The NOT clause was protecting both (team1,team2,date) and (team2,team1,date)
from deletion, preventing the stale cleanup from removing reversed-order
duplicates — e.g. both "France vs Spain" and "Spain vs France" stored for
the same SF slot. Only the exact pairing being upserted needs protection;
reversed entries with the same date are duplicates and should be deleted.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Wikipedia pre-fills QF/SF/Final bracket slots with projected team names
before the preceding round is complete. After each sync, delete any
unplayed knockout match where either team still has an unplayed match in
an earlier round — these entries are speculative and shouldn't be shown
as confirmed fixtures.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The squads Wikipedia page for 48 teams is very large and was causing the
scheduled sync to be OOM-killed (exit 137) mid-run. Squad data doesn't
change during the tournament, so it doesn't need to run on every sync.
Pass --squads to include it when needed (e.g. initial setup).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The stale-entry cleanup was deleting played matches when a team appeared
in a newly-scraped entry for the same round. Add AND score_ft_home IS NULL
so only unplayed placeholder entries are removed, not completed results.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The previous fix only removed stale entries when the team pairing changed.
It missed the case where Wikipedia updates the date/time for the same teams
(same pair, different date = different upsert key). Now any knockout match
with the same year+round involving either team is deleted unless it is the
exact same pairing AND date.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The upsert key is (year, team1_id, team2_id, date), so when Wikipedia updates
a placeholder bracket entry to real opponents, the old pairing is never
overwritten — it just accumulates alongside the new one. Before upserting a
knockout match, delete any existing match in the same year+round that involves
either team but isn't the exact same pairing.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
sync.ts: after computing standings from played matches, seed 0-0-0-0 rows
for every team in any group match, so all 12 groups always appear.
/groups: fetch all 2026 matches alongside standings; each group card now
shows results (score), live badge, and upcoming fixtures with local
kickoff time, sorted by UTC kickoff.
/tournaments/[year]: derive group list from union of standings + match
group names, so groups with no played matches still render with their
fixtures.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Detect Wikipedia plain-text rate-limit response ("You are making too many
requests") and wait 30s before retrying, rather than silently failing
- Increase inter-attempt delay from 3s to 15s per attempt
- Increase group subpage delay from 1.2s to 3s, year delay from 0.6s to 2s
- Re-scrape 1982, 1998, 2002, 2006 which had failed groups; all groups now
complete — e.g. 2002 now has 64 matches including Group E (Germany/Klose)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Merge data/wikipedia/{year}/ into data/{year}/ so there is a single
canonical location for World Cup JSON files. Update scrape and seed
scripts to use data/ instead of data/wikipedia/.
Re-scraped all 22 years (1930-2022) with fixed player name extraction
(full name from <a title="..."> rather than abbreviated display text)
so historical goals now show e.g. "Thomas Müller" not "Müller".
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Needed to recover from duplicate team entries (Bosnia & Herzegovina / USA)
that persisted because ON CONFLICT matching is on team IDs, so old rows
with wrong team IDs are never updated. --force clears all 2026 data and
orphaned teams before re-syncing clean.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add TEAM_ALIASES to lib/wiki-scraper.ts applied at extraction time so both
scraper and sync consistently produce canonical names. Removes the duplicate
alias map from seed.ts in favour of the shared normalizeTeam() export.
Aliases added:
Bosnia & Herzegovina → Bosnia and Herzegovina
USA → United States
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move all scraping logic (fetchWikiHtml, scrapeYear, scrapeSquads and all
helpers) into lib/wiki-scraper.ts as exported functions shared by both scripts.
scrape-wikipedia.ts becomes a composable CLI:
pnpm scrape [year] — matches + squads (default)
pnpm scrape [year] --matches — matches/meta/stadiums only
pnpm scrape [year] --squads — squads only
sync.ts drops the openfootball GitHub dependency entirely and scrapes
Wikipedia directly. Incremental: completed groups (all matches have FT
scores) are detected via DB query and their sub-pages are skipped each run.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add worldcup.meta.json per year with host, teams_count, winner, runner_up,
third_place, fourth_place — derived from match results (Final/Third-place
match) with infobox as fallback for edge cases like 1950's round-robin final.
Fix infobox host extraction to handle <br>-separated multi-host entries
(2002: Japan / South Korea). Fix squad scraper to filter out zero-player
phantom sections that Wikipedia appends (References, Captains, etc.).
Drop app/data/world_cup.csv and the PLACEMENTS/parseCsv code in seed.ts —
all tournament metadata now comes from the scraped JSON files.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move world_cup.csv to app/data/ directly (the only remaining Kaggle file
used by seed.ts for tournament metadata). Delete the rest of the Kaggle CSVs.
Update path constants in scrape-wikipedia.ts and seed.ts accordingly.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add scripts/scrape-wikipedia.ts that fetches all 22 World Cups (1930–2022)
from English Wikipedia via MediaWiki API, handles group sub-pages, AET/penalty
detection, and goal parsing, writing openfootball-format JSON to app/data/openfootball/.
Rewrite scripts/seed.ts to read these local JSON files instead of the Kaggle
CSV, producing 965 matches and 2716 goals with per-group assignments for all
historical tournaments (enabling group standings on tournament pages).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously each match goal sync did: DELETE (auto-commit) → N
individual INSERTs (each auto-commit). During those ~50ms readers
saw 0 goals for the match — the inconsistency window.
Now: collectGoals() builds the rows in memory, replaceGoals() wraps
the DELETE + single bulk VALUES INSERT in a transaction. Under
Postgres READ COMMITTED, readers see the old goals until commit and
the full new set after — never an empty window.
Also drop sync pool from max:5 → max:2; the job is fully sequential
and was holding idle connections unnecessarily.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
TypeScript doesn't narrow module-level consts across closure
boundaries, so the explicit process.exit(1) guard isn't enough —
add ! assertion at the usage site inside run().
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The DDL block in sync.ts was a "safety net" but caused misleading
password auth errors when Coolify's scheduled task ran without
DATABASE_URL injected — the fallback `wc:wc` password was wrong.
- Drop the silent `?? 'postgres://wc:wc@...'` fallback; exit with a
clear message if DATABASE_URL is missing so the root cause is obvious
- Remove the 90-line CREATE TABLE IF NOT EXISTS block — seed.ts runs
before the server starts and guarantees all tables exist
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Soviet Union (su), Yugoslavia (yu), East Germany, Germany DR, FR Yugoslavia,
and Czechoslovakia have no valid entry in flag-icons. Map them to null in
TEAM_ISO so getIso() returns null, and render a muted initials badge in
TeamFlag instead of a broken/empty sprite. Also drop the buggy 2-char
substring fallback that generated random valid codes for unknown teams.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- scripts/seed.ts: one-time import of Kaggle FIFA dataset (matches_1930_2022.csv,
world_cup.csv) covering all 964 matches and 2720 goals from 1930-2022 with full
scorer names, minutes, penalties, and own goals for every tournament
- scripts/sync.ts: stripped to 2026 only (openfootball live data); historical years
removed since Kaggle is now authoritative for 1930-2022
- Dockerfile: copy app/data into runner image; CMD runs seed.ts before server.js so
a fresh deployment auto-seeds on first start (skips if already seeded)
- package.json: add 'seed' script; use --force to re-import from updated CSV files
- app/data/kaggle/: bundle Kaggle CSV files in repo
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
syncGoals() was calling DELETE FROM goals WHERE match_id=X at the top,
so processing goals2 (away team) wiped out goals1 (home team) that were
just inserted. Every match with goals from both sides lost all home-team
goals — Ronaldo's hat-trick vs Spain, Kane's vs Panama, and many others.
Fix: move DELETE above the goals1/goals2 loop, executed once per match.
Result: 2018 goal count corrected from 107 → 169; hat tricks from 8 → 18.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
openfootball uses 'West Germany' for 1954–1990 era matches. All DB references
(matches, goals, group_standings, squads) have been merged into the Germany
team on both local and VPS. TEAM_ALIASES map prevents re-creation on re-sync.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Drizzle ORM mutates client.options (parsers/serializers) after the postgres
client is created, which causes the separately-passed password option to be
lost on the actual connection attempt. Root cause confirmed on VPS: raw
postgres.js query succeeded while drizzle.execute() failed with auth error.
Fix: encode the password directly in DATABASE_URL (%23 = #, %5D = ], %3D = =).
postgres.js decodes percent-encoding correctly. No separate DB_PASSWORD env
var needed in the app container anymore.
DB_PASSWORD is still used by the Postgres container (POSTGRES_PASSWORD).
Coolify env var to set: DATABASE_URL=postgres://wc:<encoded-pass>@db:5432/worldcup
Also adds resolver-level isMissingTable() guards so the app returns empty
results instead of GraphQL errors on a fresh deploy before sync runs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The sync script created its own postgres client and only read DATABASE_URL,
bypassing the DB_PASSWORD override that lib/db/index.ts already applied.
Since DATABASE_URL has no password embedded, auth always failed.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Full-stack World Cup web app (1930–2026):
- Next.js 16 + TailwindCSS 4 + GraphQL Yoga + Apollo Client 4 + Drizzle + PostgreSQL 16
- 23 tournaments synced from openfootball/worldcup.json (matches, goals, teams, stadiums, squads, standings)
- Pages: home (live), groups, stats, history, search, /tournaments/[year], /teams/[slug], /players/[name]
- Live match detection via isLive() + Apollo 60 s poll
- pnpm with node-linker=hoisted for Docker compatibility
- docker-compose.yml with Traefik labels (HTTPS redirect, TLS, security middleware)
- docker-compose.dev.yml for local dev (DB only, port 5432 exposed)
- Dockerfile: multi-stage pnpm build, standalone Next.js output, sync script bundled
- .env.example with all required variables documented
- Comprehensive README with local dev, deployment, schema, and GraphQL API reference
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
