[jbob06]

The problem

Kills per hour leaves out seed time, but seed time is only kept while a Seeding reward rule is on. A server that rewards seeding some other way (ours does it from a Discord bot) has no seed time at all, so every hour a player spends on the server waiting for the match to start counts as played.

On our server (MinimumRequiredPlayers=20) a match on 2026-10-17 sat at a score of 0 for about six hours with 1 to 7 players on before it went live. A player who seeded it and then played the match had 434 minutes on and 43 kills that day, and their K/h divided by all six hours of the wait. They asked for K/h to count only the time a match is actually being played.

The change

• Idle time is tallied beside time on (tallyLook, match-players.ts): the gaps a player was on while no side had scored yet, the same test AFK protection uses for "the server is seeding". It is written with the line as match_players.idle_seconds.
• Migration 0051 adds the column to match_players, player_totals and player_days, and restates the six functions that list the line columns: 0038's player_totals_recount and 0039's player_days_recount, player_lines_cleared, player_totals_match_ended, player_days_rebuild and player_totals_rebuild. Each has idle_seconds added and nothing else changed (checked line by line against the originals). Every existing line has 0, so nothing needs rebuilding.
• Boards: rangeBase, seasonBase and totalsBase carry idle_minutes, and kills per hour is kills / (minutes - GREATEST(seed_minutes, idle_minutes)): in SQL for the sort, and in perHour for the table and the export. The larger of the two rather than their sum, because on a server with a Seeding reward both count mostly the same minutes.
• Cash per minute leaves out the same wait, cash / (cash_minutes - GREATEST(seed_minutes, idle_minutes)), in SQL and in perMinute. Its time is whole sessions while idle time goes by when a match ended, so on a 7, 30 or 90-day board the two can differ a little at the range's first day; all time they agree. Seed time has the same kind of edge.

Things worth your view

• "No side has scored" or the feed's match state. Season 2's feed reports MatchState.PreMatch.WaitingForPlayers, which says it exactly, but only where the feed reaches Warcon. Scores work on every server, so I used them. Preferring the state when there is one would be a small follow-up.
• Idle time is kept from the deploy on; earlier matches have none, so their K/h reads as before.
• A server whose match starts scoring with anyone on records next to no idle time, and its K/h is as before.
• The first score period of a live match counts as idle (24 s on ours), which is negligible.

Tests

• tallyLook tallies idle time only while no side has scored, and none when the scores are unknown; writeMatchPlayers stores it; perHour and perMinute leave out the larger of seed and idle time.
• totals-oracle.ts carries idle_minutes, so the exactness tests compare it. The 120-step test writes lines with idle time (fixed values: a random draw would change the seeded history after it), and dayTable compares the column.
• The tests that insert lines into a database migrated only to 0037, 0038 or 0046 now use insertEarlyLines (src/test/db.ts): a raw insert of just the columns given, because Drizzle's insert names every column in the schema, idle_seconds included, which those databases do not have yet. Any later column on match_players would have hit the same thing.

Checks

• prettier --check .: clean.
• bun run check: 0 errors, 0 warnings.
• bun test with TEST_DATABASE_URL on Linux, against timescale/timescaledb:2.30.0-pg18: 1669 pass, 1 skip, 0 fail across 169 files.
• Not run against a live server or in a browser.

🤖 Generated with Claude Code

https://claude.ai/code/session_011q1rvh2JotX9onGUYQH78V