 This paper is about steering and platform self-preferencing on the Amazon Bivox and just to profess preface before I start the paper. When I started this paper there was almost no research on this topic and in the past year or two it's exploded. So I'm really lucky to have Leon as a discussant as well as Tiffany both of which have done a lot of work in this space. But let me start out with this is the disclaimer why we might care. So there are two opposing views on this topic. So the first few you can see in the US House of Representatives digital markets report on competition in digital markets and you know there's a 400 page report where they talk about all the the big tech companies but when they talk about Amazon and this is a more general theme throughout the report they talk about how the fact that it controls many different business lines means it can preference itself and disadvantage competitors in ways that undermine free and fair competition. Other businesses are beholden Amazon for success. The fact that Amazon has a bunch of hats so if you think of Amazon here Amazon is a retailer, Amazon is a fulfillment operator, Amazon is also a marketplace at host sellers. This is all a huge inherent conflict of interest and Amazon is exploiting that conflict of interest to disadvantage competitors in competition. So this is one widely held view both of Amazon, Biobox, self-preferencing but more broadly one of the big issues in antitrust enforcement in the tech space. So the other view is our country's most recent space cadet so this is Jeff B I guess he's now retired enjoying himself but this view is this is all about consumer welfare and so when he talks about the Biobox I really believe this is congressional testimony he says you know we're picking the offer that we think consumers are like the most there are a whole bunch of different attributes so there's price there's whether it's prime or not there's shipping speed other things but we're just doing what's best interest of the consumer. So that's another view and if you think about this more broadly there are lots of different sellers you're going to have to pick a default seller and Amazon is saying well we pick the default seller that we think consumers not the most. Now this is a talk in the Toulouse seminar so I'll talk a little bit about the EU. So the EU commission has been investigating this as an abusive dominance case so this is Margaret Ristauer who's head of DG Comp and I don't know if you can see it but behind her there's something about antitrust in Amazon so this is one of her big issues as well. So she first outs out and she says the Biobox is essential you know winning the Biobox is crucial because more than 80% of all transactions on Amazon come through the Biobox. So first of all this is really important for Amazon's business. She then says Amazon may have been artificially favoring itself both in terms of its retail arm and in terms of its logistic and delivery services and the EU commission is really concerned that they may be pushing retailers to use its own services and locking them into Amazon's ecosystem. So this is something of a lot of policy interest now not so much I think when I started this paper but policy interest is also blown up a lot and this is just one of many cases I believe that cases in the UK and maybe in Germany and others as well. I'll talk a little bit about this at the very end to conclude the paper. So this is going to be a very simple paper the idea here is to try to examine the Biobox algorithm and measure how much is Amazon favor itself both in terms of its retail business and in terms of its fulfillment business and then in some very simple counterfactuals how does that steering effect both through the Biobox winner is as well as prices and feel free to interrupt me if you have any questions I'm happy to take any questions. Alright so with that let's get started. So Amazon I believe in 2000 created a marketplace so when it originally started as a as a website it was just its own retailer it wasn't really a platform at this point in time there are a bunch of different players that are all playing on the Amazon website. First there's Amazon retail where Amazon is you know is the retailer it's using its fulfillment services and shipping to you directly but second there are a number of different types of third-party sellers first there's I'm going to go in a different order than the slide but first there's fulfilled by merchant so here the merchant is doing the shipping itself but it's selling on Amazon's website and then Amazon is taking a cut typically about 15% that depends on the category. Second there's fulfilled by Amazon which in this case Amazon is doing the shipping so these products are sitting in an Amazon warehouse Amazon is doing the shipping but the merchant owns the goods and so it's being sold by the merchant. So here Amazon is taking the same cut as for the other merchants but it's also taking some FBA fees and Tiffany has has done a lot of work on this she can tell you more if you want to know any details about that. Now lastly there was an option to do seller fulfilled prime where you could get the prime badge but the merchant would do the shipping at the time I got this data Amazon had decapped this program and essentially eliminated it by making it very difficult to get but that's something I may talk about again at the end of the time. Now let me show you what the buy box looks like so this is some kind of Japanese video game. If you look in the red square I showed this once in a Japanese student was like you shouldn't put this video game on it's like a bad video game or something. I have not played it so this game is if you look in the red box you can see it's being offered on Amazon for $25.10 with the prime badge and if you kind of squint at the very bottom you can see it says ship from and so by amazon.com but that's something that is not really is not really what consumers can see. Now the green box here you may need to move your zoom feed if it covers it. The green box here is the buy box and in the buy box you can see there's a free delivery it's going to be shipped on a Tuesday which is probably in two days I think I did this on a Sunday and then if you look in blue you can see a box saying there's actually 22 different sellers. So the green is the buy box and if you click the buy now button you'd buy it. If you clicked in blue which people often don't do you can see that there are actually a whole bunch of different offers here. So in red is the featured offer which is $25.10 that's coming from Amazon but you know there is a blue offer for I guess $0.13 cheaper from a third-party FBM seller. So this is going to come a little bit later they say free delivery and maybe a week or two. There's also a FBA offer from a third-party seller that is $0.11 cheaper for $24.99 and it would arrive on the same day as the featured offer. So you can see here that you know this is a case where Amazon has won the buy box it's not the cheapest offer there's both an FBM and FBA seller that are cheapest. Now there's not always a buy box winner and I know many of you may have gone to the buy box or the BYE talk earlier in the year about this. So sometimes so this is a product this is some kind of slime tear sealant there is no you can't click in green on a one click button because there's no option there but if you look in red it says it's available from these sellers and if you click on that there is a seller that's selling this for $29.99 with a prime offer that would come in two days and there's another seller that looks like an FBM seller as well. So one option is that Amazon doesn't open the buy box at all. I'm going to put this in my algorithm. There's a lot of work which I think showed up a little bit in the BYE buy box paper about Amazon restricting the buy box if somebody's selling for cheaper on a different website. So for example on eBay or Walmart if somebody's selling cheaper they might close the buy box that's not a feature of my paper but I may talk a little bit later about that's something I find hard to predict which may be because I don't have data on what's going on on other websites. But there are multiple court cases now on this topic. So that's not very clear what it is that how it is that Amazon chooses the featured merchant or who wins the buy box. If you go to their website they say a few lower prices are better and things like that in stock is better but they don't really tell you what the algorithm is and there are third parties that have tried to give some guidance. So these are typically third parties that are selling their services to sellers. So this is from I think feed advisor and the top metric with the highest impact on the buy box is fulfillment method and they say the best way to win the buy box is FBA. So what this says is if you use Amazon's fulfillment services you're much, much more likely to win the buy box than if you don't. So that's the top metric. There are others so there's price shipping time and whether it's in stock or not are the ones they call high and then there are a bunch of other things some of which are about shipping quality and some of which are about the number of ratings and feedbacks of the the rating quality and the feedbacks of the third party. So again this is a black box but it seems like the most important method is whether you use Amazon's fulfillment to ship or not. Now what I'm going to do in this paper is get data from across both different categories and different countries and then try to reverse engineer what the algorithm is given my data. So for my sample I'm going to collect offers by products at the country category level. So I'm going to stop with the top 500,000 ASINs for a country category by sales rank. So think of this as US electronics. I'm then going to collect a 10,000 product sample. The head of the sample is all of the top 5000 ASINs. So ASIN for those of you who are not as familiar with Amazon is their version of EBC code. So the head is going to be all 5000 top ASINs by sales rank. I'm then going to create samples of products that are that are in the tail and I call this the torso. The idea here is to try to get something of the entire distribution of the top 500,000. Now the top 5000 ASINs are going to be sold at much, much greater rates than the tail. So I want to make sure I have some of both. In general, what you're going to find is that the head ASINs is going to be a lot more competition with different sellers than the tail ASINs. Now for my primary sample, I'm going to sample 18 categories and then four countries, US, UK, Germany, and France. And then the secondary sample, I'm going to have six categories for the other countries I can sample, Japan, Canada, Italy, Spain, and Mexico. So these are all the different categories. The only thing I want to note here is that what I find is that there are pretty big differences between media categories and non-media categories. And I'll try to show this in some of the slides later on. But so for the primary sample, I've got about 800,000 products and three and a half billion offers in the primary and secondary sample. I've got about a million products and more than four million. Now the data that I can get, so this is coming from the KEPA API, which I know that both Tiffany and Leon, I believe, have used at some point. So at the offer level, I've got data on the price, shipping cost, the seller type, whether this is Amazon, FDR, FBM, whether the product is in stocks, it's immediately shipable, and who won the buy box. I can also see some information about the sellers, like the number of lifetime feedbacks and the average rating. And for the product, I can see data about the average buy box price and how the buy box price moved over time. Now the main things that I can't see are first, so there's really one that's important, which is I can't see shipping time. Now what this is going to mean in practice, Amazon and FBA are going to have essentially the same shipping time. But the FBM sellers are going to be much more varied, both with longer shipping times, but also typically a range. So if it's an Amazon or FBA seller, it's going to say this is going to come Tuesday, March 9th, whereas for the FBM sellers, that may be a range of March 12th to March 15th, and that range may vary across the sellers. So I think that's the main limitation in terms of data of this study. I don't have data on seller fulfilled prime. So this was this program where the sellers could do prime themselves. But at this time period, that was minuscule and it shouldn't affect the results. I also don't have data on shipping quality, which again is going to matter for the third party FBM sellers that are doing their own shipping. I guess I'll stop for questions before I get into some of the empirics. I'll move on. So let me start with these are just descriptive statistics. So these are the number of offers. So what you see is that at median, for most of the categories, the number of offers are one. So just one person selling this. But the average is about two. And again, this is going to be more in the head aces that are selling a lot than the tail aces. But you should think median product, there's one seller average, there's two. It's quite different in the media categories. So for books, for example, there are at median five and an average about seven sellers for CDs and DVDs. It's like three or four at median four at five on average. So there's a lot more competition for these media goods than there are for the non-media goods. Now in terms of who wins the buy box, it might surprise you, but Amazon is often not winning the buy box. So this is just on average. This is not conditional on there being an Amazon offer. But for most of the product categories, Amazon is the winner about 10 to 20% of the time. If you include Amazon and FBAs, where Amazon is doing the shipping, it's winning about 25 to 50% of the time. That's again, a lot higher for the media categories. So if you look at DVDs, Amazon is winning more than 25% of the time for CDs, something like 40% of the time. And then for books, it's a very clear outlier here. Amazon retail is winning about 80% of the time. And if you look at UK books, it's winning 100% of the time. So it looks like whatever this buy box is doing, and if you remember from the slide before, there are a lot more offers for books than other categories. But at the same time, it seems like Amazon is winning much more often than the typical category. Now, how often is it that the winner is not just the lowest price item? So for the non-media categories, that's true about five to 10% of the time. For all products, more like 20% of the time where there are multiple offers, there's more than one offer, it is a lot higher in the media categories. So if you look at books, for example, the winner is not the lowest price, about 30% of the time. It's even higher for products with multiple offers for CDs and DVDs, video games, where it's 30 to 40% of the time. So again, once there are multiple offers, it's a decent chunk of the time when the product with the lowest price is not the winner. If you look at the percentage price difference, if the winner is not the lowest price, at median, this is pretty small. This is like a couple percent. The average is higher, maybe more like five to 10%. So these are small differences. If you remember the example I showed you earlier, the difference between Amazon's offer and the lowest price offer was something like 15 cents. So it's not breaking the bank, but there are these differences. Now, the empirical strategy I'm going to do is pretty simple. I'm going to try to predict the buy box winner looking at the different offers. So I'm going to do what we often do in an industrial organization and I treat the offer as a bundle of characteristics and then estimate how consumers, not how consumers trade off these characteristics, but how the Amazon algorithm trades off these characteristics. So instead of taking a car and saying that car is there's a price and there's a horsepower and there's a safety rating and so on, I'm going to take the offer and say there's a price, but there's also who does the fulfillment and the feedbacks and so on. Now, in terms of showing you these results, what I'm going to do is I'm going to compare Amazon to what I'm going to call a perfect seller. So that's somebody with 100% rating and 1 million feedbacks. So 100% rating is obviously the highest rating you can get here and 1 million feedbacks is about the max of what any individual seller has. Now, the reason I have to do that is that Amazon retail does not allow people to rate itself. So there's no way to directly compare Amazon to the third parties. That's something I'll talk about a little bit at the very end. Now, the offer characteristics I include are price and here I'm going to just compare price to the average buy box. That's going to allow Amazon to penalize sellers that say have high prices relative to the average. So you could imagine they don't want price gouging or something like that. Then there's the seller type. Then there's whether the product is in stock or immediately shippable and then the number of feedbacks in the average seller rate. And I'm also going to have an outside option here, which is that there is no winner. So again, that's the way the buy box is closed. And so the empirical strategy is going to be to normalize the seller type or the characteristics using price and to show you the price premium. That's a way to sort of make these results understandable. And then I'm going to compare Amazon to a perfect seller. That's a seller with a million feedbacks and 100% rating. So this is just for those of you who like math, this is the equation. I presented this to lawyers before, so I was like, we got to skip this equation. So these are the top level estimates. So these aren't, I think, super helpful on their own, but I think it's helpful, again, to just normalize these. So if there is an offer, so it's in stock, sorry. So first of all, having an offer is very much preferred, which is what you'd expect Amazon wants in general, somebody to win this. If the product is in stock, that's like a 16% decline in price. Going from 1,000 ratings to 100,000, 100,000 ratings is like a decline of a price of 5% for FBA, 3.5% for FBM. So just to be clear here, I'm interacting the rating things with both FBA and FBM. The reason I do that is that for the FBM seller, the seller is responsible for shipping as well as, say, whether the product works or not. Whereas for FBA, it's not responsible for the shipping, so you could imagine the feedback ratings and quality to be, to matter differently. And then for the rating difference, going from an 80% rating to 100% rating, so that's roughly speaking from the bottom 5% or 10% to the top percentile is like a 1.5% decline in price for FBA, 2.5% decline in price for FBA. Now here, the main object of interest is... Sorry, can I ask a question about the rating difference, please? Yeah, sure. So the rating typically is for the product, not for the vendor, is it? So I want to be, that's a great question. I want to be clear here, so Amazon does have product, does have ratings for the product. These are not the product level ratings, these are the ratings of the seller. So you can also rate the seller and that's the rating, you know, going from zero to 100%, but most of them are in the 80 to 100% range. And the number of feedbacks is, again, the number of feedbacks of the seller. So if you buy from a third-party seller on Amazon, you may get an email saying, I hope you really liked our service, please rate us on Amazon, you know, that's to affect these ratings, which then it may affect whether or not they win the buy box in the future. So I mean, one reason for this, just getting back to the very beginning slide, is that Amazon is going to worry about fraud and poor quality sellers, and so they may want to have sellers only when the buy box have had some customer experiences already where they know that this is a, you know, not a scammer or something like that. But so the main object of interest here is, is it an FBA seller or is it an FBM seller? So again, this is comparing Amazon to those sellers with a million feedbacks and 100% rating, and what this is finding is being an FBA seller is like a 16% higher price, being an FBM seller is like a 46% higher price. Now, this is similar if you look at all the different countries in terms of the signs and such. But now I want to go, so this is the main I think slide of this talk. So here I've got all the different categories broken out. And then the X axis is the price premium. The red lines are Amazon's retail over third party FBA. The blue lines are Amazon retail over third party FBM. So what you find is, for all of the sellers except, sorry, for all of the categories except auto, there is a significant price premium of Amazon or for FBA. It's roughly speaking for the non media categories between 10 and 20%. The premium over the FBM sellers is not surprisingly larger. That's more like 10 to 50%. Sorry, 20 to 50%. And then if you look at the media categories, it's much, much larger. So for example, the largest is books, where the retail, the premium over FBA is on the order of 50%. And over FBM is on the order of 150%. A lot of this seems to be driven by UK books. So here I'm aggregating across the US, UK, Germany and France. And in UK, at least it seems like Amazon always wins. And the algorithm is going, the multinomial logic framework is going to rationalize that as a really large premium. But I want to highlight, you know, if you think about pro competitive justifications, so you're worried about the quality of the product, like you might imagine for electronics, you're worried that the product doesn't work. Or in groceries, you're worried about the product is spoiled. It's a lot harder to rationalize those for the media categories, like books and DVDs, you know, so books don't spoil. It doesn't even really matter that much if they come in there so late. But it is exactly in these media categories, which around Amazon probably has the largest market power that you see the largest price premium. I can do this just restricting on products where only there's only a winner. So removing the cases when there's no winner, this reduces the premium slightly maybe by about a third, but you still see the same general pattern. I also look across countries. So this is just using the sample where I've got six different categories. I've got nine countries here. And you see the same general pattern across all the countries. And, you know, other than Mexico, I think the Amazon over FBA premium is actually pretty similar across the different countries. So this is a worldwide phenomena. You know, this may be changing as I'll talk about at the end of the talk. And I've got about eight minutes left. So I'll briefly talk about predictive power. So this algorithm is correctly predicting who wins about 90% of the time, about 80% of the time when there are multiple offers. That's actually pretty good. And if you look at other work of mine where we look at predicting which hospitals people go to, you know, we're getting that at maybe 30 or 40% of the time. So much, much less than this paper. If you want to know where we get things wrong or where I get things wrong, this is a confusing table. The columns are who we predict, whereas the rows are who actually wins. But it's the no winner that is very hard for this model to predict. That's where it's doing the worst at and to some extent, the FBM sellers. I think the reason for that is that a lot of the no winner and potentially also FBM maybe about other things like whether this seller is selling cheaper on another platform or potentially characteristics of shipping quality that both of which I'm not going to observe. Now, in the last few minutes, I'm going to talk a little bit about counterfactuals and a little bit about policy. So I'm going to do two counterfactuals here. The first is equating Amazon retail to a perfect FBA seller. The second is equating Amazon retail to both the perfect FBA seller and a perfect FBM seller. So again, this is still building in a whole bunch of advantage of Amazon retail over third parties, you know, because essentially you're comparing them against a seller that doesn't exist. But I think it's useful to try to see what would happen if you remove the self-prepensing both towards FBA, both towards Amazon retail and towards FBA. And I want to be very clear here, this is holding a lot of things fixed. You know, so if this was to happen in practice, which may happen soon, a lot of things are going to change on the platform. You know, so Amazon could change its revenue share and its fees. A lot of third party sellers could potentially enter because it's easier for them to win. They may change their prices. They may also change whether they use FBA or not. You know, so for example, if you removed a lot of the advantage of using FBA, I think that would change what they did. But I think these counterfactuals are helpful to try to see sort of a baseline with this algorithm is doing. So what I find is that equating Amazon retail to a perfect FBA seller, it doesn't actually have huge changes for what happens on the platform. So Amazon share of the buy box falls by about a half a percent. But Amazon plus FBA share goes up by about a half a percent. That's because Amazon is losing the FBA, but FBA is winning over FBM. On average, the buy box price rises a little bit by about 0.1%. That's because some of these FBA sellers are displacing FBM cheaper sellers. And then the average change in the buy box price when Amazon retail was the baseline winner is going to fall by about 0.1%. So that's cheaper FBA sellers winning out over Amazon. So these are pretty modest changes. You get very different results if you equate Amazon to a perfect FBM seller. Then Amazon share the buy box is going to fall by 7%. Amazon plus FBA by 10%. On average, the buy box price is going to fall by about a half percent and about a 2% when Amazon was winning before. Now this is a plot and so this is kind of a boring plot. If you equate Amazon retail to perfect FBA, the share changes, Amazon retails across categories are going to fall between 0 to 1%. Amazon retail plus FBA is going to rise by between 0 to 1.5%. So again, not huge changes. If you equate Amazon retail to a perfect FBM seller, Amazon share is going to fall by 30%, roughly speaking, in books and CDs by something like 10% to 20% in DVDs. So again, there are going to be huge changes in particular in the media categories where Amazon had this large Amazon had this large soft preference effect. All right, so I've got a couple minutes left. So I'm going to skip the conclusion to talk about policy levers. So you can think of three broad policy levers. So first would be some kind of platform regulation, which would be restrictions on the algorithm. The second would be structural separation. That's breaking the company up so they're no longer incentives to sell preference. And lastly would be user interface changes, which would be some way where you can maybe more easily see alternative sellers. Now I think it's easier to talk about this in the context of what the EU is about to do. So the EU commission investigation found, preliminary found that Amazon is unduly favoring its own retail business as well as its FBA business. This is relatively recent. I think this is coming from June or July and that this kind of bias can harm sellers, harm marketplaces, harm consumers, and also harm alternative delivery vehicles. So the EU, I guess, agrees with me. So that's good. Not that I'm saying that this is a bad thing, but certainly finding self-preferencing. So Amazon has given an offer, which takes two of the three different policy levers. The first is a platform regulation. It says it's going to equally treat all sellers when ranking their offers. I'm not exactly sure how they're going to do this because as I showed you, there are a bunch of different metrics where Amazon doesn't have a comparison to the third parties. So exactly how you do equal treatment when you have variables for Amazon, but I don't have variables for Amazon, but have variables for the others is not so obvious. And again, it might be hard for a regulator like the commission or the antitrust agencies to know whether Amazon is really doing this equal treatment or not. And second, they've put in a user interface change, which is to display a second competing offer to the buy box winner. There's a second offer that's different enough. So this is a change that says, well, there's some consumers that say want the lowest price offer or want a different type of offer, they can see both. And maybe they instead of just blindly clicking the one click button, they'll make a more informed choice. But so I think the EU commission now is deliberating on whether or not to accept Amazon's offer or not, which goes beyond this. It goes into other things like what it takes to win prime and so on. But I think it's quite interesting because it's quite possible that the algorithm changes relatively soon. So with that, I think I'm just at time. So I'll let Leon take over. Perfect. Thank you. Thank you so much, Devesh. Tiffany, do you want to introduce me or should I? Yeah, just go ahead. Yeah. So we are very happy to have Leon to have a short discussion of the paper. Perfect. Well, thank you so much. And thank you so much for the organizers for getting me to discuss this paper, because I've been looking for a reason to read this one really carefully, given how closely related it is to my own research. And I think Devesh does an amazing job here analyzing a really topical question, the self-preferencing like he emphasized is really important for EU regulators and also US regulators at the moment. And I wanted to start out by saying that I find it very reassuring that we at least qualitatively find very similar results in our papers, even though we use completely different data sources. So Devesh uses this extensive data from keepers has way more products and has, you know, sort of the scrape data, which I think here is actually a positive in that Devesh, you have the data that the users actually see right the way keeper collects this data is when the user actually visits the product page. They are collecting what the user sees. So I think that is exactly the sort of data you would want to use to analyze that. And using that data, you find something like a 10% roughly benefit in terms of price to Amazon over FBA. And I think you said like 20 to 50% for Amazon over FBM. And we find in our paper something like 14% and 28%. So these are roughly actually quite well aligned, given that these are different data sources and for sure there's composition things and so on. Now I did want to mention a couple of things here that I just think might make the paper even stronger. The first one is just a sort of a general comment about framing. You refer to the practice that Amazon does a lot as self-preferencing. And that's also a term I use in my paper, but I think people are a little dubious as to whether it's self-preferencing when it could just be a sort of like a quality effect. You have this table with all these various things that feature in the algorithm that you can't observe. And of course, Amazon could be better on all of these things. And it's a bit unclear to me whether it's then a self-preferencing or whether it's just like a fair preferencing of things that are unobservable to you. And I agree it's a tight rope to walk. And I also have trouble communicating appropriately the concern, but yet the inability of us to see everything. And of course, Amazon won't share their data. So it is a bit tough there. There were two modeling concerns I wanted to briefly talk about because they came up when we initially modeled the Bybox. And I think they improved our estimates. And given what I saw from your estimates, I think you should try them. The first one is that I strongly believe this is actually a nested loger model, which is to say that Amazon more easily switches between inside offers than switched to saying, oh, actually, I'm just not going to show you anything in the Bybox. I think that only happens when really all offers are quite bad. And when we switched to the nested loger, our price coefficient went up by a lot. And I think it could explain why you find in the media categories quite different results. Because the media categories, as you showed earlier, have way more offers. And when you have way more offers, that's when the difference emerges between the nested loger and the non-nested loger. And similarly, it could explain why you find the situation where there's no winner so hard to predict. So I think that I would definitely just try to run that. And the second modeling concern I had is that you don't talk at all about price endogeneity at the moment. We normally, when we estimate demand systems, we worry that the price coefficient is biased towards zero, because offers that are better are also going to have prices that are higher. And I do think that's a relevant concern for you because you do divide all the FBA indicators and so on by the price coefficient to translate them into human understandable terms. And so it is important for you that you estimate that correctly. I will say though in your defense that when we did this in our paper, we didn't find that the results changed a lot. They changed a little bit, but I don't think it will change any of your conclusions at all. So yeah. Okay. And these were sort of just some concrete modeling concerns, but I wanted to say that in general, I think this is exactly the right approach. I've heard rumors that at some point, I doubt while we analyze the data, but at some point, the Bible might even literally have been eloged. So I think we are not that far off from sort of modeling it. So I think this is a really nice paper and a really topic question. And with that, I think I should pass it over to the Q&A for general discussion or back to Tiffany, actually.