When should the starter come out, and who should come in?

Shorten Keller: two times through.

Mitch Keller: two times through (about 18 batters, 66 pitches), +3.3 ± 1.6 runs and 29 relief innings at the plan’s 128 innings. Ended 2026 on the 60-day IL (Right arm teres major muscle injury). The plan assumes he is ready for spring; that is a medical call we can't see. The call leans on that season; on 2024–25 alone he grades two times through, a toss-up (+.022 ± .015 against the bar; first two passes .318 expected vs. .350 in 2026). The innings go to the 2027 pen’s 4th–8th arms (.309 after a penalty for hitters seeing a reliever twice); its multi-inning arms are Dotel (8.1 batters an outing) and Mlodzinski (11.7 batters an outing).

If Keller isn’t ready: Keller opens 2027 on the injured list: −0.3 wins. Starter innings: Keller 128 → 94 IP. No one else on the 40-man picks up starts. 35 starter IP go to replacement level.Keller and Jared Jones both miss the season: −2.9 wins; Mlodzinski starts: he grades tandem (a toss-up), but at the branch’s 105 starter innings that hands the pen 59 innings, more than the 56 left under the cap, so two times through (14 relief innings, +2.3 ± 1.0 runs); Dotel takes over as the long man (8.1 batters an outing); Harrington starts (no call: too few 2026 starts).

Third time through: Skenes, Ashcraft and Jones.

Skenes: third time OK, clear of the line, pending the review of his −1.4 mph fastball drop. Ashcraft: third time OK, but a toss-up. Jones: by matchup (the bottom of the order a third time), a toss-up by nature. A toss-up sits within one standard error of a call line; one season of third-time results can’t separate those calls.

Jones, back from a 60-day IL stint (right elbow UCL sprain) that cost him 2025: a leash of 22 batters (about 81 pitches) instead of an open third time. It hands the pen 2 innings and costs about nothing (± 0.1 runs). The leash is a club setting the model prices, not a model output.

Ashcraft: 15-day IL Sep 10–28 (right arm discomfort); innings +40% (118 → 165, majors and minors). The call is on performance; none of it is a leash trigger here (a 60-day arm stint or a lost season), so it’s a Health & Workload watch item.

Development innings for Chandler, on a leash.

He grades two times through (.014 worse than the relief bar the third time), but at 24 the club can buy him third-time innings. Leash: 22 batters (about 82 pitches), once through the top of the order a third time. Cost against a two-times-through hook: 1.0 ± 1.2 runs.

Goal: his first two passes (.310 expected) are worse than the relief bar (.309) by .002. At .306 or better his third time comes off the hook; a third-time skill isn’t the target (one season of that doesn’t repeat). A season of first-two-pass batters (about 462) measures it to ± .017, more than the .004 he needs: a direction to work toward, not a pass/fail test. His fastball holds the third time through (+0.3 mph vs. MLB starters’ −0.4, 28 starts); keep it there. Finish 2027 without a late-season velocity warning (he ended 2026 −1.3 mph against his baseline; Health & Workload). The leash is a club setting; the model prices it.

2027 ladder: Montgomery and Ramírez late; Sisk and Weaver set up.

Ranked by skill within the 2027 pen; Montgomery, Ramírez, Sisk and Weaver are each within one standard error of the next, so their order is a toss-up. Montgomery LHP, Ramírez RHP, Sisk LHP and Weaver RHP. Montgomery: arm angle −6.9° late in 2026, the largest change on the staff; on its own a context sign, but get the trainer’s read before he pitches the late innings. A command caveat: Pitch Lab lists Ramírez (Location 79, 3rd percentile; 12.2% walks) and Sisk (Location 75, 2nd percentile; 12.7% walks) for command work, and a walk costs more in a big spot. The Winter Plan lists the same arms in its win engine’s order, which sets their innings (Montgomery, Mlodzinski, Dotel and Weaver first); read that order as innings, not roles.

Gone since 2026: Soto (34), Santana (14), Doval (5) and Lawrence (5) took 40% of the club’s big-spot entries. Priced context-neutral at the 2027 pen’s 4th–8th arms (.300), the departures’ 2026 batters come to Soto +3.5, Doval +1.0, Santana 0.0, Bidois −2.4 and Lawrence −4.0 runs: −1.9 net (−0.2 wins). Context-neutral runs leave out leverage; the big spots move up to the ladder’s top arms.

Replacing the weakest arm the plan keeps, Brandon Eisert (skill −0.85 ± 0.40, 52 innings in 2027), with a reliever the Free-Agent Board prices for Pittsburgh. Two prices of each: this page’s skill model on his 52 innings, and the board’s gain from adding him to the plan (the win engine moves innings off the lowest-rated arms):

• Gregory Soto LHP · re-sign · skill +0.74: +3.3 runs on Eisert’s innings; the board +0.51 wins (≈+5 runs) for 1 yr × $9.2M, 0.55 wins per $10M.
• Adrian Morejon LHP · skill +2.85: +9.6 runs on Eisert’s innings; the board +1.38 wins (≈+13 runs) for 2 yr × $23.2M, 0.60 wins per $10M, at the board’s price cap (an extrapolation past last winter’s market), a qualifying-offer toss-up (may cost a pick).
• David Bednar RHP · skill +0.55: +4.9 runs on Eisert’s innings; the board +0.90 wins (≈+9 runs) for 2 yr × $16.9M, 0.53 wins per $10M.
• Sean Newcomb LHP · skill −0.39: −1.9 runs on Eisert’s innings; the board +0.62 wins (≈+6 runs) for 2 yr × $12.6M, 0.50 wins per $10M.

For scale, the Winter Plan’s spending table: Jack Flaherty 0.98, Isaac Paredes by trade 0.90, Jazz Chisholm Jr. 0.68, Lowe on the qualifying offer 0.57, Lowe re-signed 0.46 wins per $10M. Money: under today’s rules there is no floor, so each costs his salary. Under the owners’ proposed floor (a proposal, not a rule) the plan’s 2027 payroll sits $34.0M below it ($38.0M in 2028), room for any one of these deals: that money is free if nothing better would use it, and costs face value if it displaces an equally good use. The Winter Plan signs none of them and keeps Eisert. Roster Builder.

Protect Isaac Mattson from back-to-backs.

On the second straight day he loses 1.0 ± 0.4 mph against his rested outings within 14 days (9 back-to-backs); the average reliever loses 0.2.