League History audit package

1. What League History is today

Missing everywhere: an owner page, head-to-head game lists, records with sources, and any per-season confidence. The only flag today is one yellow banner on every past year.

2. Proposed layout (information architecture)

History becomes one area with seven sub-views, the same on desktop and phone. The default is Seasons.

1. Seasons, one page per season:
   • the UPS champion and runner-up, and the Hawk Tuah Bowl champion and finalist
   • final standings 1–12, with seed, division title, record, All-Play, Points For and the rookie pick it earned
   • both brackets with real scores
   • a status chip on every fact, and a "How we know" panel listing the sources and every discrepancy
   It replaces today's Playoff Bracket, Draft Order Finish and Historical Finishes.
2. Champions: the Hall of Champions list. IDX is renamed "Scoring vs league" (section 7). The stories stay behind the review banner until checked.
3. Owners: one page per person: titles, finals, playoffs, division titles, Hawk Tuah titles and every season. The identity questions (7a–7e) show as Provisional until you rule.
4. Head to head: owner v owner. Records are shown for every season; games are listed only where MFL kept them (2012 and 2020 on).
5. Records: league records, each with its season, its source and a status. A record built on a Provisional season says so.
6. Divisions: names and members each season, division titles, and division strength over equal-week samples (section 8). It replaces Division Power Rankings and absorbs Eras.
7. Draft order: each season's finish → pick table. It's on desktop only; the phone shows the pick on Seasons.

The rule on screen: every fact carries its status. Facts with a Conflict or Unknown status are shown as such, never as a plain number. Stories and records only use Verified or Reconciled facts (section 9).

Desktop: History → Seasons (wireframe; chips: V Verified, R Reconciled).

Desktop: History → Owners.

Phone: League → History → Seasons.

3. Source hierarchy

1. MFL's archived exports for that season's own league: playoff brackets and placement games, leagueStandings records, divisions and names, and the schedule and weekly results where they were kept.
   • This is the authority for what happened.
   • League ids: 2010 60671 · 2011 40832 · 2012 37227 · 2013 42721 · 2014 30590 · 2015 29015 · 2016 27191 · 2017 on 74598.
2. Canon (docs/league_context_v1.md): the authority for rules, such as finish → pick (§A1) and seeding (§F.1). Today's rules are never applied backward to a finished season.
3. Your approved rulings: the corrections overlay (canon §E). This is the only way a judgment call or identity decision changes a stored value.
4. D1 src_* tables: the stored copy the screens serve. They're checked against tier 1 and never edited directly.
5. The pick-trade replay and the legacy local database: corroboration only.
6. Stories, the forum, Discord and recollection: never a source for a fact, and they can't raise a status.

4. Season-by-season confidence audit, 2010–2025

Tap a season for each fact's status, what disagrees and the sources.

5. Playoff-matchup data by season

Every season keeps its playoff games with teams and scores in MFL's brackets and in D1, and they match.

• 2025: D1's game rows kept three Week 17 scores from before MFL's stat corrections. The winners are unchanged.
• Regular-season games: listed only for 2012 and 2020 on. The other seasons keep only MFL's official records.

6. Genuine pregame projections

2026, Weeks 1–6: ups_player_projections (MFL projectedScores).

• What counts: a row is genuine pregame only if its capture time is before that player's own NFL kickoff (MFL nflSchedule).
• First capture: the table keeps each player's first capture (first_projected, first_captured_at), and that's pregame for nearly everyone.
• Latest value: it's overwritten by the Wednesday re-capture, so it's post-game for about 95%.

• Why 70–85 rows a week are late: the first capture runs on Friday, after Thursday night's kickoff, so those players have no pregame number. The fix is a Thursday pre-kickoff capture, a Wire automation change for a separate PR.
• Week 1 was first captured Aug 2: pregame but stale.
• Frozen evidence files (site/wire/data/projection_evidence_2026_wkNN.json) exist for Weeks 1 and 3–6, not Week 2.
• Team-level pregame forecasts exist as the Wire's week previews and season simulations:
  • Week 2: generated 2026-09-17 00:01Z
  • Week 4: 09-30 12:22Z
  • Week 6: 10-14 01:56Z
  • Week 7: 10-21 14:02Z

Recommendation: History and the Wire call a projection "pregame" only from a capture taken before that player's kickoff, and never from the latest value. Weeks without one say "no pregame projection kept".

7. IDX: plain English and a recommendation

What's wrong:

• "IDX" means nothing to an owner.
• 2011: the stored Points For includes the playoffs, and the champion played the most playoff games. That inflates 2011 to 1.179, ranked #1; on regular-season scores it's 1.130, #6.
• 2010: rests on reconstructed scores.
• 2012–2025: already exact.

Recommendation:

• Rename it "Scoring vs league" and show it as "+16%" with a one-line note.
• Compute it from regular-season weekly scores, not the stored Points For.
• Mark 2010 Provisional.
• Fixing the 2011 Points For (correction C10) makes the stored number agree.

8. Maddie Mondays, validated on equal weeks

• All-Play, Weeks 1–6: Verified
  • 67–131 (.338), the lowest of all 64.
  • But only by one All-Play win over 2018 MAKE IT RAIN (68–130). A single stat correction could tie it. Say "lowest through six weeks", not "worst ever".
• Scoring vs league: 0.918, #3 lowest.
• Head to head against other divisions: 24–39 (.381), #5 worst of 32.
  • Only 2012 and 2020 on can be compared, because only those seasons list their games.
  • 2026 plays more interdivisional games, so compare rates, not counts.
• Division history: "Maddie Mondays" is new in 2026. It's division 02, which was MAKE IT RAIN 2014–2025 and Raytorious L52 2011–2013.
  • Members were re-drawn every cycle, and 2026 was owner-drafted.
  • So no lineage claim by division id holds. Compare division-seasons, as here.

Lowest All-Play through Week 6

Worst head to head against other divisions through Week 6

Division names and stored membership, by season

9. Proposed shared UPS Wire story-lead contract

The problem today: the Wire computes history its own way, straight from D1 (H2H series, record book, Survival line), and Standings does it another way. Nothing marks a fact's confidence, and a lead can't say which season it left out.

Proposal: one module computes historical facts once, and both the Wire and History read them.

• The module is pipelines/etl/history/history_facts.py.
• Every week it writes site/wire/data/history_leads_<season>_wkNN.json.
• Wire: the UPS Center's story box lists them as suggestions.
• History: the Records and Divisions views show the same facts.

{
  "schema": "ups.history_lead.v1",
  "lead_id": "division-allplay-low-thru-wk06-2026-02",
  "kind": "division_through_week",          // also: first_since, streak, h2h_series, record, season_compare
  "subject": {"season": 2026, "division_id": "02", "owners": ["Ryan Bousquet", "Josh Martel", "Keith Creelman"]},
  "sample": {"weeks": [1, 6], "basis": "regular season", "through_week": 6},
  "population": {"what": "three-team division-seasons", "seasons": [2011, 2026], "n": 64,
                 "excluded": [{"season": 2010, "why": "two six-team divisions"}]},
  "facts": [
    {"key": "allplay", "value": [67, 131, 0], "rank": 1, "of": 64, "margin_to_next": 1,
     "status": "Verified",
     "sources": [{"name": "D1 src_franchise_weekly_score", "read_at": "2026-10-17T14:45Z"},
                 {"name": "MFL weeklyResults 2026 W1-6", "read_at": "2026-10-21T14:06Z"}]}
  ],
  "status": "Verified",                     // the lowest status of its facts
  "publishable": true,                      // Verified, or Reconciled with an approved correction
  "caveat": "Lowest by one All-Play win; a stat correction could tie it.",
  "generated_at": "2026-10-21T00:00:00Z", "code_sha": "…"
}

Rules:

1. Numbers live only in facts. The editor writes the words; the Wire's existing check refuses digits in lead text.
2. Conflict and Unknown are never publishable. Provisional appears in the editor box only, with its caveat. Reconciled is publishable only after its correction is approved.
3. Every comparison names its population, its sample weeks and its exclusions. "Ever" may only be used when the population is every season.
4. Game-level claims (series, streaks, "first time since") use only seasons with listed games: 2012 and 2020 on. Record-level claims may use every season.
5. Owners are people, not franchise ids. Until identity 7a–7e are ruled, any claim touching those seasons is Provisional.
6. The validator re-computes every fact from the sources before the UPS Center can publish, like today's Heat Check check.

10. Every proposed historical correction needing your approval

Nothing below has been written.

• How a correction lands: after your ruling, through the corrections overlay (canon §E), with a dry run shown first and the same fix in the local database.
• Size: about 120 stored rows across all groups. The exact count comes from the dry run.
• Order: one decision at a time.

A. Settled by MFL's own record (decision 5)

B. Judgment calls (decision 6)

C. Still open from before

• Identity 7a–7e (comments on #1329):
  • 2010 → 2011 franchise lineage
  • Dave/David Murray
  • 0002 in 2018
  • 0005 in 2017–18
  • the 0006 handover date in 2024
• Decision 8: owner career stats (#1335). It needs a production rebuild.
• Decision 9: recorded seeds in the API.

D. Code fixes (no history change; for the build stage)

• /api/eras counts Hawk Tuah wins as finish 12. It should be 7.
• Hall of Champions: the seed heuristic and the fallback champion guess.
• IDX basis and its name (section 7).
• The migration 0033 comment says "12 = Hawktuah champion".

Not verified:

• 2025's finish → pick (the 2026 draft's trades aren't in the trace).
• 2010's weekly scores, which are partly corrupt; a reload needs one more MFL read.
• Owner names before 2011, beyond what's listed.