Field Atlas / Animal Behaviour
A dog gets it right.
What did it learn?
Dark yellow led to food. Against light blue, “pick yellow” and “pick dark” both worked. Choose Separate the clues to test the two rules.
A question without words
Original illustration
Inspired by a real study
Both rules point to dark yellow. Give them different answers to tell them apart.
Reported for these test cards
Labels carry colour and brightness. Your screen is not calibrated paper or a dog-vision simulation.
The training puzzle
Two clues.
The same answer.
These are predictions from two stated rules. The illustration does not generate animal choices.
The predictions agree. The explanations don’t.
A choice of dark yellow fits both rules. Change the pairing so yellow and dark no longer identify the same card.
Go further: design a trial plan that can tell competing rules apart ↗. Keep hypothetical choices separate from the reported study.
The test design
Put the clues on
different cards.
To find out whether an animal learned the signal you meant to teach, give the possible rules different answers. Light yellow keeps the trained colour. Dark blue is the darker option in the new pair. Researchers call this a cue-conflict test.
Exchanging left and right checks another possibility. The colour and brightness predictions move with their cards. A habit of choosing the left-hand box would still point left.
The 2013 study
What did eight dogs choose?
Which clue would dogs follow if colour and brightness both led to food? In training, each dog chose between two boxes. Only one opened for food. Its card differed from the other in both ways.
One test card kept the rewarded colour. The other was light or dark like the rewarded card. Both test boxes stayed locked so researchers could count choices without rewarding either new card.
All eight chose the trained colour more often in these tests. Compare La and Co to see how consistent their choices were. Each mark records one choice by the selected dog.
La
Could a dog that always goes left look like it’s following yellow? Design a card test ↗
Source: Table 1 in the original paper ↗. Individual counts are transcribed, not a new statistical analysis. Repeated choices are not extra independent animals.
Inspect all eight records and the experimental controls
| Dog | Positive training card | Test cards C / B | C choices | B choices | Total |
|---|
Both feedboxes contained meat. One opened in training; both were locked in tests. Positions varied. The paper used three training assignments, whereas our illustration uses one. Ji had twenty test choices; each other dog had ten.
Read the less tidy record: Co chose the trained colour seven times out of ten. That row does not establish a uniform response in every dog. More choices from one individual don’t increase the number of independent animals.
Outdoor daylight, particular paper stimuli and eight previously untrained dogs bound this result. Screen colours are not the study’s spectra. See the source for procedures and reported statistics.
Keep the question
A notebook, not a black box.
Keep the setup and selected dog record in a link or file. Export the illustration’s rule assignments separately from the study’s reported choices.
Methods, scientific source and credits
The interactive schematic calculates the predictions of “choose yellow” and “choose dark.” It does not simulate learning, animal vision or movement, or generate trial outcomes. Moving the cards changes their screen side without changing either rule. The dog drawing is original artwork.
The optional study view is an attributed factual transcription of eight Table 1 test-count rows. It retains the unequal counts and original assignments. No significance calculation or general training recommendation is added. Publication-status history has not been comprehensively assessed. The original paper’s text and figures retain their own rights; none are bundled.
- Kasparson, Badridze and Maximov (2013), original methods and tables ↗
- PubMed bibliographic record ↗
- DOI: 10.1098/rspb.2013.1356 · source checked 3 November 2026.
Aptifold’s AI agent researched the sources, implemented the models, designed the visuals, wrote the code and ran the tests. Aptifold’s founder set the goals and reviewed the results. Software: MIT License. Scientific authors retain their credit. This independent educational view implies no institutional endorsement.