Evidence grades
The three-grade notation used on every substantive claim in this course, how to assign one, and the claims we currently grade as contested.
Every substantive claim on this site carries one of three grades. You meet the notation in the week 1 seminar, you start assigning grades yourself in week 3, and by the cognitive profile every claim you make has to carry one. This page is the definition the rest of the site points at.
The three grades#
[E] — established. Replicated, or uncontested primary evidence. The claim has survived contact with people who would have liked to falsify it. A single well-designed study is not automatically [E]; a single well-designed study that nobody has successfully contested in twenty years usually is.
[C] — contested. There is live disagreement in the field, and the disagreement is the content rather than an inconvenience. A [C] grade does not mean “probably false” and it does not mean “we don’t know yet”. It means that competent people have looked at the same evidence and reached different conclusions, and that you are expected to be able to name both positions. Roughly a third of this course is [C], by design.
[S] — speculative. A reasoned extension beyond the evidence, labelled as such. Speculation is permitted here and it is not a lesser register — the course ends in it deliberately. What is not permitted is speculation that arrives wearing the clothes of evidence.
Why the notation exists#
The subject matter makes it necessary. Animal cognition and consciousness research is unusually full of results that are overclaimed, contested, or have failed to replicate, and the failure mode for a course with this title is reciting a satisfying canon that has not held up. Week 11 is largely about how much of the famous literature on human error did not survive replication.
Hence the rule the whole site follows: the three registers never blur inside a paragraph. If you cannot tell which one you are reading, that is a fault in the writing, and on this site it is a fault you should report.
How to assign a grade#
Grade the claim, not the paper, and grade it at the resolution the evidence supports. Most disputes about grades turn out to be disputes about what claim is actually being made.
The worked example is in week 3. Gagliano and colleagues reported associative learning in pea plants in 2016; Markel failed to replicate it in 2020 with blinded scoring; both parties published further and neither conceded. That is [C]. In the same year Dener, Kacelnik and Shemesh reported risk-sensitive foraging in pea roots, and it has held up. That is [E]. Two papers, same year, same plant, same broad claim family, different grades — and being able to say why is most of what this notation is for.
Conversely, week 9 carries a claim graded [S] leaning [C]: the Stubbs and Stubbs proposal that chromatic aberration plus a peculiar pupil shape could let a colourblind cephalopod recover spectral information. It is ingenious, it is not established, and nobody has settled it. Compound grades are allowed when they are honest. What is not allowed is picking the flattering half.
What we currently grade [C]#
These are the standing contested claims in this course. The list is maintained deliberately rather than pruned, because a course that quietly resolved its own disagreements would be doing the thing it warns you about.
| Claim | Grade | Where |
|---|---|---|
| Associative learning in pea plants | [C] | Week 3 |
| The free energy principle as a unifying framework | [C] | Week 6 |
| Introspection as simply unreliable (the strong reading of choice blindness) | [C] | Week 7 |
| Mirror self-recognition in cleaner wrasse | [C] | Week 7 |
| Play and REM-like sleep in insects | [C] | Week 8 |
| Single-channel colour vision in cephalopods | [S] leaning [C] | Week 9 |
| Whether social influence improves a group estimate | [C] | Week 10 |
| The dual-process account of reasoning | [C] | Week 11 |
| The classic Dunning-Kruger curve | [C] | Week 12 |
| Whether integrated information theory is testable | [C] | Week 12 |
Note that two of these are graded on a reading of a result rather than on the result itself. The choice blindness data are not in dispute; what is in dispute is whether they license the strong conclusion, and Petitmengin and colleagues’ 2013 work is why. Grading the reading rather than the finding is often the precise thing to do, and it is worth more marks than grading the finding bluntly.
You are accountable for the grades you assign#
This is in the policies and it is worth repeating here. A grade you assign is a claim you have made. If you mark something established and the literature turns out to be split, that is an error in your work in the same way a miscalculation would be — not a difference of opinion, and not excused by how you arrived at it.
The same standard applies to us. If you think a grade on this site is wrong, tell Halvard. Corrections get made and attributed, and finding one is a better demonstration of what this course teaches than anything on the assessment page.