Marks

Load efficiency: two indexes

Raw

Mechanical work divided by heart-rate load (TRIMP). Higher is better. It also moves with the session: an easy week reads higher on its own.

Adjusted

How much heart-rate load the session cost, against what the same amount of work usually costs this player. 1.00 is his usual. 1.10 means 10% less heart-rate load than usual for that work (better). 0.90 means 10% more (worse).

Which one

Read a player with Adjusted. Use Raw while Adjusted has no value yet: it needs 15 sessions in the current training block.

How Adjusted is calculated

1. Take the player's sessions in the current training block (a break of 14 days or more starts a new block), leaving out the session being scored. At least 15 are needed.
2. Fit a straight line to log(TRIMP) against log(mechanical work): log(TRIMP) = a + b × log(work). The line is the player's own cost curve.
3. For the session, predicted TRIMP = e^{a + b × log(work)}. Adjusted = predicted TRIMP ÷ actual TRIMP.
4. A session under 40 units of mechanical work or 20 TRIMP, or with work more than 20% outside the range the line was fitted on, gets no value.

The arrow

The median of the last sessions against the median of the sessions before them, inside the current training block. Up or down only when the difference is bigger than the player's noise band, otherwise steady. The popup shows the band.

The noise band, per player

1. Take how much the player jumps from one session to the next, over all his sessions in the current training block (at least 13 sessions).
2. Read the typical size of those jumps in a way one wild session cannot inflate (the median absolute deviation, scaled like a standard deviation, divided by the square root of 2). This is his typical error for one session.
3. The two block medians carry less noise than one session: 1.25 × typical error × square root of (1 ÷ recent sessions + 1 ÷ sessions before). This is the standard error of the difference.
4. Band = 2 × that standard error, or 3% of the level if that is larger. A consistent player gets a narrow band, a jumpy player a wide one, and more sessions in the blocks make it narrower.

Example: a player who typically jumps 0.39 between sessions, read at 5 vs 20, has a band of 0.49; on a baseline of 2.05 that is 24%, so his arrow moves only for a change bigger than that.

Sessions compared

5 vs 20

The last 5 sessions (about 8 days) against the 20 before. Answers fastest and sees only big changes, about 25% or more. For the week.

7 vs 20

The middle ground between the two.

10 vs 20

The last 10 sessions (about 3 weeks) against the 20 before. Steadier, fewer false changes, sees changes of about 20%. For a decision about a player over a month. Needs 30 sessions.

Default

Adjusted, 10 vs 20. Your pick is remembered on this browser. The attention flag reads Raw 5 vs 20 from the daily calculation.