57.9 0.89
McII 24.67 19.3 0.78
McIII 23.20 25.0 1.08
Ilium 47.65 46.4 0.97
Pubis 69.91e 42.1 0.60
Tibia 105.55** 107.8 1.02
MtII 45.09 45.2 1.00
MtIV 49.47 53.6 1.08
* Obviously something went wrong here since the radius is subequal to the ulna. Which makes me question the written portion even more...
** I'm treating this as tibiotarsal length, as neither in situ figure or reconstruction shows separate tarsals, despite the description saying "The calcaneus and astragalus are unfused with one another or the tibia."
So that's pretty abysmal even ignoring the obviously wrong listed radius length. The ulna, metacarpal II and pubis are increasingly way too short with the latter being 60% of what it should be! So 5 of 10 elements are within 10%?
In conclusion, just call Xu et al.'s reconstruction a fictional animal with the right jugal and mtII/femur ratio. And on that note, I apologize for introducing my AI modified skeletal into the web, since apparently it was ultimately based on an imaginary creature despite looking rather good. So Petersian of me...
But lucky for us, Yuezhuornis was only published in an obscure, non peer-reviewed rag called... NATURE?! Only (along with Science) supposedly the most prestigious journal of all whose peer reviewers and editor would have SURELY noticed such nonsense? Yet we are blessed as the peer review data was actually made public! So reading it for the first time, let's see what they say...
Novas says "I have read with attention the manuscript about the newly discovered Yuezhuornis, and I must say that the anatomical evidence afforded for the specimen is not convincing enough for its location as a basal avialan more derived than Archaeopteryx" Ha! When your intuition agrees with Novas, you might be on the right track...
Anonymous Referee #2 (come on...) says "Overall, I have sufficient doubts about the most important inferences in the paper, that I do not recommend publication of this fossil in Nature. ... I have numerous reservations about this, which include: (1) I'm not confident that this is a mamber of Avialae, rather than a non-availan paravian (e.g. similar to Anchiornis)." Ha again!
Anonymous Referee #3 (ditto) says "What I am not convinced of is that it is a flightless bird. Let me unpack what that means. If it were a flightless bird, that means that: 1) it is a member of the clade Avialae (the ‘bird’ clade capable of powered flapping flight: Archaeopteryx + modern birds) and more derived than Archaeopteryx; 2) it evolved from flying ancestors. I think neither of these things is clear." Ha thrice! "Remarkably, the new species has a hyperextensible second pedal digit! That is amazing! Or
it would be amazing if it truly was an avialan, more derived than Archaeopteryx. Because this is a classic deinonychosaur feature. And I think, ultimately, a giant red flag: if this species has a classic deinonychosaur feature, then it most likely is a deinonychosaur." You don't say?
First mention of the figures! O'Connor says "The figures in the MS (both main figures and extended data figures) are generally good and convey the main points of the paper." Hmmm...
In round 3, Novas says "3. Comments on FIGURE 1 (EXTENDED DATA): Please, review drawing because the knee seems dislocated (femoral distal condyles are anteriorly displaced respect to proximal end of tibia)." How bad can that figure have been then if the knee was the biggest issue?!
Novas apparently also said (maybe in the non-public pdf review file?) "About Extended Data Fig. 1. I have counted 29 presacral vertebrae, which contrast with most early paravians in which the count does not surpass 23 presacral elements. Ungual of digit II seems large. Is that big?" The authors replied "We thank the Reviewer for these comments. Yes, we have revised this reconstruction accordingly. The ungual of digit II was incorrectly reconstructed, and this has been fixed now." So it was once even worse! And yes, it has 23 vertebrae now (assuming a tiny hidden atlas).
Not directly relevant, but "23. "The calcaneus and astragalus are unfused with one another or the tibia. In contrast to what is observed in most other Jurassic avialans and non-avialan paravians18"
[the referee asks] >>Could be more specific about which taxa. They are unfused in
Anchiornis and Archaeopteryx.
Response: No, we intended to mean that the fibula reaches the proximal tarsals in Anchiornis and Archaeopteryx here, not about the fusion degree of calcaneus and astragalus."
So.... ARE the astragalus and calcaneum unfused?! It's... just... ... I'm sure you readers get it.
So in (final-er) conclusion, no- the reviewers didn't notice the reconstruction did not match the in situ figures for anything except the jugal. You would hope you didn't need to care about that, but maybe we do now?! At least Novas saved us from 29 presacral vertebrae!
And in final(ist-er) conclusion, I haven't even read all of the Yuazhuornis paper yet, and (surprise!) do not trust their tracings of any element. So maybe their tracings are wrong, thus my reconstruction is wrong. the whole thing is so undependable. Who knows! BUT my point is partly that their tracings and reconstruction can't both be right.
Not so long ago, I thought the most flawed part of our field was that the children phylogenetic analysis scorings were wrong. But if the reconstructions can be mostly fictional and make their way past Nature, then OOF! I'll say one other thing- when reading the next report of the birdiest bird-like thing from the Nanyuan Formation, DON'T TRUST IT!
Reference- Xu, Wang, Lin, Dong, Deng, Xu, Du, Lin, Chen and Zhou, 2026. Jurassic flightless bird reveals early ecological diversity of avialans. Nature. 658(8136), 690-697.
It has seemed to me for a long time—since at least the discoveries of _Sinosauropteryx_ and _Confuciusornis_ and the continual, early assertion that they are Late Jurassic in age—that somewhere within Chinese paleontology there is a strong current of pressure to ensure that China has both the oldest bird(s) as well as the best and most continuous evolutionary record of early avian evolution. For that reason, I've viewed the latest round of Late Jurassic ostensibly bird-y fossils with some trepidation: are they really more derived contemporaries of _Archaeopteryx_, or is there a nationalistic tendency at play to force the issue that leads to misleading (at best) or erroneous (at worst) interpretations of the evidence? If such pressure exists, I do not know from where it ultimately comes...the scientists themselves? Political pressure? Add to this the long-standing rumors that Chinese paleontologists get big monetary bonuses to (a) erect new taxa, and (b) publish sexy and/or paradigm-shifting interpretations in "top-tier" journals, and now there's the appearance of even _stronger_ pressure to do the science less rigorously than it deserves. I'm not at all saying that Chinese paleontologists are poor scientists; they have produced excellent work! I merely worry that their hands are being forced, consciously or unconsciously, to view new fossils through certain lenses that color their interpretations.
ReplyDeleteWhile the critique of the artistic reconstruction in Extended Data Fig. 1 is largely valid—2D skeletals often contain anatomical errors, and this one certainly appears to have misplaced elements and exaggerated proportions—the blog post makes a fundamental error by conflating an ancillary illustration with the paper's primary scientific data. To argue that these drawing errors invalidate the study's conclusions is to attack a straw man.
ReplyDeleteThe taxonomic, phylogenetic, and ecological conclusions of Xu et al. (2026) are not derived from the skeletal reconstruction. They are based on direct observations of the holotype (IVPP V34427), a comprehensive character matrix (853 characters across 170 taxa), and quantitative morphometric analyses. Here is why the core arguments of the paper remain robust:
1. The Phylogenetic Analysis Specifically Tested the "Deinonychosaur" Hypothesis
The blog post suggests that Yuezhuornis might be a deinonychosaur rather than an avialan. The authors anticipated this exact objection. They explicitly constrained their analysis to force Yuezhuornis into a monophyletic Deinonychosauria. The result? The constrained tree was 14 steps longer than the unconstrained most parsimonious tree, and the Bremer support value for uniting Yuezhuornis with deinonychosaurs was -13. This is strong quantitative evidence that the deinonychosaur hypothesis is significantly less parsimonious than its placement as a basal avialan.
2. The "Mosaic" is Fossil Reality, Not Artistic License
The paper describes Yuezhuornis as having a mosaic of features, and this is based on the actual fossil bones, not the drawing. The authors document:
-Avialan-like postcrania (trapezoidal coracoid, obtuse interclavicular angle, short fibula, reduced fifth metatarsal).
-Troodontid-like cranial traits (closely packed premaxillary teeth, serrated maxillary teeth, posterodorsally directed postorbital process of the jugal).
-Deinonychosaur-like pedal features (a hyperextensible second pedal digit).
This combination is genuinely present in the matrix and the specimen, illustrating the complex, mosaic nature of early avialan evolution.
3. The Flightless/Terrestrial Ecology is Supported by Quantitative Data
The conclusion that Yuezhuornis was flightless and terrestrial is not based on a visual assessment of a drawing. It is supported by three independent lines of evidence:
-Raw Measurements: The forelimb is shorter than the hindlimb, and the humerus is shorter than the femur.
-Ecomorphometric PCA: Principal Component Analysis of limb proportions places Yuezhuornis squarely within the morphospace of extant terrestrial birds, distinct from arboreal taxa.
-Phylogenetic Comparative Methods: Mapping limb ratios onto a dated theropod phylogeny confirms this taxon has a prominently short forelimb, even shorter than some dromaeosaurs.
4. The Peer Review Process Worked
The blog implies that Nature's peer review failed. However, the published peer review file shows that reviewers explicitly challenged the avialan status and the flightless interpretation. The authors defended their conclusions using the quantitative data outlined above, and the reviewers ultimately recommended publication.
In summary, yes, the skeletal reconstruction is flawed and should probably be revised or removed in future versions. But scientific papers are judged on their data and analyses. The morphometric data, the character matrix, and the explicit testing of alternative phylogenetic hypotheses robustly support Yuezhuornis as a flightless, terrestrial avialan. The errors in an artistic reconstruction do not change the fact that the fossil itself is a striking example of early ecological diversification in birds.