# Who stole the red panda? Author: Florent Martineau Language: en Published: 2026-10-15T00:00:00.000Z Updated: 2026-10-15T00:00:00.000Z Canonical: https://florentmartineau.com/en/blog/who-stole-the-red-panda/ A painting has vanished. Three detectives with very different methods take the case. All are right, and all are wrong. ![A fictional newspaper front page: The Red Panda has disappeared. The photograph shows an empty spot in a museum and an open window.](https://florentmartineau.com/images/articles/panda-roux-journal-blank.webp) THE MUSEUM GAZETTE THE RED PANDA HAS VANISHED No alarm, an open window. Coincidence? We think not. 8 p.m. The guard enters the museum’s main gallery. Shock! The world-famous painting, “_The Red Panda_”, has vanished!! 😱 The window is open. The alarm did not go off. Three strong-minded detectives arrive. Each has a method of their own. ![Sinalors, Déjàvu and CQFD examine the painting’s empty spot in the museum, beside an open window.](https://florentmartineau.com/images/articles/panda-roux-detectives-scene-initiale-v2.webp) ## Inspector Sinalors Rules · “If… then…” > “An open window? I look for signs of entry. A missing painting? I check whether its move was authorized. I follow the procedure.” ## Commissioner Déjàvu Frames · “This looks like…” > “I’ve seen similar cases before. I look at what fits… and what doesn’t.” ## Lieutenant CQFD Logic · “What follows from that?” > “One moment. What are the facts? What are our assumptions? Before drawing conclusions, I want evidence.” ## Was the painting stolen? ### Sinalors > “The painting is missing: did someone authorize its move? The window is open: let’s look for traces. Two clues, two checks.” ### Déjàvu > “A painting disappears, a window is open… It smells like a burglary. Let’s see where they could have entered and how they could have carried the painting out.” ### CQFD > “The painting is gone, agreed. But ‘missing’ and ‘stolen’ are not the same thing.” Sinalors consults the procedure: IF the painting is missing from the gallery THEN check authorized movements IF the window is open THEN look for signs of entry For each situation it covers, Sinalors’s procedure specifies a check. But what if no case matches? No rule applies: the procedure gives her no next step. Déjàvu opens his notebook and carefully fills in his case sheet: BURGLARY? ├── missing object : _The Red Panda_ ├── possible access : _window_ ├── perpetrator : _unknown_ ├── signs of forced entry: _to look for_ └── removal of the loot : _to explain_ Déjàvu uses a _frame_: a case sheet describing a type of incident and its usual features. For a burglary, he looks for where someone might have entered, what traces they might have left and how the painting might have been taken out. These are hypotheses to test. They give him a lead, but he may persist in the wrong direction. CQFD sticks to the facts: Absent(TheRedPanda, Gallery3, 20:00) Open(Window, 20:00) Stolen(TheRedPanda): not established CQFD describes the facts and states what can be concluded from them. To him, an open window does not prove theft. Logic helps check a conclusion. On its own, it does not tell him which questions to ask next. Where would you start? **How we represent the facts changes how we investigate.** Sinalors sees conditions that trigger checks. Déjàvu recognizes a scenario and looks for the clues he expects to find. CQFD records facts and checks what follows from them. So they do not ask the same questions first. Déjàvu looks for how the thief got in. Sinalors checks authorized movements. CQFD asks what warrants saying that a theft occurred. ![Déjàvu sketches the empty spot and open window in his notebook. His sketch keeps these features of the museum but leaves out the sculptures, frames and lighting.](https://florentmartineau.com/images/articles/panda-roux-scene-et-carnet.webp) The notebook captures some aspects of the museum but leaves out plenty of details. A substitute for the real world A case sheet is not the crime scene: it represents it. It stands in for the real world. **It keeps some details and leaves others out.** ## The silent alarm The head of security explains: > When the alarm is working and remains armed, opening this window triggers an alert. Yet no alert was recorded. ![A silent alarm near the open window. Two thought bubbles show possible explanations: disarming and a fault.](https://florentmartineau.com/images/articles/panda-roux-alarme-silencieuse-v2.webp) The same silence, several possible explanations. ### Sinalors > “No alert? Let’s check that the alarm was armed, that it was working and that alerts were being recorded properly. In that order.” ### Déjàvu > “No alarm… It reminds me of cases involving someone who knows the place well. I’d look at the regulars.” ### CQFD > “Or a fault. Or an alarm that wasn’t armed. We can’t decide yet.” Déjàvu suggests a lead: the regulars. Sinalors knows which checks to carry out. CQFD points out that we still do not know whether someone deliberately disabled the alarm. A way of reasoning Depending on the desired result, each approach may have an advantage: * Knowing which checks to perform: Sinalors has the advantage. Her rules give her a procedure to follow. * Finding a lead, even at the risk of being wrong: Déjàvu has the advantage. He moves quickly, but may rush headlong down a false trail. * Avoiding hasty conclusions: CQFD has the advantage. He requires evidence before accusing anyone, even if that lets some culprits slip away. The same information can therefore help us propose a lead, choose a check or examine a conclusion. **How we represent the world influences the questions we will answer.** ## The badge seems to implicate the technician An access log provides a new clue: 18:42 | GALLERY 3 | TECHNICIAN’S BADGE Yet the technician says she left at 6:30 p.m. ### Sinalors > “The times don’t match? Let’s check the clock.” ### Déjàvu > “An alibi that doesn’t hold up… I’ve seen that before.” ### CQFD > “Someone used her badge. Nothing proves it was her.” A few minutes later, the conservator explains that he had borrowed the badge. ![The conservator holds a badge bearing the technician’s portrait up to a reader: the person using the card is not its owner.](https://florentmartineau.com/images/articles/panda-roux-badge-emprunte-v3.webp) The log was accurate. It was Sinalors and Déjàvu who were reading more into it. If the log had identified the person using the badge, it would have said: 18:42 | GALLERY 3 | USER: CONSERVATOR The technician would then not have been accused in the conservator’s place. Choosing what to represent What should we record: the badge’s owner or the person using it? That choice changes the possible conclusions. Here, we need to distinguish two relationships: “owns the badge” and “uses the badge”. **If our description confuses them, we risk accusing the wrong person, whatever our method of reasoning.** None of the three detectives is immune to this trap. A rule can rest on a false assumption. A frame can send us looking for clues to a scenario that never happened. And purely logical reasoning can start from incorrect information. ## The right information, in time A miracle! Sinalors, Déjàvu and CQFD agree! All three want to check the artwork movement register to see whether the painting was moved. A well-organized Excel file? Filter for “The Red Panda” and there it is! A pile of unsorted documents? Good luck. ![On the left, the three detectives find a row in a filtered spreadsheet. On the right, they struggle to find information among unsorted files and papers.](https://florentmartineau.com/images/articles/panda-roux-tableur-paperasse-v2.webp) The same search, with or without organization. A way to find answers How we represent and organize information can make a search quick or laborious. **Being able to find an answer is not enough: we must be able to find it within a reasonable time.** ## The painting was in the workshop The register records an authorized transfer at 6:45 p.m. The conservator confirms the move. The painting is found in the workshop. The window had been opened to air the room after closing, with the alarm disabled under the usual procedure. ![Sinalors, Déjàvu and CQFD surround The Red Panda, found on an easel in the conservation workshop.](https://florentmartineau.com/images/articles/panda-roux-atelier-detectives-carre-v3.webp) The painting had simply been transferred to the workshop. ### Sinalors > “Authorization checked, painting found. Both boxes ticked.” ### Déjàvu > “Right. This looks much more like a trip to the conservation workshop than a burglary.” ### CQFD > “The painting had simply been transferred to the workshop. QED.” We had started looking for a thief before checking where the painting was. ## How do we avoid making the same mistake again? The museum wants to remember the lesson: > A badge can be used by someone other than its owner. How should we write that down? ### Sinalors Revise the rules: use of a badge is not enough to identify the person who entered. ### Déjàvu Allow for a borrowed badge in his case sheet. If the investigation depends on the person’s identity, remember to check. ### CQFD Record separately who owns the badge and who used it: `Owner(Badge17, Technician)` `User(Badge17, Conservator, 18:42)` All three can record that a badge was borrowed. Their colleagues still need to understand that information and correct it if necessary. CQFD shows the guard a logical notation she does not understand. Sinalors restates the same facts in sentences the guard understands. ? Owner(B17, T) User(B17, C) Got it! The technician owns the badge. The conservator used it. The same information, expressed for the person who needs to read it. A shared language We need to write down what we know, make it understandable to others and correct it. **A representation must be understandable to its users, not just its author.** ## There is no magic representation The three detectives have the same clues. Their descriptions help them do different things. ### Rules **For following a procedure:** they connect a condition to a conclusion or an action. **The limit:** overlooked cases and bad rules. A procedure cannot anticipate everything. ### Frames **For recognizing a situation:** they bring together what we know and what we expect in a typical case. **The limit:** the real case may be different. What usually happens is not necessarily what happened here. ### Logic **For checking a conclusion:** it states what follows from the facts and relationships we have described. **The limit:** on its own, it does not say which action to take or which lead to explore. Correct reasoning from false premises can lead to a false conclusion. These approaches overlap: rules can be written in logic, and scenarios can be described through facts and relationships. **The same information can be easy to use in one form and cumbersome in another.** **Before choosing a representation, we need to know which task we want to accomplish.** Do we want every detective to follow the same rigorous procedure to limit bias? Do we want to find a lead quickly, at the risk of following the wrong one? Or to check that a conclusion really follows from the facts? In [_What Is a Knowledge Representation?_](https://doi.org/10.1609/aimag.v14i1.1029), Davis, Shrobe and Szolovits distinguish five roles explored during the investigation: * **A substitute for reality:** the notebook lets us study the scene without staying at the museum. * **Choosing what to represent:** the badge’s owner and its user are two different people. * **Guiding reasoning:** depending on the method, checking a conclusion, suggesting a lead or triggering a check. * **Getting an answer quickly enough:** an organized register saves us from searching through thousands of documents. * **Sharing what we know:** the guard must be able to understand and correct the report. Representations can be combined: a frame to suggest a lead, rules to organize checks, logic to examine conclusions. **We can choose several representations to perform different tasks.** **There is no ideal representation in the abstract: its value depends on what we want to do.** > You know, I don’t believe there is an inherently good or bad representation. I’d say it starts with goals. Procedures to follow, leads to explore, conclusions to check… So I give thanks for case sheets, I sing the rules, I dance with logic! And when people ask, “How do you find the culprit?”, I say: first I check that there is one. Did your favorite detective change during the investigation?