Final Conclusions / Postmortem
Page Header / Scope Note
This page is the repo-only terminal synthesis layer for the finished 2026 tournament. It summarizes what the completed Forecast stack supports at tournament close, links back to the men and women Forecast pages for dossier-level detail, and keeps any later external-enrichment page out of scope in this pass.
Primary detail surfaces: Men’s Forecast | Women’s Forecast | Portfolio | Men’s Reports | Women’s Reports
Final Bottom Line
Men
Supported
The single-entry anchor missed the title, but the diversified FM-01 core still covered the actual champion. Michigan-title sleeves finished 1st through 4th, while every Duke-title sleeve finished below them.
Women
Supported
The repo's women book stayed inside the real title tier, but the anchor was too concentrated on UConn. UCLA-title sleeves finished 1st and 2nd after an otherwise strongly top-heavy tournament.
Final Portfolio Outcome
| Track | Submission-facing anchor | Final result | Winning sleeve read | Evidence status |
|---|---|---|---|---|
| Men | E01 / FM-01 / BP-01 / Duke |
8 of 10, 970 points, 43/63 correct picks |
E05-E08 (FM-01 diversified core, Michigan champion) finished 1-4; best score 1320 |
Supported |
| Women | W-E01 / W-LEG-FM-01 / W-BP-01 / UConn |
4 of 10, 1200 points, 50/63 correct picks |
W-E07 and W-E08 (women diversified core, UCLA champion) finished 1-2; best score 1510 |
Supported |
Portfolio concentration at submission remained explicit:
- Men: the saved ten-entry book split champion exposure
Duke:4,Michigan:4,Arizona:1,Illinois:1. Supported - Women: the saved ten-entry book split champion exposure
UConn:4,Texas:2,UCLA:2,LSU:1,South Carolina:1. Supported - Men: the primary single-entry answer
BP-01and the deterministic chalk anchor both ended on Duke, so the winning outcome came from diversification rather than the anchor. Supported - Women: the primary single-entry answer
W-BP-01ended on UConn, so the winning outcome again came from diversification rather than the anchor. Supported
Tournament Morphology and Calibration Read
| Track | Repo-local pre-tournament calibration read | Realized tournament read from saved realized-results files | Evidence status |
|---|---|---|---|
| Men | BP-01 was labeled structurally credible but chalk-leaning; top-4 champion share 0.8158 and R64 upset rate 0.1418 in saved calibration summaries |
Realized upset counts were 9/32 in R64, 4/16 in R32, 3/8 in S16, then 1/4, 0/2, 0/1 from E8 through the title game. That reads as mixed rather than purely chalk: early volatility, but a top-tier title outcome with Michigan. |
Supported |
| Women | Saved scenario summary called the women bracket scenario_supported_top_heavy; top-4 champion share 0.9304 |
Realized upset counts were 3/32 in R64, 3/16 in R32, 2/8 in S16, then 0/4, 0/2, 0/1 from E8 through the title game. All four No. 1 seeds reached the Final Four and UCLA won the title. |
Supported |
| Men | How far the realized field was from the full saved simulation distribution beyond these summary objects | Partial, because the repo has strong round/concentration references but no dedicated final realized-vs-simulation distance report | Partial |
| Women | Same deeper realized-vs-simulation distance question | Partial for the same reason | Partial |
Model Knowledge and Resolved Disagreement Read
| Track | Repo-supported read | Evidence status |
|---|---|---|
| Men | The saved pre-live disagreement layer showed meaningful FM-01 versus FM-03 differences, but the live tracking artifact only covers the first four completed First Four games. Inside that saved window, FM-03 correctly flipped Prairie View A&M over Lehigh where FM-01 missed, while both models missed Texas over NC State and Miami (OH) over SMU. |
Partial |
| Women | The saved disagreement layer is narrower: one explicit W-LEG-FM-01 versus W-FM-03 disagreement card (Samford vs Southern). The live tracking artifact shows the primary path was correct on that game while the challenger was wrong. |
Supported |
| Men and women | Stronger statements about full-tournament challenger win rates, systematic blind spots, or model causal failure modes are not supported by a repo-local final resolved-disagreement rollup. | Unresolved |
Consequential Tournament Turns
| Track | Turn | Why it mattered for the saved portfolio | Evidence status |
|---|---|---|---|
| Men | East E8: UConn 73, Duke 72 on 2026-03-29 |
This ended every Duke-title sleeve. The four Duke-champion entries (E01-E04) finished 5-8 rather than contending for the top four. |
Supported |
| Men | National title game: Michigan 69, UConn 63 on 2026-04-06 |
This locked the four Michigan-title diversified sleeves (E05-E08) into ranks 1-4. |
Supported |
| Women | National semifinal: South Carolina 62, UConn 48 on 2026-04-03 |
This ended the anchor/UConn title branch that still dominated the women portfolio by entry count (W-E01-W-E04). |
Supported |
| Women | National title game: UCLA 79, South Carolina 51 on 2026-04-05 |
This elevated the two UCLA-title diversified sleeves (W-E07, W-E08) to ranks 1 and 2. |
Supported |
| Men and women | A complete ranked game-by-game leverage table for the full tournament | The repo keeps live-state, exposure, and round EV objects, but not one final compiled leverage ranking artifact that spans every game end-to-end. | Partial |
Evidence Boundary at Tournament Close
| Claim class | Men | Women |
|---|---|---|
| Final portfolio placement, scores, and correct-pick totals | Supported | Supported |
| Submission concentration and which sleeves covered the actual champion | Supported | Supported |
| Tournament morphology from realized upset counts plus saved calibration summaries | Supported | Supported |
| Narrow resolved-disagreement read from saved artifacts | Partial | Supported |
| Full final game-by-game leverage ranking | Partial | Partial |
| Off-repo narrative explanations, betting-market expectedness, and popularity reads | Unresolved | Unresolved |
What Remains Outside Repo-Only Support
- Betting-market expectedness judgments, consensus pick-popularity comparisons, and richer public contest context remain outside this page. Unresolved
- Injury, lineup, broadcast, and game-script storytelling remains outside this page unless a later bounded enrichment pass adds off-repo support explicitly. Unresolved
- This page does not implement the dedicated external-enrichment surface. That later task remains separate on purpose so repo-only claims stay auditable. Supported
Links Back to Forecast Pages and Underlying Detailed Surfaces
- Men’s Forecast: championship-complete dossier stack, matrix, lower support band, and cumulative ledgers.
- Women’s Forecast: championship-complete dossier stack, matrix, lower support band, and cumulative ledgers.
- Portfolio: public home for whole-portfolio architecture, concentration/diversification read, and links into track-level portfolio detail.
- Men’s Reports: model, forecast, and deep-dive report wrappers.
- Women’s Reports: model, forecast, and deep-dive report wrappers.
- Repo-local evidence families synthesized here:
men_live_state_entry_ranking_championship.csv,womens_live_state_entry_ranking_championship.csv,scenario_calibration_summary.csv,womens_scenario_calibration_summary.csv, the men/women realized-results files fromR64throughChampionship, and the saved disagreement/live-tracking artifacts already used by the Forecast layer.
This page is intentionally narrower than the Forecast pages. It closes the tournament at the synthesis layer and leaves the full dossier, matrix, and ledger architecture in the Forecast stack.