Your log already ran the exercise comparison study you keep searching for.

The hamstring question sitting one app over

A lifter with four years of logs types "Romanian deadlift vs stiff-leg deadlift" into a search bar and gets the same ten articles everyone gets. EMG comparisons. A diagram of hip angles. A conclusion that says both work and it depends.

It does depend. On them, specifically.

And the thing it depends on is sitting one app over: a spring block of Romanians, a winter block of stiff-legs, a summer where both showed up in the same week because the program called for it. Every set, every load, every RPE, with a date on it. That lifter has already run the experiment, maybe more than once. They just haven't read the results.

The accidental crossover trial

Clinical researchers have a name for a study where the same subject gets treatment A, then treatment B, and serves as their own control. It's a crossover trial, and it's one of the cleanest ways to strip out between-person noise, because the person never changes.

Serious lifters run crossovers without meaning to. Programs rotate variants every block. Injuries force swaps. Boredom forces more. Across three or more years of consistent logging, most of the hinge, press and squat variants a lifter cares about have had at least one honest run, with load, reps and effort written down.

Call it the accidental crossover. The design is messy, there's no blinding, and the washout periods are whatever life handed over. But it's the only dataset anywhere that speaks to the question the lifter is actually asking: what does this exercise do for this body?

Which exercises build muscle fastest? The research answers a different question

The hypertrophy literature is good and getting better. But a trial comparing two exercises reports a group mean, and a meta-analysis pools group means. Its answer is accurate for the average subject in those samples, who is often less trained than someone with years of logged blocks behind them.

Inside those averages, individuals scatter widely. When researchers put a large group of men and women through the same supervised arm-training program, some people barely changed while others gained a lot in both size and strength. Same program, same supervision, very different outcomes.

So the average is a fine starting point and a poor finishing point. Generic content has to stop at the average because it can't see anyone's squat history. A four-year log can.

What a working exercise looks like in the log

Exercise response leaves a signature in training data, and reading it doesn't take a lab. Five signals carry most of the information, and a decent log already records all of them.

Early strength gains on a new variant are largely skill acquisition. So the useful number is how many weeks the climb keeps going after the movement stops being new. A variant that keeps paying out in weeks five through eight is telling you something a fast first fortnight can't.

Incline vs flat: walking one comparison through

Say a lifter in their fourth year ran flat dumbbell press as the secondary press in one hypertrophy block, and a low-incline press in a comparable block the following year. Same three sets of eight to ten, twice a week, same bench program on the main day, bodyweight roughly stable. (The details here are illustrative.)

Put the two blocks side by side and the read takes about five minutes. On flat, the top set added weight for the first three weeks, then sat at the same load at RPE 9 for the rest of the block. No rep PRs after week three. On incline, progress started slower. But RPE at the opening working weight drifted down by a couple of points by week five, rep PRs showed up at two different loads, and the main bench moved during the block.

That's a verdict: incline, for this lifter, in this kind of block. The flat block still did something. Its curve is the classic novelty shape, fast early gains from learning the movement and then a wall.

One comparison is a single data point. When a second pair of blocks a year later points the same direction, that's about as strong as single-lifter evidence gets.

The meta-analysis is the prior

If the accidental crossover holds up, exercise selection picks up a step most lifters skip: a retrospective audit before the next program gets written. Read the research to decide which variants are reasonable candidates. Then read the log to see which of those candidates this body has already voted for.

That changes what rotation is for. Plenty of lifters swap variants on a schedule, or because a movement got stale, and the swap carries nothing forward. A lifter who reads their own blocks rotates on evidence. Keep the variant with the late plateau and the falling RPE. Retire the one that stalled in week three twice. Retest anything with only one block on record.

And it ends a specific kind of waste. Sumo vs conventional, low-bar vs high-bar, Romanian vs stiff-leg: these get argued in forums for years because the population answer is close and individual bodies differ in limb lengths and mobility. For one lifter the gap can be wide, and the evidence has been piling up in their own sets the whole time.

The meta-analysis is the prior. Your log is the posterior.

Reading three years of blocks without a spreadsheet weekend

The barrier here is mostly friction. The data exists, but it's spread across an old app's export, a current app, maybe a notebook, and lining up two blocks means an evening of pivot tables most lifters never get around to.

Platepusher is built so that history stays readable. A CSV from Strong or Hevy imports every lift, set and date as native data, so a Romanian block from three years ago sits on the same exercise history as last month's sessions. Plateau detection flags the week a top set stops moving, which is the plateau-onset row from the table above, marked on the chart without a calculation. The lifter does the reading. The instrument keeps the record straight. The earlier piece, Your training history is the experiment, makes the wider case for keeping that record whole.

Before reading another meta-analysis on exercise selection, open the log and find the last two times you ran the variants you're deciding between. There's a decent chance the question is already settled.