Here are a few fantasy baseball thought experiments designed to raise questions about how we go about constructing our rosters:

For starters, let's imagine instead of buying/drafting players for this year's season, we bought/drafted players for 2015. Given it's February of 2016, that means we'd have the advantage of knowing exactly what happened. Instead of looking at spread sheets filled with stat projections, we'd look at ones filled with last year's results. Every player's stat line (unlike his projection) would be equally reliable. There would be no surprises - no Carlos Correas or Dallas Keuchels to be taken in the late rounds. How would the results differ from a draft or auction for this coming season?

For starters, the first 5-10 players drafted should be pitchers. This is because pitchers contribute to five of your total 10 categories, i.e., they are just as important as hitters. When we use projections as the basis of our dollar values, normally there's roughly a 70/30 split in the apportionment of funds toward hitting. But that's because (according to Peter Kreutzer) there's more "free loot" on the waiver wire when it comes to pitchers than hitters. Moreover, drafted pitchers as a group don't live up to their values as well as hitters. The reason for this is too many pitchers have negative value, given the ERA and WHIP blowups, whereas few drafted hitters have negative value from just bad batting average. But when you compare top pitchers to top hitters, the former actually hold their value even more reliably than the latter. And in our hypothetical, everyone's value holds up perfectly because we're simply drafting their 2015 stat lines, not their prospective 2016 seasons.

In a scenario, then, where there's no free loot (waiver wire adds) and reliability isn't a factor, the pitching/hitting split ought to be 50/50, not 30/70. In that case, Clayton Kershaw, who earned roughly the same as Mike Trout on a 30/70 basis, absolutely destroys him on a 50/50 one. Put differently, Kershaw's massive contributions to four categories when there are only nine pitching slots (2-3 of which will be filled by low-impact closers) dwarf Trout's contributions to five categories when there are 14 hitting slots (admittedly, catchers and middle infielders might make it more like 12 full hitting slots.)

It would be interesting to hold drafts from last year's player pool and results, let people take Trout, Bryce Harper and Paul Goldschmidt with the first three picks and see them lose every time to the teams that took Jake Arrieta, Kershaw and Zack Greinke.

What does this mean for our leagues in 2016? It seems Kreutzer did a good job in explaining why the 70/30-ish split makes sense, but when you look at the elite pitchers, they're more reliable than pitchers generally. It might make sense to price the top-10 pitchers at 63/37 or so, and maybe Kershaw - whose reliability in the ratio categories has been unmatched - at 60/40. If you do that, Kershaw is easily the first pick, and you'll probably take Max Scherzer or Chris Sale (or whatever pitcher you've projected higher) in the late first or early second round.

Another thought experiment would be to use, say, 20 pitchers instead of nine. How would that change the apportionment of money or draft slots? On the one hand, you'd think pitchers would increase in value because you'd have more slots to fill, but on the other elite starters like Kershaw would occupy a much smaller portion of your team's overall stats. Moreover, you'd still need to compete in five hitting categories, so even if your roster required 40 pitchers, you couldn't skimp that much on hitting. In the end, I think it would push up the value of middle relievers and low-end starters significantly, but would that extra value come out of your pitching or hitting budget?

A third thought experiment is to imagine hitting projections were completely unreliable, i.e., a hitter's projections had zero correlation with his actual results. In that case you'd spend $14 on hitting and $246 on pitching because there would be no point in prioritizing one hitter over another. That's the easiest case, but what if hitters were 70 percent likely to be within one standard deviation of projected their roto values and pitchers were only 50 percent likely to do the same?

Clearly, pitching projections would have to be discounted more significantly than hitting ones. If that were the case, though, wouldn't that kind of regression be priced into the pitching projections themselves? Put differently, wouldn't Steamer, knowing their pitching projections were inherently more uncertain, simply regress outlying pitching stats toward the league mean more than they would for hitters? If your model said Kershaw is likely to strikeout 285 hitters this year, but you also knew pitchers were less likely to be close to their projections, wouldn't you simply give him (and other outliers) fewer strikeouts if you wanted to be more accurate overall?

But if in fact, there were more uncertainty with pitching projections, and it were baked in by making them more conservative, we'd be double-counting if we also re-apportioned pitching dollars to the hitting side of our budgets.

Setting that aside, it's clear projected stats alone don't tell the entire story, and the reliability of the projections (assuming it's not baked in by making them more conservative) is a separate variable that should greatly affect our valuations. While hitting projections might be more reliable than pitching ones (I'm not sure about this, but it's plausible at least given the volatility of ERA and WHIP and the team dependency (and smaller sample) with wins and saves), it's important to know the precise extent.

Finally, just as Kreutzer found a subset of pitchers who were every bit as reliable as the hitters (at least after discounting them via the 70/30 split), we should be conscious of not overgeneralizing when constructing our budgets. In other words, maybe the unreliable subset of pitchers should be discounted significantly from their projected stats while others less so, if at all. To that end, perhaps the ideal roster construction in an AL- or NL-only would be the two best pitchers for $65, two closers for $30, and five $1 pitchers for $5. That would be $100 on pitching, $160 on hitting for a 62.5/37.5 split. Or you could buy one closer for $15 and six pitchers for $10. That leaves you at $170/$90 or a 65/35 split. You'd be spending more on the pitching that pulls its weight and less on the negative subset without sacrificing too much hitting.

These and other thought experiments we might entertain - like imagining every player got the same number of plate appearances or innings pitched, assuming every player had perfect health or adding a replacement-value stat-line for every game a player misses - might distort our draft boards slightly, but they also open our eyes to possibilities we're overlooking.