untraded

Open, transparent Counter-Strike 2 match metrics.

The metrics engine of Untraded, a self-hostable CS2 demo analyzer and coach. It turns a demo into canonical tables, a round model and 81 versioned metrics: base stats, duels and trades, positioning, utility, economy, HLTV 1.0 and an approximation of HLTV 2.0. Every metric ships with its formula, its parameters, its version and its tests (metric pages), and the shared stats are checked against CS Demo Manager on 30 matches (report).

Alpha. The library and the command line are usable now; the web app comes later. Metric versions follow semver, but the Python API may still change before 1.0.

Command line

untraded analyze match.dem.zst --mode premier              # terminal report
untraded analyze match.dem.zst --mode premier --side CT    # CT rounds only
untraded analyze match.dem.zst --mode premier --format json > report.json
untraded analyze match.dem.zst --mode premier --out out/   # also writes Parquet

The report lists every player's main stats (K, D, A, ADR, HS %, KAST, opening kills, untraded deaths, HLTV 1.0 and an approximation of HLTV 2.0), the same per side, and the key rounds: multi-kills, won clutches, eco or force rounds won against a full buy. The JSON output holds every metric of the registry, with its version. --out writes the canonical tables, the round model and the metrics as Parquet.

Demos can be .dem, .dem.bz2, .dem.gz or .dem.zst. The modes are premier, faceit, esea, scrim and pro.

Several matches:

untraded parse demos/*.dem.bz2 --mode premier --out parsed/     # in parallel, as the free RAM allows
untraded reference parsed/matches/* --player 7656119... --session <match id>

untraded parse takes many demos at once and adapts to the machine (low priority, a demo starts when the free RAM covers it). untraded reference compares a session with the player's own 30 days before it, recomputed over all the rounds, and shown from 5 matches and 100 rounds. untraded calibrate measures the adjustable thresholds on a corpus (method).

untraded compare-csdm checks our stats against CS Demo Manager's analyzer (see docs/validation-csdm.md).

untraded fixture extract match.dem.zst --mode premier --rounds 11-15 --name my-bug --out my-bug/ writes an anonymized extract of a few rounds: steamids, names, server, teams, date and rank are removed. It is enough to reproduce most metric issues, so you can attach it to a bug report instead of your demo.

Library

from pathlib import Path

from untraded.derive import derive
from untraded.metrics import Dataset, compute
from untraded.parsing import parse_demo

match = parse_demo(Path("match.dem.zst"), source="upload", mode="premier")
tables = match.tables()
ds = Dataset.from_matches([{**tables, **derive(tables).tables()}])
compute(ds)  # every value metric, per player
compute(ds.split(side="CT"), ["base.kast"])  # one metric, CT rounds only
compute(ds, params={"trade.window_s": 3.0})  # another trade window

Results record each metric's version and the parameters used. The ratings are also available on match totals:

from untraded import rating_hltv_1

rating_hltv_1(rounds=24, kills=25, deaths=15, multikills={1: 10, 2: 4, 3: 1, 4: 1})

rating_hltv_1 implements HLTV's public Rating 1.0 formula:

(KillRating + 0.7 × SurvivalRating + MultiKillRating) / 2.7

KillRating      = (kills / rounds) / 0.679
SurvivalRating  = ((rounds − deaths) / rounds) / 0.317
MultiKillRating = ((1K + 4×2K + 9×3K + 16×4K + 25×5K) / rounds) / 1.277

rating_hltv_2_approx is a community approximation of Rating 2.0, whose formula HLTV does not publish: always present it as "approx.".

Untraded is not affiliated with HLTV or Valve.

License

MIT