How the live audience is measured

A concise guide to what the dashboard counts, how streams are labeled, and where the data has limits.

What “comp” means

Comp is the live audience of the Dead by Daylight comp scene: the sum of the viewer counts of the streams that belong to it at the time of collection. It counts viewers, not people who play competitively.

Two different signals decide whether a stream is counted. The first is the channel: a channel on the maintained competitive-channel list, or a player on a tracked team roster, counts all of its viewers whatever it streams that day, because the channel itself is part of the comp community and its audience is the scene's audience. The second is the broadcast: any other stream counts only when its title matches a competitive rule. Title rules do not establish that a creator is affiliated with the comp community.

So the headline measures the audience of the comp scene, not only competitive matches: a comp player streaming public matches still counts. How much of it is watching competitive play is shown below.

Competitive play

Inside the comp audience, every comp stream is also read for what it has on air at each check. Competitive play is two kinds of stream. A match: tournaments of any size, including charity and community ones, their watch parties and casters, friendlies and showmatches, a “vs” against a named opponent, scrims, ladder and 1v1. A streak: a run of wins against top lobbies. Everything else a comp channel streams is counted as other comp streams. The comp headline does not change: competitive play is a part of it, never more.

Only what is on air counts. A title such as “pubs, scrims later”, “ladder at 9PM EST” or a chat command like “!ladder” does not make a stream competitive play; “pub/ladder, DBDL later” does, for the ladder. Streams of known tournament organizers and streams the event detector tags as part of a match always count as a match.

What the comp audience watched, last 7 days

Matches47%
Streaks19%
Other comp streams34%

Shares of comp hours watched. Competitive play is 66% with streaks and 47% without them.

Why streaks count as competitive play. A streak has no opponent team, but it is still a competition: the player keeps winning against the highest-MMR lobbies, one loss ends the run, and every match is played fully locked in. That is closer to competitive play than to casual content, so streak streams are counted here. They were 19% of the comp audience in the last 7 days; tournaments and matches against a named opponent are reported apart from them, so either figure can be read on its own.

How streams are classified

The live scan reads Twitch streams in the Dead by Daylight category. Each stream is checked in the order below, and the first rule that matches decides. Lists can come from the database, with static lists used as a fallback when classification lookup is unavailable. Channel tags never decide content.

Excluded channel → Casual. Channels that look competitive by title but are not are kept on an excluded list and always count as casual.
Listed channel or team roster → Comp. Listed channels and tracked roster players count as comp whatever the stream shows.
Title heuristics → Comp. The broadcast title alone is checked against these rules:
- Ladder or scrim wording, such as “ladder” or “scrims”. These stand alone because the scan is already limited to Dead by Daylight.
- Russian tournament and scrim terms.
- An English competitive term next to “DBD”, such as “DBD tournament” or “Competitive DBD”, with at most one word between them. A term that sits near another game in the same title does not count.
- Spanish and Portuguese tournament wording: a tournament term next to “DBD”, such as “Torneo DBD”; a tournament with a bracket stage, such as “Campeonato de 1v1 fase 1” or “Final del torneo”; or “A vs B” with tournament or stage wording, such as “Copa - Ronda 1 - A vs B”.
- “A vs B” with league, season, round, playoff, or cup wording, such as “Team A vs Team B - Week 5”.
- “Team X” with streak, scrim, tournament, or league wording, such as “Team X winstreak”.
- “1v1 @opponent” with a named opponent. Lower confidence: a bare “1v1” does not count.
Anything else → Casual. The default is casual.

Live classification returns a result immediately; the live scan does not wait for per-stream approval. Separate review uses curator labels to validate and evaluate classifications, and curated list changes are explicit.

Illustrative example: the streams and viewer numbers are not live data.

Separate curator review

Reviewed candidatesProject curatorCOMPNOT COMP

Labels validate and evaluate candidates; they do not gate live streams. Curated list edits are explicit.

Rules version and accuracy

The title rules are versioned. The current version is 2026-10-27.1, and it is raised whenever a rule or term changes, so results can be compared across versions.

Accuracy is checked against curator labels. A curator marks channels COMP or NOT COMP, and the classifier is scored on two measures: precision, the share of channels it marks as comp that curators confirm, and recall, the share of confirmed comp channels it finds. Rules are tuned on one set of samples and then checked on later samples, because a rule tuned and tested on the same data looks better than it will perform on new streams.

Viewer totals and comparison

Total is the sum of valid viewer counts across unique live streams returned for the DBD category. Comp is the classified subset; casual is calculated as total minus comp. The home comparison uses the mean comp share from stored samples collected 24–48 hours before now, and is left out with fewer than two valid samples.

Analytics compares a period with the equal period just before it only when that earlier period holds at least 80% of the checks it should (one every 5 minutes). Otherwise the comparison is left out rather than measured against missing data, for example before tracking began on 9 October 2026. Changes describe what was observed; they do not explain why the audience moved.

Comp hours watched add up the comp viewers of every stored check, each check standing for 5 minutes. A missing check adds nothing: viewers are never estimated for time without data, so an outage can only lower the figure, never raise it. Charts break the line where checks are missing instead of drawing across the gap.

A match's peak adds up the players' channels within the same 5-minute check, each stream once, and takes the highest of those sums. Channel lists show what each channel had in that same check, so they add up to the peak; a channel's own highest count can come from another moment. Co-streams are reported separately and only added in the figure “with co-streams”: watch parties, casters and the official broadcast, any channel that streamed the match without playing it. That figure takes its own highest check, and its list shows the co-streams in that check on top of the players' viewers at the same moment, so it adds up too.

Collection, status, and freshness

Twitch collection can span multiple pages and records the start and finish timestamps for each scan. A complete scan is labeled complete; a scan that stops early or encounters a collection problem is partial. A failure before usable data is available can leave the last cached result in view, marked delayed, or leave collection status unavailable when metadata is missing. Partial scans do not replace the last complete cache or create stored viewer-stat snapshots.

The home page refreshes its live request every 90 seconds and the server cache also lasts 90 seconds. Stored history is written by authorized scheduler collection requests, but the deployed invocation schedule is not represented by this page; sample spacing can vary. The displayed collection time comes from the source scan’s completion timestamp, not an inferred last-success time. Missing timestamps or status metadata are shown as unavailable.

Limitations

Viewer counts are concurrent snapshots, not unique viewers.
Classification is rule-based, not machine learning. It reads broadcast titles and channel lists, never the video.
A competitive player who is not on the list and streams under a title without competitive wording is missed until a curator adds the channel.
Listed channels count all their viewers, including casual content. Titles can also be stale or misleading, so a stream can be counted or missed for as long as its title stays the same.
Title rules cover English, Spanish, Portuguese and Russian wording only; titles in other languages are counted only through the channel lists.
Competitive play is read from titles too: a match under a title without match wording counts as other comp streams until the event detector or a curated list recognizes it.
Changes in the totals do not establish that an event caused a change.
The data does not measure viewer retention.
No coverage or confidence percentage is reported because the current data does not provide a computable basis for one.
Category results and classifications can be incomplete; partial scans, missing samples, and stale cached data are surfaced where their metadata is available.