# Corey’s Double Play — 2026 post-season decision audit

**League:** Yahoo H2H-Pts 203711  
**Team:** Corey’s Double Play (team 5)  
**Evidence cutoff:** September 27, 2026, after all championship games reached 0 remaining  
**Final-score note:** Yahoo shows **Corey’s Double Play 863.3, Sane Moderate Person’s Gems 701.0** with 113/113 and 109/109 games complete. Ordinary postgame stat corrections can still adjust a fantasy total, but no MLB games remain.

## Evidence labels

- **Exact:** directly reported by Yahoo or calculated from Yahoo’s preserved scoring-category totals using the league’s scoring formula.
- **Derived:** calculated from exact inputs; formula is stated.
- **Estimated:** a reconstruction with stated assumptions.
- **Unavailable:** Yahoo’s surviving interface does not expose the required historical state.

## 1. Executive assessment

Corey’s Double Play finished the 22-matchup regular season **16–6, second of 10**, with **17,513.5 points**, also second in the league. The team earned a first-round bye, beat ⚾️Young Guns 🔫 **896.4–670.1** in the semifinal, then defeated Sane Moderate Person’s Gems **863.3–701.0** to win the league championship and finish **17–6** in Yahoo’s playoff-inclusive record. The archived early-week projection slightly favored the opponent, **816.36–805.55**. After the games closed, Yahoo displayed a recomputed “original” projection of 797.47–787.10; both versions made Corey an approximately 10-point underdog, but the discrepancy shows that Yahoo’s projection field is not immutable. Sources: [final standings](https://baseball.fantasysports.yahoo.com/b1/203711?module=standings&lhst=stand#lhststand), [playoff bracket](https://baseball.fantasysports.yahoo.com/b1/203711?module=standings&lhst=playoff#lhstplayoff), [championship matchup](https://baseball.fantasysports.yahoo.com/b1/203711/matchup?week=26&mid1=5&mid2=7).

The clearest explanation for the season is **a high-output drafted spine plus unusually active replacement and streaming work**. Across every week preserved in Yahoo’s team log, drafted players produced **11,774.2 captured points (59.6%)**; players not drafted by Corey produced **7,972.7 (40.4%)**. The 19,746.9 denominator includes Weeks 2–26: 17,513.5 regular-season points, 473.7 during the Week 24 bye, 896.4 in the semifinal, and 863.3 in the championship. Those figures preserve only statistics earned for this fantasy team; they are not full MLB-season player totals. Source: [Corey team log](https://baseball.fantasysports.yahoo.com/b1/203711/5/teamlog).

### Strongest advantages

1. **Drafted star/core performance.** Yordan Alvarez (1,532.3 captured points), Brandon Lowe (1,169.8), Yandy Díaz (1,122.6), Freddie Freeman (1,107.1), Bryan Reynolds (1,073.2), Juan Soto (962.9), Oneil Cruz (929.6), Cristopher Sánchez (870.4), and Gavin Williams (834.6) all cleared 800 captured points.
2. **Pitching replacement depth.** Non-drafted pitching supplied major multiweek contributors, not just one-day darts: Drew Rasmussen (571.1 captured), Taj Bradley (462.5), Logan Henderson (379.5), José Soriano (275.5), Bryce Miller (200.8), Michael Soroka (194.4), Tyler Glasnow (165.6), and others.
3. **Playoff adaptability.** The team used all six championship-week acquisitions; Zebby Matthews, Drew Anderson, Connor Prielipp, Trevor Rogers, Kade Anderson and Bryce Miller combined for **88.9 captured points**.

### Largest weaknesses

1. **Shortstop and third-base instability.** The drafted answers, Geraldo Perdomo and Alex Bregman, produced only **37.8** and **35.0** captured points. Replacement volume prevented a total collapse, but the churn consumed attention and acquisitions.
2. **Excessive roster motion at the margin.** Yahoo records **117 moves**, second only to the opponent’s 120. The team extracted real value from activity, but several short-term pitching starts were negative and some churn merely exchanged similar low-ceiling options.
3. **Lineup-state observability.** Yahoo’s season team log preserves captured statistics but not a downloadable day-by-day active/bench ledger. Exact season-long bench points, empty-slot losses, and late-scratch avoidability are therefore unavailable from the surviving interface. This is both an audit limitation and a process-design weakness: the manager should preserve daily lineup snapshots next season.

## 2. League environment and settings

| Setting | Exact value |
|---|---:|
| Teams | 10 |
| Format | Head-to-head points |
| Active hitters | C, 1B, 2B, 3B, SS, 3 OF, 2 Util |
| Active pitchers | 2 SP, 2 RP, 4 P |
| Bench / IL | 5 BN / 4 IL |
| Weekly adds | 6 |
| Season adds | Unlimited |
| Minimum innings | 20 per week |
| Lineup deadline | Daily — Today |
| Playoffs | 6 teams, Weeks 24–26; top two seeds bye |
| Playoff tiebreak | Higher seed |
| Trade deadline | August 6, 2026 |

Scoring strongly rewards power, strikeouts, length, wins, and saves while imposing meaningful damage for baserunners and earned runs. Batting: R 1.9, 1B 2.6, 2B 5.2, 3B 7.8, HR 10.4, RBI 1.9, SB 4.2, BB/HBP 2.6. Pitching: W/SV 8, OUT 1, H/BB/HBP −1.3, ER −3, K 3. Source: [Yahoo league settings](https://baseball.fantasysports.yahoo.com/b1/203711/settings).

## 3. Season timeline

| Phase | Evidence-based inflection points | Decision-quality read |
|---|---|---|
| Draft | Balanced 23-player build; nine drafted players later captured 800+ points. The vulnerable spots were SS/3B and several injury/return bets. | Strong overall process; risk was concentrated in the middle/late rounds rather than at the top. |
| Opening month | April 9 reset: Chapman, Perdomo, Bregman and Iglesias were dropped for O’Brien, Dubón, Giménez and Cantillo; Iglesias was reacquired April 21. | Decisive, but abandoning both drafted SS/3B answers together created a season-long repair burden. |
| Early season | May cycled Correa, Crawford, Brooks Lee, Rocchio and Schmitt through infield roles while pitching churn accelerated. Altuve was dropped May 13. | The pitching shift fit K/OUT/W scoring; the hitter decisions require contemporary injury/news context. |
| Midseason | June–July became a rotation laboratory; Rasmussen arrived June 5, Logan Henderson July 17, Kwan July 26 and Hoerner July 30. Foundry grading began in June. | Good adaptation produced durable holds, but familiarity/re-add bias remained a risk. |
| Trade-deadline period | August 6 passed without a documented trade changing the core. Raleigh was dropped Aug 13 and reacquired Aug 25; Stott, Sogard and injured/prospect arms rotated through. | Waivers were the main lever; Raleigh’s 12-day reversal merits scrutiny. |
| Playoff push | Hoerner stabilized MI for 50 days, then September shifted to Sanoja/Dubón plus repeated SP scheduling. | Aggression was justified by the value of the No. 2 seed and bye. |
| Semifinal | Beat Young Guns 896.4–670.1. | High-margin result; do not infer that every daily decision was optimal from a 226.3-point win. |
| Championship | Entered as an approximately 10-point projection underdog, used all six adds, attacked early pitching volume, built a 168.5-point lead by Saturday, and won 863.3–701.0. | Process improved title odds; the 162.3-point win also contained large favorable pitcher variance. |

## 4. Draft audit

Overall draft grade: **A− process / A− outcome**. The team did not draft a perfectly stable roster; it drafted enough elite and mid-round surplus to survive major misses. Captured points below are exact team-log calculations through the completed Week 26 matchup, including playoff and bye-week production. “Held time” and full raw MLB production are included only where transaction evidence supports them; otherwise they are unavailable rather than guessed.

| Rd | Overall | Player | Captured pts | Process | Outcome | Audit note |
|---:|---:|---|---:|:---:|:---:|---|
| 1 | 4 | Juan Soto | 962.9 | A | A− | Yahoo preseason ADP 4.2; correct anchor, with availability limiting captured volume. |
| 2 | 17 | Cal Raleigh | 633.3 | B | C+ | Early catcher investment bought an advantage when active, but not a full elite season. |
| 3 | 24 | Cristopher Sánchez | 870.4 | A | A+ | Yahoo preseason ADP 27.1; strong price and ace-level captured return. |
| 4 | 37 | Yordan Alvarez | 1,532.3 | A | A+ | Yahoo preseason ADP 40.2; a mild reach that became the roster’s top captured scorer. |
| 5 | 44 | Freddie Freeman | 1,107.1 | A− | A | Stable high-end production at a fair cost. |
| 6 | 57 | Geraldo Perdomo | 37.8 | B− | F | Reasonable SS solution on draft day; availability/result failed. |
| 7 | 64 | Oneil Cruz | 929.6 | B+ | A− | High-variance profile paid off when active; 30 captured steals mattered in this scoring. |
| 8 | 77 | Alex Bregman | 35.0 | B | F | A sound positional-stability idea that failed through availability/result. |
| 9 | 84 | Jose Altuve | 82.1 | B | F | Plausible value bet; little production was captured. |
| 10 | 97 | Aroldis Chapman | 22.8 | B− | D | RP role/value did not persist for this roster; only three captured appearances. |
| 11 | 104 | Shohei Ohtani (P) | 441.9 | A− | A | Injury/role uncertainty was priced in; 69 captured appearances created real surplus. |
| 12 | 117 | Yandy Díaz | 1,122.6 | A | A+ | One of the draft’s best values and an ideal points-format profile. |
| 13 | 124 | Raisel Iglesias | 446.4 | A− | A | Thirty captured saves plus useful ratios/strikeouts. |
| 14 | 137 | Gavin Williams | 834.6 | A | A+ | Late rotation anchor; among the season’s clearest skill wins. |
| 15 | 144 | Tanner Bibee | 27.6 | B | D | Defensible SP depth; six captured appearances did not work. |
| 16 | 157 | Bryan Reynolds | 1,073.2 | A | A+ | Massive late-round surplus and lineup stability. |
| 17 | 164 | Brandon Nimmo | 169.1 | B | C− | Useful in a short captured window, but not a season-long hit. |
| 18 | 177 | Roki Sasaki | 19.9 | B− | D | Upside thesis was defensible; five captured appearances were low value, and he later appeared on the championship opponent. The drop deserves more scrutiny than the pick. |
| 19 | 184 | Shane McClanahan | 54.8 | B | C− | Appropriate late injured-player speculation; limited usable return. |
| 20 | 197 | Adley Rutschman | 0.0 | C− | F | A second catcher after a Round 2 catcher had poor marginal utility in a one-C league; no captured points. |
| 21 | 204 | Brandon Lowe | 1,169.8 | A | A+ | Best late-round hitter outcome; 30 HR and 86 RBI were captured. |
| 22 | 217 | Aaron Nola | 61.4 | A− | C+ | Excellent price for the talent, but only three captured starts. |
| 23 | 224 | José Caballero | 139.6 | B | B− | Flexibility and steals produced a useful short stint. |

Best selections by combined process and outcome: **Yordan Alvarez, Cristopher Sánchez, Yandy Díaz, Gavin Williams, Bryan Reynolds, Brandon Lowe**. Worst outcomes: **Perdomo, Bregman, Altuve, Rutschman, Sasaki**. The first three are important examples of outcome failure that should not automatically be relabeled as obviously bad process.

## 5. Position-by-position evaluation

Exact historical points by lineup slot are unavailable. Yahoo’s team log attributes captured production to players, not to the C/1B/2B/etc. slot occupied on each date. The table therefore reports player-source pools and a qualitative league-strength estimate; multi-eligible players can appear in more than one discussion but are not double-counted in team totals.

| Position | Evidence | Comparative read | Verdict |
|---|---|---|---|
| C | Raleigh 633.3; Liam Hicks 760.3 from C/1B eligibility | Strong aggregate catcher-source production, though exact C-slot share is unavailable | Not a true season weakness |
| 1B | Freeman 1,107.1; Yandy Díaz 1,122.6; Hicks also eligible | Elite depth and stability | Major strength |
| 2B | Brandon Lowe 1,169.8; smaller stints from Altuve, Giménez, Hoerner, Stott and flex bats | High production despite some churn | Strength |
| SS | Perdomo 37.8; replacement work from Caballero, Crawford, Rocchio, Hoerner, Stott, Sanoja, Dubón and others | Low stability; replacement pool prevented disaster | Genuine process weakness, less damaging than it felt |
| 3B | Bregman 35.0; Schmitt 277.2, Dubón 198.9, Karros 165.6 and multi-eligible replacements | High churn, modest ceiling | Genuine weakness |
| OF | Soto, Yordan, Cruz, Reynolds and Nimmo supplied 4,667.1 captured points before counting later OF pickups | Top-tier | Major strength |
| Util | Exact slot attribution unavailable; deep 1B/OF pool kept the slots populated | Strong, but cannot quantify exact utilization | Strength |
| SP | Sánchez/Williams plus Rasmussen, Bradley, Henderson, Soriano, Miller, Soroka and others created heavy volume | High ceiling and high volatility | Strength produced with real churn cost |
| RP | Iglesias 446.4, Bryan Baker 519.9 and Riley O’Brien 422.1 formed the main captured closer core | Excellent waiver-backed save production | Major strength |

Shortstop and third base were not imaginary frustrations: the drafted incumbents combined for only **72.8 captured points**. The important correction is that the team did not leave those positions wholly unproductive; it paid for replacement production with acquisition volume and management attention.

## 6. Transaction audit

Yahoo reports **117 moves**. Using exact final team-log totals, non-drafted players produced **7,972.7 captured points**, or **68.1 points per recorded move**. This is a portfolio metric, not a claim that each move generated 68.1 points; repeated adds and zero-point acquisitions prevent that interpretation. It also includes postseason/bye-week production, because Yahoo does not expose a regular-season-only player-source split.

The full chronological add/drop table is in Appendix A. A stint-level “points captured” figure is exact only when the player had one roster stint. Yahoo’s team log aggregates multiple stints, so repeated adds such as Zebby Matthews cannot be separated without a daily transaction/stat archive. Successful- and failed-move rates are therefore reported under an explicit definition in the appendix rather than inferred from aggregate player totals.

The ledger contains **149 transaction records: 117 adds, 116 drops, 84 paired add/drops, 33 add-only, and 32 drop-only**. There were **81 distinct added players**. August was the busiest month by ledger rows (37); April 25, May 27, June 23, July 17, September 20. The pattern was not purely indiscriminate: Riley O’Brien was held 160 days, Liam Hicks 167, Taj Bradley’s first stint 67, Bryce Miller 66, Hoerner 50, and Joey Cantillo’s later stint 48.

A move-level success/failure rate is **unavailable** because repeated stints cannot be allocated separately. As a descriptive lower-resolution substitute, 71 of 81 distinct added names finished with positive aggregate captured points for Corey across all their stints; 10 were zero or negative. That **87.7% distinct-player positive rate** must not be presented as an 87.7% move success rate, and it is partly contaminated by drafted players who were later reacquired.

### Five best transactions

1. **Drew Rasmussen** — 571.1 captured points across 19 appearances.
2. **Bryan Baker** — 519.9 across 53 appearances, including 34 saves.
3. **Taj Bradley** — 462.5 across 22 appearances.
4. **Riley O’Brien** — 422.1 across 56 appearances, including 34 saves.
5. **Logan Henderson** — 379.5 across 12 appearances.

### Five worst captured outcomes among used pitchers

1. **Gage Jump:** −21.1 across three appearances.
2. **Jackson Jobe:** −10.6 in one appearance.
3. **Cade Cavalli:** −10.1 across two appearances.
4. **Paul Sewald:** −7.5 across two appearances.
5. **Sean Manaea:** −5.0 in one appearance.

Negative results do not by themselves prove bad decisions. Contemporary matchup, role, health, and alternatives are evaluated in Appendix A.

### Five defensible processes with poor results

1. **Gage Jump (−21.1):** injury/rotation speculation with real upside, but role and workload uncertainty were large.
2. **Jackson Jobe (−10.6):** prospect/skills bet; one start is too noisy to condemn the process.
3. **Cade Cavalli (−10.1):** returning-injury upside was reasonable with four IL slots, though active-roster cost mattered.
4. **Sean Manaea (−5.0):** one-day probable-start stream; poor result can arise from a coherent schedule play.
5. **Stephen Kolek (−2.1):** role/IL speculation; defensible only if the expected activation and workload were documented.

**Net value above replacement is unavailable.** It requires the contemporaneous best available alternative and his eligible active-lineup points on the same dates. Full-season MLB points or post-drop production alone would not answer that counterfactual.

## 7. Starting-pitcher streaming and the Baseball Pitcher Foundry

An exact season-wide streaming sample is **unavailable** because Yahoo aggregates a player’s multiple roster stints and does not expose an export joining acquisition time, active slot and appearance. A defensible stream definition also requires tenure (for example, held seven days or fewer), which the aggregate team log cannot reconstruct. The championship subset is exact and separately reported below; the report does not silently extrapolate it to the full season.

### What the Foundry has actually demonstrated

The Foundry reports 106 graded nights and 2,655 starts, with season rank correlation **ρ = 0.27**, 80% interval coverage of **78%**, QS Brier score **0.23**, and an online edge of **+0.044 ρ** over a plain xFIP sort. Its 161-slate full backtest reports **+0.062 [0.044, 0.079]** versus xFIP, **+0.255 [0.223, 0.287]** versus opponent OPS, **+0.065 [0.051, 0.078]** versus K%, and **−0.016 [−0.032, 0.000]** versus recent fantasy points. Sources: [Foundry track record](https://casa.tailb77fc.ts.net/record), [Foundry engine room](https://casa.tailb77fc.ts.net/engine).

Interpretation: the engine adds a modest, credible ordering edge over simple skill/matchup shortcuts, but its ρ of 0.27 leaves most start-level variance unexplained. It is useful for ranking candidates and exposing uncertainty; it is not strong enough to justify ignoring transaction scarcity, role certainty, or roster context.

### Signals that survived or failed testing

| Signal | Foundry evidence | 2027 use |
|---|---|---|
| Recent “hot” performance | Kept in the June fitted model, but full backtest was level/slightly behind recent-points sort | Use as a secondary tiebreaker, not a veto |
| Opponent OPS alone | Engine decisively beat this shortcut | Do not rank primarily by opponent OPS |
| K% alone | Engine decisively beat this shortcut, though K% remains valuable in this scoring | Keep K% as a core component, not the whole model |
| xFIP alone | Engine beat it by a modest but positive margin | Baseline/guardrail, not final ranking |
| Home/road splits | Multiple trials timed out with no clear effect | Remove from headline reasons; show only as context |
| Long-rest “rust” | Timed out | Do not penalize automatically |
| Opponent recent form | Timed out | Do not overweight |
| Opposing starter ERA / win-support proxy | Repeated trials timed out | Treat win probability as low-confidence; do not chase motivation narratives |
| Hard-hit reweighting | One upward reweight hurt; other directions timed out | Keep fitted contact model, suppress manual overrides |
| Opener/bulk cap | Timed out | Require role confirmation instead of a generic cap |

### Foundry changes recommended

1. Display **expected points above the best available alternative**, not only raw projected points.
2. Add a **weekly-acquisition shadow price** that rises as the six-add budget is consumed.
3. Separate **role certainty** (confirmed starter, pitch-count cap, opener/bulk, return from IL) from performance uncertainty.
4. Add alerts for **probable-pitcher changes, postponement risk, weather, and late scratch** with timestamps.
5. Report calibration and realized value specifically for **manager-selected streams**, avoiding selection-bias confusion with the full slate.
6. Tag prospects/call-ups/returnees and publish cohort metrics; do not infer from a handful of memorable hits.
7. Add a two-start score that discounts the second start for schedule-change probability and acquisition opportunity cost.
8. In playoff mode, optimize **win probability and opponent denial value**, not season-long expected points alone.

## 8. Lineup management and injury-slot audit

An exact season total for points lost to benching, empty positions, postponed games, or missed starts is **unavailable**. Yahoo’s current team log tells us what counted but not the historical active/bench state for every date; the matchup totals page also warns that a player currently shown on the bench may have accumulated those points earlier while active. Treating current “BN” totals as bench points would be wrong.

One exact historical checkpoint was recovered. On April 1, active hitters scored 117.9 and active pitchers 80.8; Altuve (4.5) and Nimmo (7.8) were on the bench for 12.3 raw points. That is not automatically an error because all eligible hitter slots were filled and the pregame information set/eligibility counterfactual is unavailable. Ohtani (P) and McClanahan, both Yahoo-tagged IL, remained in active pitching slots, but no scoring bench pitcher was displaced that day, so the measured one-day cost was 0.0. The state is a process warning, not proof of lost points.

The four IL slots created meaningful option value, particularly for Ohtani (P), McClanahan, Christian Scott and other speculative/returning pitchers. At the final championship snapshot only one of four IL slots was occupied, which is not itself evidence of poor management: IL value depends on the availability of eligible, roster-worthy stashes. The detailed injury chronology and any identifiable delayed placements appear in Appendix A.

### Prevention checklist for 2027

- 10:00 a.m. ET: check IL activations, postponement/weather and probable-pitcher changes.
- 90 minutes before first pitch: confirm MLB lineups and move non-starters out where an eligible alternative exists.
- Before adding a streamer: confirm an active slot exists on the start date and record the drop contingency.
- Preserve a daily roster export/screenshot after lock; this makes later bench-loss auditing exact.
- Friday morning: count remaining team games by active hitter and remaining probable starts by pitcher.
- Never read weekly points beside a current BN label as “bench points” without the daily ledger.

## 9. Weekly performance and schedule luck

Regular-season scoring was high but not unusually stable: **796.07 mean**, **793.35 median**, **590.7 floor**, **1,026.0 ceiling**, and **111.59 population standard deviation**. Corey’s exact all-play record was **161–37 (.813)**, equivalent to **17.89 expected wins** across 22 weeks. The actual 16 wins were therefore **1.89 below** the all-play baseline. Opponents averaged **647.18**, slightly below the 652.71 league team-week mean and third-easiest in the league, but the timing was unfavorable: Corey went 16–5 above the weekly league average and 0–1 at/below it, with no low-score wins.

Using a transparent close-game definition of an absolute margin under 50 points, Corey went **1–3**: losses by 14.2, 21.7 and 15.6; win by 39.1. At 100 points the record was 2–4. Yahoo projected Corey to win 20 of 22 matchups; four projected-favorite losses show that the projections described roster strength but overstated week-level certainty. The clearest timing hit was Week 10: Corey posted the league’s second-highest score and lost only because the Gems were first.

## 10. Playoff and championship autopsy

The initial championship position was close to a coin flip by projection but unfavorable: the archived early-week Yahoo snapshot projected **805.55 for Corey and 816.36 for the Gems**. After the matchup closed, the page displayed a recomputed “original” projection of 787.10–797.47, still favoring the Gems by roughly 10 points. Early points mattered because both managers had a six-add ceiling and could convert expendable pitchers into later-week starts; an early lead forces the trailing manager to accept more downside and creates denial value in scarce Friday/Saturday starters.

Corey’s 16 active starts produced **404.2 points (25.26/start)** with an **18.7 median**, **16.05 population SD**, **15/16 positive**, **1/16 negative**, **7 quality starts**, **5 wins**, and **104 strikeouts (6.50/start)**. Home starts went 175.0 over six (29.17/start); road starts went 229.2 over ten (22.92/start). Jared Jones and Michael Soroka were the two-start pitchers: their four starts combined for **137.4 points (34.35/start)**, all positive. These splits are descriptive, not causal; the sample is small and pitcher/opponent quality is confounded with venue.

### The six-acquisition plan

| Add | Acquired | Captured championship points | Process read | Outcome read |
|---|---|---:|---|---|
| Zebby Matthews | Sep 21, 9:16 p.m. | 16.6 | Volume-first, volatile matchup/skill profile | Positive but noisy |
| Drew Anderson | Sep 22, 7:35 a.m. | 17.5 | Reasonable scheduled-start conversion | Positive |
| Connor Prielipp | Sep 22, 9:14 p.m. | 11.2 | Prospect/role uncertainty; acceptable at playoff leverage | Slightly positive |
| Trevor Rogers | Sep 23, 5:38 a.m. | 14.5 | Lower-ceiling floor play in a difficult context | Positive |
| Kade Anderson | Sep 24, 9:54 p.m. | 15.5 | Added before the opponent could; real denial value | Positive and strategically useful |
| Bryce Miller | Sep 25, 4:34 a.m. | 13.6 | Preferred over David Peterson; defensible but not clearly optimal ex ante | Positive outcome; choice still debatable |
| **Total** |  | **88.9** |  |  |

The six adds averaged **14.82 points/start**, median **15.0**, with **32 strikeouts, zero wins, one quality start, six positive and zero negative starts**. The narrow 2.06-point observed SD is an outcome fact, not proof the candidates were inherently low-risk. All six avoided a catastrophic negative score; with this scoring’s heavy ER and baserunner penalties, that is favorable variance.

The matchup-sized margin was driven more by favorable variance from pitchers already rostered than by the six adds. On Tuesday, Taj Bradley, Drew Rasmussen, Jared Jones and Michael Soroka scored **188.5** against contemporaneous Foundry projections totaling **77.2**, a **+111.3** swing above projection—larger than all six acquisitions’ 88.9 points.

### Sunday closeout

Corey entered Sunday ahead **788.4–619.9**. Active hitters added 40.0 points and the second starts from Jared Jones (24.4) and Michael Soroka (10.5) added 34.9, for a **74.9-point Sunday**. The Gems scored 81.1, led by Yusei Kikuchi’s 16.6-point stream and late offensive production, but gained only 6.2 points. The final margin was **162.3**. Sunday therefore validates the strategic value of the earlier cushion, not a claim that Corey won the final day.

### Friday’s six-versus-four volume

The opponent, not Corey, had six Friday starts versus Corey’s four. Corey’s starters scored **63.6** (15.90/start; one negative); the opponent’s scored **67.5** (11.25/start; median −0.65; three negatives). Including relievers, the opponent won Friday pitching only **76.5–74.6**. The opponent’s Eury Pérez and Reid Detmers combined for 98.2 while the other four starts totaled −30.7. More starts raised both expected volume and downside; it was not automatically a decisive advantage.

### Kade denial value and Bryce choice

Foundry projected Kade Anderson at 14.8 and the opponent’s eventual Saturday alternative Davis Martin at 11.4. If the managers would otherwise have swapped those arms, acting first created an estimated **6.8-point expected matchup allocation swing**. The realized swing was +21.7 (Kade 15.5, Davis −6.2), most of which was variance. Kade also recorded only 10 outs, contradicting the confident “normal-length final start” narrative; the short workload’s cause is unavailable.

For Friday, Foundry ranked Bryce Miller #11 at 16.2 (STREAM, raised by the LLM to START), David Peterson #19 at 13.3 (SPEC, raised to STREAM), and Brayan Bello #25 at 10.8 (PASS, raised to STREAM). Bryce’s process was defensible because of the better park/opponent and higher engine projection. Actual Yahoo-equivalent points were Bryce 13.6, Peterson 26.5 and Bello 11.5. Peterson would have added 12.9 realized points, but Bryce carried a 2.9-point ex-ante model edge; hindsight does not reverse the process grade. External line verification: [Bryce Miller box score](https://www.baseball-reference.com/boxes/SEA/SEA202609250.shtml), [David Peterson game log](https://www.cbssports.com/mlb/players/2835074/david-peterson/game-log/), [Brayan Bello box score](https://www.baseball-almanac.com/box-scores/boxscore.php?boxid=202609252BOS).

The Foundry should surface engine-versus-LLM disagreement and absolute deltas. Three simultaneous prose upgrades make the tiers sound more decisive than the numerical separation warrants.

### Counterfactuals

1. **Peterson over Bryce Miller.** This was the strongest simple alternative. Bryce’s actual 13.6 was useful, but the process comparison depends on contemporaneous projection, opponent, role/pitch count, and weather. The actual outcome does not settle the choice.
2. **Brayan Bello over the weakest-risk stream.** A plausible higher-floor alternative if his start and availability were confirmed. The opportunity cost is the start that would have been blocked and any denial value lost.
3. **Hold one add until Saturday/Sunday.** Lower early volume but preserves optionality for probable-pitcher changes, postponements, or a need to chase/defend. This is preferable when leading and forecast uncertainty is high.
4. **Max-volume opponent mirror.** Add any confirmed starter and accept tail risk. This raises mean points but can reduce win probability when already ahead because negative starts are costly.

The strongest realistic alternative was Peterson over Bryce, worth 12.9 more realized points but still leaving Corey far ahead. Holding a move for Sunday would have added flexibility, yet the actual final state shows it was not necessary: both two-start incumbents were positive and the opponent’s Sunday gain was small. The six-add plan could still have failed through a postponement, surprise workload cap, or one or two heavily negative early starts; the championship result should not erase those ex-ante risks.

## 11. Opponent benchmark: Sane Moderate Person’s Gems

The Gems finished the regular season **21–1**, first with **18,604.5 points**, 1,091.0 more than Corey, and used **120 moves**. The manager entered with a Diamond rating and a long history (30 teams managed; five first-place trophies), reached the championship after a **924.3–825.9** semifinal, and lost the final 701.0–863.3. Source: [opponent team](https://baseball.fantasysports.yahoo.com/b1/203711/7), [standings](https://baseball.fantasysports.yahoo.com/b1/203711?module=standings&lhst=stand#lhststand).

The roster’s structural lesson is not merely “stream more.” Its draft began Judge, Junior Caminero, Tatis, Matt Olson, Chris Sale, Cody Bellinger, Jackson Merrill, Tyler Soderstrom, CJ Abrams, Nico Hoerner and Eury Pérez—a blend of elite batting, scarce infield positions, and later pitching. The manager also cycled pitchers aggressively during the final week. Copy the **combination of positional stability, lineup discipline, and tactical pitching volume**; do not copy every churn decision or assume a 21–1 record can be reproduced without favorable weekly timing.

In the final, the Gems kept chasing volume after falling behind. Their six Friday starters produced only 67.5 points because four combined for −30.7, and Sunday’s added volume helped them win the day 81.1–74.9 without materially reducing the deficit. The opponent’s approach was rational for a trailing team, but its failure illustrates that maximum starts and maximum championship win probability are not the same objective.

The 21–1 record was deserved but somewhat flattered. The Gems’ all-play record was **177–21 (.894)**, or 19.67 expected wins—1.33 fewer than actual—and their opponents averaged a league-low 631.48. A single 1,549.4-point Week 17 inflated the total-points gap; excluding it, the Gems averaged 812.15 versus Corey’s 796.07, while the median gap was only 12.15 points (805.5–793.35).

## 12. Manager profile

The evidence describes a **high-aggression, high-information, high-variance manager**:

- **Aggressiveness:** 117 moves, full use of the six-add championship budget.
- **Patience:** mixed. Core bats were held through normal fluctuation; back-end pitchers and infield replacements moved quickly.
- **Recency bias:** present as a risk, but the Foundry’s own tests do not show that recent-form chasing consistently added value. The manager built a model partly to discipline this impulse.
- **Risk tolerance:** high, especially with prospects, injured returnees and volatile starters.
- **Attachment to drafted players:** not excessive at the roster fringe; several drafted misses were discarded or de-emphasized.
- **Response to uncertainty:** active and creative, sometimes converting uncertainty into too many small bets.
- **Playoff pressure:** good. The six-add plan was coherent, fast, and opponent-aware.

Intuition added value when it recognized roster scarcity and acted early (notably Kade). It created risk when narrative signals—recent form, motivation, opposing starter quality—were allowed to sound more predictive than the Foundry’s own trials support.

## 13. Ten decisions requiring the most scrutiny

| Rank | Decision | Contemporary alternative | Actual result | Lesson |
|---:|---|---|---|---|
| 1 | Build around a batter-heavy core before deep SP volume | Draft more early aces | Second in points; deep batting core | Correct for this format when paired with active SP replacement |
| 2 | Draft Sánchez at 24 | Take a hitter/other ace | 870.4 captured | Small ADP reach was justified by scoring fit |
| 3 | Draft Yordan at 37 | Wait toward ADP 40 | 1,532.3 | Reach that worked; judge by pre-draft projection gap, not result alone |
| 4 | Draft Perdomo/Bregman as SS/3B plan | Prioritize scarce infield earlier | 72.8 combined | Positional fragility imposed season-long transaction cost |
| 5 | Draft Ohtani (P) at 104 | Healthy SP/RP | 441.9 | Correctly priced asymmetric upside |
| 6 | Add/hold Rasmussen | Continue pure streaming | 571.1 | Promote a good stream into a hold when skills/role justify it |
| 7 | Build closer value through Baker/O’Brien | Pay more draft capital for saves | 942.0 combined | Strong repeatable waiver skill, though roles are volatile |
| 8 | Use all six championship adds early | Preserve late optionality | 88.9 from named six | Coherent and successful; still exposed to schedule changes |
| 9 | Secure Kade before opponent | Wait for more confirmation | 15.5 plus denial value | Early blocking is valid only when replacement gap exceeds flexibility cost |
| 10 | Bryce Miller over Peterson | Add Peterson / hold add | Bryce 13.6 | Positive outcome does not prove best process; retain the counterfactual |

## 14. 2027 playbook

### Draft rules

1. By the end of Round 8, roster at least **two stable MI/3B-eligible bats**, unless a clearly superior value falls.
2. Draft **one SP by Round 4 and two by Round 8** only when they project at least 3 points/start above the replacement-stream baseline; otherwise keep taking bats.
3. Draft no more than **two** players whose Opening Day workload/health is materially uncertain, excluding late-round IL stashes.
4. Pay for one reliable closer only if the role is at least 80% secure; reserve the second/third RP jobs for waivers.
5. Treat multi-eligibility as worth roughly one bench spot only when the player also clears replacement-level projection.

### In-season and streaming rules

1. Define a stream as a pitcher expected to be held **7 days or fewer**; track each stint separately.
2. Add only when projected points exceed the best free alternative by **5+ points**, or by **3+** when the move also blocks an opponent in the playoffs.
3. Require three green lights: confirmed start/role, projected pitch count at least 75 (or credible bulk role), and no high postponement risk.
4. Use two-start pitchers only when each start independently projects above replacement after a 15% discount to the second start.
5. Do not add solely for “playoff motivation.” Motivation may change leash; it is not a standalone run-prevention skill.
6. Promote a streamer to a hold after two starts only if underlying K-BB%, velocity/arsenal and workload all support it; ignore fantasy points alone.
7. Budget adds: **2 by Tuesday, 2 for Wednesday–Friday, 2 reserved for weekend**, unless early-week denial value or a must-win deficit justifies acceleration.
8. When leading in playoffs, optimize downside-adjusted win probability: avoid starts with a bottom-decile negative tail even if their mean is slightly higher.
9. Add IL/prospect stashes only when the expected active date is within 21 days and the upside clears the best active bench alternative by 75+ rest-of-season points.
10. Store a daily roster snapshot and the Foundry projection used for every add.

### Foundry operating checklist

- Use the engine rank as a shortlist, not an automatic add order.
- Display median, p10, p90, role certainty, acquisition cost, and opponent-denial value together.
- Ignore home/road and long-rest stories unless new out-of-sample evidence changes the result.
- Give recent form no more than tiebreaker weight.
- Refit on fantasy points only after preserving an untouched time-based test set.
- Report manager-selected-stream ROI separately from full-slate rank accuracy.

## 15. Final verdict

| Dimension | Grade | Confidence |
|---|:---:|:---:|
| Overall manager | A− | High on result; medium-high on process attribution |
| Draft | A− | High on outcome; medium on process context |
| In-season management | B+ | Medium |
| Streaming | B+ | Medium-low; season-wide stint join unavailable |
| Lineup management | B | Low; historical active/bench ledger unavailable |
| Playoffs | A | High on outcome; medium on counterfactual optimality |

**Skill versus luck:** the second-place regular-season finish and championship were primarily skill-supported: second-most regular-season points, a strong drafted core, substantial waiver production, a 161–37 all-play record, and a coherent playoff-volume plan. The 16–6 record contains schedule/close-game variance, while the title run benefited from several volatile pitchers landing on the positive side. The correct synthesis is **repeatable top-tier process with meaningful favorable playoff variance**, not either “all skill” or “all luck.”

### Ten lessons to retain

1. Draft points-format bats with walks, power and lineup stability.
2. A small, evidence-based ADP reach can be correct.
3. Promote successful streamers into holds when skills support it.
4. Mine saves rather than paying for every RP role.
5. Multi-eligibility is most valuable at weak infield spots.
6. Early playoff points create strategic leverage.
7. Blocking an opponent has real but replacement-gap-limited value.
8. Projection ranges matter more than point estimates for volatile SPs.
9. Keep process and outcome grades separate.
10. Preserve daily data so next year’s audit can be exact.

### Ten changes for next season

1. Solve SS/3B earlier.
2. Cap simultaneous injury/workload bets.
3. Track every acquisition as a separate stint.
4. Reserve weekend moves by default.
5. Price the six-add constraint in the Foundry.
6. Add role/weather alerts.
7. Stop elevating home/road, motivation and opposing-starter narratives.
8. Use downside thresholds when protecting a playoff lead.
9. Record best available alternative at decision time.
10. Archive daily active/bench rosters and projections.

---

# Appendices

## Appendix A — Transaction table

The complete 149-row Yahoo ledger—every add and drop, exact date/time, pair, transaction type, representative hold lengths, injury notes and position-churn analysis—is preserved in the companion [transaction, lineup and injury evidence appendix](/Users/corey/Projects/fantasy-baseball-2026-audit/agent2_transactions_lineups_injuries.md).

| Ledger measure | Exact value |
|---|---:|
| Transaction records | 149 |
| Add events | 117 |
| Drop events | 116 |
| Paired add/drop records | 84 |
| Add-only records | 33 |
| Drop-only records | 32 |
| Distinct added players | 81 |

Yahoo’s aggregate team log does not split repeated stints. Therefore active appearances, captured/bench points, post-drop points, best alternative, and outcome grade are marked unavailable at the **stint** level except where the championship’s daily reconstruction supplies them. The companion table does not fabricate those fields.

## Appendix B — Draft table

See Section 4. Overall pick numbers are derived from the 10-team snake draft board. Source: [Yahoo draft results](https://baseball.fantasysports.yahoo.com/b1/203711/draftresults).

## Appendix C — Weekly results

| Wk | Opponent | PF | PA | Orig proj (C–O) | Proj | Actual | Margin | Lg avg | Rank | All-play |
|---:|---|---:|---:|---:|:---:|:---:|---:|---:|---:|---:|
| 2 | NAPOLI | 761.3 | 690.9 | 731.33–694.58 | W | W | +70.4 | 630.70 | 2 | 8–1 |
| 3 | Hot Corner Mito | 762.3 | 600.5 | 746.90–669.21 | W | W | +161.8 | 641.43 | 3 | 7–2 |
| 4 | Speedy Brian K | 980.9 | 646.5 | 856.89–666.32 | W | W | +334.4 | 669.46 | 1 | 9–0 |
| 5 | arthur’s Destroyers | 590.7 | 652.8 | 761.26–651.85 | W | L | −62.1 | 628.40 | 7 | 3–6 |
| 6 | Line Drive Ramon | 706.6 | 813.9 | 713.84–699.87 | W | L | −107.3 | 641.78 | 4 | 6–3 |
| 7 | Stealing kenneth hopkins | 801.7 | 617.8 | 784.46–687.05 | W | W | +183.9 | 635.66 | 2 | 8–1 |
| 8 | Ashwin’s Bats | 802.3 | 389.8 | 789.27–421.49 | W | W | +412.5 | 630.42 | 2 | 8–1 |
| 9 | ⚾️Young Guns 🔫 | 886.8 | 738.9 | 770.94–757.31 | W | W | +147.9 | 623.24 | 1 | 9–0 |
| 10 | Sane Moderate Person’s Gems | 895.2 | 909.4 | 871.67–892.26 | L | L | −14.2 | 656.93 | 2 | 8–1 |
| 11 | NAPOLI | 942.2 | 694.4 | 832.62–718.08 | W | W | +247.8 | 632.97 | 1 | 9–0 |
| 12 | Hot Corner Mito | 797.4 | 568.2 | 771.07–571.00 | W | W | +229.2 | 649.44 | 2 | 8–1 |
| 13 | Speedy Brian K | 657.5 | 679.2 | 816.93–588.52 | W | L | −21.7 | 627.56 | 6 | 4–5 |
| 14 | arthur’s Destroyers | 777.5 | 441.6 | 799.19–531.20 | W | W | +335.9 | 667.80 | 2 | 8–1 |
| 15 | Line Drive Ramon | 789.3 | 684.4 | 873.96–733.46 | W | W | +104.9 | 662.69 | 3 | 7–2 |
| 16 | Stealing kenneth hopkins | 826.1 | 787.0 | 905.29–765.85 | W | W | +39.1 | 630.82 | 1 | 9–0 |
| 17 | Ashwin’s Bats | 1,026.0 | 668.5 | 1,140.64–670.41 | W | W | +357.5 | 947.17 | 3 | 7–2 |
| 18 | ⚾️Young Guns 🔫 | 935.1 | 650.3 | 780.32–700.82 | W | W | +284.8 | 657.24 | 1 | 9–0 |
| 19 | Sane Moderate Person’s Gems | 689.4 | 986.9 | 773.06–904.45 | L | L | −297.5 | 642.59 | 4 | 6–3 |
| 20 | NAPOLI | 670.3 | 685.9 | 787.77–710.44 | W | L | −15.6 | 630.31 | 5 | 5–4 |
| 21 | Hot Corner Mito | 646.0 | 486.0 | 805.32–635.42 | W | W | +160.0 | 621.81 | 5 | 5–4 |
| 22 | Speedy Brian K | 734.0 | 464.4 | 814.68–726.09 | W | W | +269.6 | 577.88 | 1 | 9–0 |
| 23 | arthur’s Destroyers | 834.9 | 380.7 | 859.38–448.95 | W | W | +454.2 | 653.27 | 1 | 9–0 |

### Playoff matchups

| Wk | Round | Opponent | PF | PA | Archived orig proj (C–O) | Proj | Actual | Margin |
|---:|---|---|---:|---:|---:|:---:|:---:|---:|
| 24 | Bye | — | 473.7 team-log points | — | — | — | Bye | — |
| 25 | Semifinal | ⚾️Young Guns 🔫 | 896.4 | 670.1 | 865.94–842.98 | W | W | +226.3 |
| 26 | Championship | Sane Moderate Person’s Gems | **863.3** | **701.0** | 805.55–816.36 | L | **W** | **+162.3** |

Yahoo’s original projections were recovered from the historical matchup pages; they are point projections, not calibrated win probabilities. The closed Week 26 page later displayed a recomputed 787.10–797.47 “original” projection, so the archived early-week figure is retained as the genuine decision-time snapshot. Sources: [Yahoo schedule module](https://baseball.fantasysports.yahoo.com/b1/203711/?module=standings&lhst=sched&sctype=team&scmid=5) and the corresponding matchup pages.

### All-play and schedule-luck summary

| Metric | Corey | Gems benchmark |
|---|---:|---:|
| All-play record | 161–37 | 177–21 |
| All-play pct | .813 | .894 |
| Expected wins | 17.89 | 19.67 |
| Actual wins | 16 | 21 |
| Actual minus expected | −1.89 | +1.33 |
| Average opponent score | 647.18 | 631.48 |

## Appendix D — Championship timeline

| Date/time ET | Corey action | Roster consequence |
|---|---|---|
| Sep 21, 9:16 p.m. | Added Zebby Matthews; dropped Ryan Johnson | First championship add |
| Sep 22, 7:35 a.m. | Added Drew Anderson; dropped Zebby Matthews | Converted one start into the next |
| Sep 22, 9:14 p.m. | Added Connor Prielipp; dropped Christian Scott | Prospect/start bet |
| Sep 23, 5:38 a.m. | Added Trevor Rogers; dropped Drew Anderson | Continued cycling |
| Sep 24, 9:10 a.m. | Dropped Connor Prielipp | Opened roster spot |
| Sep 24, 9:54 p.m. | Added Kade Anderson | Secured/blocked scarce starter |
| Sep 24, 10:41 p.m. | Dropped Taj Bradley | Opened final-add path |
| Sep 25, 4:34 a.m. | Added Bryce Miller | Sixth and final add |

Source: [Yahoo transactions, team 5](https://baseball.fantasysports.yahoo.com/b1/203711/transactions?transactionsfilter=all&mid=5).

### Final daily scoring

| Day | Batting | Pitching | Daily | Cumulative |
|---|---:|---:|---:|---:|
| Mon | 0.0 | 16.6 | 16.6 | 16.6 |
| Tue | 78.6 | 209.4 | 288.0 | 304.6 |
| Wed | 84.7 | 43.1 | 127.8 | 432.4 |
| Thu | 71.1 | 36.5 | 107.6 | 540.0 |
| Fri | 44.2 | 74.6 | 118.8 | 658.8 |
| Sat | 112.4 | 17.2 | 129.6 | 788.4 |
| Sun | 40.0 | 34.9 | 74.9 | **863.3** |

### Final active starters

| Day | Corey active starters and points |
|---|---|
| Mon | Zebby Matthews 16.6 |
| Tue | Drew Anderson 17.5; Taj Bradley 33.2; Drew Rasmussen 52.8; Jared Jones 54.7; Michael Soroka 47.8 |
| Wed | Logan Henderson 19.9; Connor Prielipp 11.2 |
| Thu | Tyler Glasnow 36.5 |
| Fri | Gavin Williams −1.7; Cristopher Sánchez 37.2; Bryce Miller 13.6; Trevor Rogers 14.5 |
| Sat | Kade Anderson 15.5 |
| Sun | Jared Jones 24.4; Michael Soroka 10.5 |

## Appendix E — Methodology and limitations

1. Captured points were calculated from Yahoo’s team-log category totals and the exact league scoring formula. Example: pitcher points = 8W + 8SV + OUT − 1.3H − 3ER − 1.3BB − 1.3HBP + 3K.
2. Regular-season weekly totals sum exactly to 17,513.5, matching Yahoo’s standings.
3. “Drafted” means selected by Corey; production remains assigned to the draft source even if the player was later dropped and re-added.
4. “Non-drafted” is a source-of-points split, not necessarily points above replacement.
5. Yahoo’s player-level team log spans Weeks 2–26 even though its headline total remains the regular-season standings total. The player rows sum to **19,746.9**: 17,513.5 regular season + 473.7 Week 24 bye production + 896.4 semifinal + 863.3 championship.
6. Yahoo’s team log aggregates multiple roster stints and does not preserve exact active lineup slot by date in the current summary. Exact stint points, bench losses and position-slot totals are unavailable without daily archived rosters.
7. The championship had 0 games remaining at the final capture. As with any Yahoo result, ordinary postgame stat corrections could still alter tenths of a point; the 162.3-point margin makes reversal implausible.
8. The Foundry’s backtest and online records are self-reported by the application. This audit evaluates the reported design and metrics but did not independently reproduce the model from raw training data.
9. Confidence is highest for settings, standings, scores, draft order and captured category totals; medium for process grades; low where contemporary news/projection snapshots or daily lineup states are missing.

## Appendix F — Source index

- [League home / playoff bracket](https://baseball.fantasysports.yahoo.com/b1/203711)
- [Corey’s team](https://baseball.fantasysports.yahoo.com/b1/203711/5)
- [Corey team log](https://baseball.fantasysports.yahoo.com/b1/203711/5/teamlog)
- [Final regular-season standings](https://baseball.fantasysports.yahoo.com/b1/203711?module=standings&lhst=stand#lhststand)
- [League settings and scoring](https://baseball.fantasysports.yahoo.com/b1/203711/settings)
- [Draft results](https://baseball.fantasysports.yahoo.com/b1/203711/draftresults)
- [Transactions](https://baseball.fantasysports.yahoo.com/b1/203711/transactions?transactionsfilter=all&mid=5)
- [Championship matchup](https://baseball.fantasysports.yahoo.com/b1/203711/matchup?week=26&mid1=5&mid2=7)
- [Opponent team](https://baseball.fantasysports.yahoo.com/b1/203711/7)
- [Baseball Pitcher Foundry](https://casa.tailb77fc.ts.net/)
- [Foundry track record](https://casa.tailb77fc.ts.net/record)
- [Foundry engine room](https://casa.tailb77fc.ts.net/engine)

## Appendix G — Adversarial review

### What conclusions are least certain?

Lineup-management grade, exact position-slot production, stint-level acquisition efficiency, and the causal effect of individual championship choices. These are constrained by missing historical daily states and the absence of true contemporaneous counterfactual lineups.

### What evidence contradicts the main narrative?

Aggressive streaming was not uniformly successful: several acquisitions scored below zero, and the Foundry’s rank correlation is only 0.27. The opponent used slightly more moves and still scored 1,091 more regular-season points, so activity alone is not the advantage. Shortstop/third-base instability also shows that the strong overall finish masked real construction errors.

### Which apparent mistakes were reasonable at the time?

Bregman, Perdomo, Altuve, Sasaki, McClanahan and several high-upside pitching streams. Injuries, role changes and start-level variance can produce bad outcomes from defensible priors.

### Which apparent successes were probably luck?

The fact that all six named championship acquisitions avoided a major negative outing; the 188.5-point Tuesday from four already-rostered starters; the exact timing of Kade remaining available; and any win/save awarded behind otherwise similar pitching lines. Yordan/Lowe-level draft profit contains real selection skill but also outcome variance far beyond the draft-day projection gap.

### What would a highly skilled opposing manager criticize?

Too much churn without a recorded counterfactual, insufficient early investment in SS/3B, consuming all championship adds before late-week uncertainty resolved, and treating self-reported model backtests as stronger evidence than independent validation.

### Which recommendations change under different settings?

If strikeouts or outs were worth less, streaming volume would fall. If losses, HR allowed or negative starts were penalized more, downside thresholds would tighten. Weekly lineups would greatly reduce daily churn and elevate two-start scheduling. Fewer IL slots would reduce injured-player speculation. A lower add cap would increase the shadow price of every move. Category leagues would require balance rather than total-point maximization.