mintopia/blur

Watch

Fork 0

Feed: use and collect events, per-racer totals #35

New issue

Closed

opened 2026-10-20 18:41:26 +00:00 by mintopia · 2 comments

mintopia commented

2026-10-20 18:41:26 +00:00

Owner

Events

use: one per use of a powerup by any racer: a bolt volley = 3, placing a mine = 1, Nitro/Repair/Shield included. Sources: channel-9 event 0 (fired, ADR 0017), event 4 self-applied (Nitro/Repair, ADR 0016), event 14 (Shield). data: racer, powerup, serial.
collect: inferred from the 5 Hz held-queue diff (ADR 0017). A use (or Bribe grant) of a kind never seen held backfills a collect with inferred: true.
hit unchanged (one per strike, mine hit separate from its placement).

Totals

Per racer, live in standings and final in race_result, every mode: totals with per-kind maps uses, collections, hits_dealt, hits_taken, hits_shielded, damage_dealt, damage_taken; plus wrecks_dealt, wrecks_suffered (powerup + collision), self_hits. Counted from feed events so they agree on every instance. Decoy counts come with #36.

Acceptance

Verified on SP, split-screen and 2-instance LAN with the bot scenarios: counts match event logs.

Rules

Wire format: ADR 0019 (docs/decisions/0019-feed-message-contract.md). Repo data/tools: ADR 0020.
Fix and verify on BLUR_ENGINE=recomp only.
Update docs/state-websocket.md and examples/live-timing for any feed change.
Test races: use the scripts/ scenarios (windowed, low settings) once #33 exists.

## Events - `use`: one per use of a powerup by any racer: a bolt volley = 3, placing a mine = 1, Nitro/Repair/Shield included. Sources: channel-9 event 0 (fired, ADR 0017), event 4 self-applied (Nitro/Repair, ADR 0016), event 14 (Shield). `data`: `racer`, `powerup`, `serial`. - `collect`: inferred from the 5 Hz held-queue diff (ADR 0017). A `use` (or Bribe grant) of a kind never seen held backfills a `collect` with `inferred: true`. - `hit` unchanged (one per strike, mine hit separate from its placement). ## Totals Per racer, live in `standings` and final in `race_result`, every mode: `totals` with per-kind maps `uses`, `collections`, `hits_dealt`, `hits_taken`, `hits_shielded`, `damage_dealt`, `damage_taken`; plus `wrecks_dealt`, `wrecks_suffered` (powerup + collision), `self_hits`. Counted from feed events so they agree on every instance. Decoy counts come with #36. ## Acceptance - Verified on SP, split-screen and 2-instance LAN with the bot scenarios: counts match event logs. ## Rules - Wire format: ADR 0019 (`docs/decisions/0019-feed-message-contract.md`). Repo data/tools: ADR 0020. - Fix and verify on `BLUR_ENGINE=recomp` only. - Update `docs/state-websocket.md` and `examples/live-timing` for any feed change. - Test races: use the `scripts/` scenarios (windowed, low settings) once #33 exists.

mintopia added this to the Feed: standard messages, powerup events, race stats milestone

2026-10-20 18:41:26 +00:00

mintopia referenced this issue

2026-10-20 18:41:34 +00:00

Bribe's starting powerup missing from the feed #37

mintopia added a new dependency

2026-10-20 18:41:38 +00:00

#33 Test-race scenarios and race bot in scripts/

mintopia added a new dependency

2026-10-20 18:41:40 +00:00

#36 Decoy drop and decoy hit events

mintopia added a new dependency

2026-10-20 18:41:41 +00:00

#37 Bribe's starting powerup missing from the feed

mintopia added the

ready-for-agent

label

2026-10-20 18:41:44 +00:00

mintopia added a new dependency

2026-10-20 22:34:36 +00:00

#49 JSON Schema for the game-state feed

mintopia self-assigned this

2026-10-21 09:41:32 +00:00

mintopia commented

2026-10-21 10:59:06 +00:00

Author

Owner

Implemented on branch harmonic/task-33 (commit 018dba7).

Events

use: channel-9 event 0 (fired; bolt volley = 3, mine placement = 1, Shield included, since it arrives as event 0 kind 7) plus event 4 self-applied (Nitro/Repair). data: racer, powerup, serial. Event 14 is not hooked; event 0 already covers Shield.
collect: from the 5 Hz held-queue diff (multiset diff, so reorders are ignored; out-of-range kinds are skipped). A use of a kind never seen held sends collect with inferred:true first, and the later queue gain is not counted twice.
hit unchanged, except a strike the game repeats with the same serial and victim is now sent once. Without that, LAN instances disagreed on hit counts.

Totals

Added totals to every standings entry, so it is live and also in race_result.standings. Fields: per-kind maps uses, collections, hits_dealt, hits_taken, hits_shielded, damage_dealt, damage_taken, plus wrecks_dealt, wrecks_suffered, self_hits.
Self-hits count only in self_hits.
Powerup wrecks come from the hit event. Collision wrecks come from the health poll.
Totals reset when a race ends, because the first powerups are fired before the feed sees lap 1.

Docs and example

docs/state-websocket.md updated.
The live-timing example gets HITS and WRECKS columns in the final classification.

Verification (recomp, bot scenarios; I regenerated the recompiled code first because the checked-in copy predated the event-send hook)

SP: 298 use / 160 collect / 92 hit events; final totals match the event log for all 20 racers.
Split-screen (2 pads): final totals match the event log.
2-instance LAN: use and hit totals identical on both instances after mapping racer indices. Collections differ only by polling noise (one remote out-of-range kind, now filtered). The LAN bot race did not reach a finish within 500 s, so I compared the live standings totals.
race_feed, hit_event, game_state and state_ws ctests pass. I did not run the full test suite or /code-review, and I did not check the live-timing page in a browser.
Decoy counts are left for #36.

Implemented on branch harmonic/task-33 (commit 018dba7). **Events** - `use`: channel-9 event 0 (fired; bolt volley = 3, mine placement = 1, Shield included, since it arrives as event 0 kind 7) plus event 4 self-applied (Nitro/Repair). `data`: `racer`, `powerup`, `serial`. Event 14 is not hooked; event 0 already covers Shield. - `collect`: from the 5 Hz held-queue diff (multiset diff, so reorders are ignored; out-of-range kinds are skipped). A `use` of a kind never seen held sends `collect` with `inferred:true` first, and the later queue gain is not counted twice. - `hit` unchanged, except a strike the game repeats with the same serial and victim is now sent once. Without that, LAN instances disagreed on hit counts. **Totals** - Added `totals` to every `standings` entry, so it is live and also in `race_result.standings`. Fields: per-kind maps `uses`, `collections`, `hits_dealt`, `hits_taken`, `hits_shielded`, `damage_dealt`, `damage_taken`, plus `wrecks_dealt`, `wrecks_suffered`, `self_hits`. - Self-hits count only in `self_hits`. - Powerup wrecks come from the hit event. Collision wrecks come from the health poll. - Totals reset when a race ends, because the first powerups are fired before the feed sees lap 1. **Docs and example** - `docs/state-websocket.md` updated. - The live-timing example gets HITS and WRECKS columns in the final classification. **Verification** (recomp, bot scenarios; I regenerated the recompiled code first because the checked-in copy predated the event-send hook) - SP: 298 use / 160 collect / 92 hit events; final totals match the event log for all 20 racers. - Split-screen (2 pads): final totals match the event log. - 2-instance LAN: use and hit totals identical on both instances after mapping racer indices. Collections differ only by polling noise (one remote out-of-range kind, now filtered). The LAN bot race did not reach a finish within 500 s, so I compared the live `standings` totals. - `race_feed`, `hit_event`, `game_state` and `state_ws` ctests pass. I did not run the full test suite or `/code-review`, and I did not check the live-timing page in a browser. - Decoy counts are left for #36.

mintopia closed this issue

2026-10-21 11:02:25 +00:00

mintopia commented

2026-10-21 11:02:25 +00:00

Author

Owner

Completed and merged by Harmonic (task 33).

mintopia referenced this issue

2026-10-21 11:16:46 +00:00

Game award stats and awards in the feed #38