Radar Six Follow @RadarSixHQ

CLASSIFICATION METHODOLOGY LIVE EVIDENCE

Six source families. One classification rulebook.

Radar Six classifies every observation into one of six surfaces before it appears on the live feed. This page records the classification rules, the failure modes we watch for, the cadence each surface emits at, and what the desk is observing right now.

Outlet audit

Reconciliation pages list threshold-clearing outlets that appear in Radar Six claim source chains. Counts are descriptive, not a ranking.

Open outlet audit

Live observation index

LIVE Generated
  • Reddit
    33 obs · 24h 7d 46

    Last observed:

    Latest: Take-Two/Rockstar's patents for GTA VI are unreal. How is ANY of this running on a PS5?

  • Rockstar
    1 obs · 24h 7d 1

    Last observed:

    Latest: Take-Two Interactive Software, Inc. Reports Results for Fiscal Second Quarter 2026

  • YouTube
    0 obs · 24h 7d 0

    Last observed: no data

    Latest: no recent observation

  • Search
    0 obs · 24h 7d 0

    Last observed: no data

    Latest: no recent observation

  • X
    0 obs · 24h 7d 0

    Last observed: no data

    Latest: no recent observation

  • Press
    3 obs · 24h 7d 3

    Last observed:

    Latest: GTA 6 hopefuls encouraged by Sony to upgrade to PS5 now

CLASSIFICATION

How each surface is classified.

  1. reddit

    Match keys: reddit

    WHY INCLUDED

    The longest-running enthusiast venue. Thread velocity and comment-rate behavior are observable in real time.

    GOOD AT

    • First reactions land here.
    • Community moderators push back on weak claims.
    • Fact-checking happens inside the thread.

    WEAK AT

    • Source-chain compression (Reddit reposts are often two layers downstream).
    • Platform-native noise (karma incentives shape what surfaces).
    • Creator-cohort bias (the same 10 power users dominate).

    CADENCE

    Cadence: continuous; activity is heaviest during US weekday evenings.

    AGING POLICY

    Active Now: 24h. Public feed: 7d. Older observations remain in the operator audit log.

  2. rockstar

    Match keys: rockstar, rockstargames, rockstar_games, rss_atom_feed:take2_ir_rss, rss_atom_feed:rockstar_newswire_rss

    WHY INCLUDED

    The only first-party publisher channel. Cadence and metadata behavior are themselves the signal.

    GOOD AT

    • First-party silence patterns.
    • Metadata-shift detection (page-modified header changes without body changes).
    • Newswire cadence comparison.

    WEAK AT

    • First-party silence is the dominant state, so most weeks emit zero observations.
    • False-positive metadata changes from CDN regeneration.
    • No cadence floor.

    CADENCE

    Cadence: weeks to months between Rockstar publishing anything meaningful.

    AGING POLICY

    Active Now: 24h. Public feed: 7d. Older observations remain in the operator audit log.

  3. youtube

    Match keys: youtube, yt

    WHY INCLUDED

    Tracked-creator behavior often telegraphs intent before content drops.

    GOOD AT

    • Pre-publish behavioral changes, like comment moderation toggled or a video unlisted within minutes.
    • Creator cross-talk between channels in the watched cohort.
    • Tracked-channel publishing cadence relative to baseline.

    WEAK AT

    • Creator-cohort bias (the watched cohort is small and visible).
    • Platform-native noise (algorithmic boost masquerading as interest).
    • Source-chain compression.

    CADENCE

    Cadence: daily; activity peaks around content release windows.

    AGING POLICY

    Active Now: 24h. Public feed: 7d. Older observations remain in the operator audit log.

  4. search

    Match keys: search, google_trends, gtrends, search_trends

    WHY INCLUDED

    Public query interest is the cleanest crowd-attention proxy and the hardest to fake.

    GOOD AT

    • Geographic-distribution shifts.
    • Week-over-week interest changes.
    • Topic-graph adjacency.

    WEAK AT

    • Search-interest ambiguity (interest does not specify what triggered it).
    • Time aggregation: Trends reports buckets, not events.
    • Holiday and event spillover.

    CADENCE

    Cadence: bucketed; daily and weekly resolution.

    AGING POLICY

    Active Now: 24h. Public feed: 7d. Older observations remain in the operator audit log.

  5. x

    Match keys: x, twitter, x_twitter

    WHY INCLUDED

    Highest-velocity venue; tracked accounts often surface a post within minutes of first publication.

    GOOD AT

    • Repost timing across tracked accounts.
    • Very fast surface.
    • Cross-account corroboration.

    WEAK AT

    • Timing ambiguity (account local time, repost vs original).
    • Platform-native noise (engagement farming).
    • Source-chain compression (the most-shared post is rarely the source).

    CADENCE

    Cadence: continuous.

    AGING POLICY

    Active Now: 24h. Public feed: 7d. Older observations remain in the operator audit log.

  6. press

    Match keys: press, media, news, rss_atom_feed:eurogamer_feed, rss_atom_feed:gamespot_news_feed, rss_atom_feed:polygon_feed, rss_atom_feed:ign_feed, rss_atom_feed:kotaku_feed

    WHY INCLUDED

    Editorial provenance gives a different reliability bar than user-generated surfaces.

    GOOD AT

    • Editorial provenance preservation.
    • Named-source attribution.
    • Fact-checked summaries.

    WEAK AT

    • Press echo (multiple outlets republish the same source within hours, looking like independent confirmation).
    • Sourcing opacity (named sources sometimes redacted).
    • Embargo windows mean press observations can lag the underlying event.

    CADENCE

    Cadence: irregular; spike-driven.

    AGING POLICY

    Active Now: 24h. Public feed: 7d. Older observations remain in the operator audit log.

GLOBAL RULES

What classification means here.

Classification is the first step. Every observation is matched to one of these six surfaces by source key. Anything that does not match is dropped, not reclassified into a residual bucket. Match keys are the literal substrings the desk recognizes; an observation that arrives with an unfamiliar source key is treated as no observation at all.

Radar Six does not republish leak material, does not infer release dates from observation patterns, and does not reconstruct source chains beyond first-degree attribution. The published score reflects what the pattern looks like at the moment of observation. It is not a forecast.

FAILURE MODES

Seven ways an observation can mislead.

  1. Source-chain compression

    A leak that originates on a private channel surfaces on Reddit, then on X, then in press coverage within hours. Each subsequent venue looks like an independent observation but is downstream of the same source.

  2. Timing ambiguity

    A timestamp records when an observation reached the surface, not when the underlying event happened. Repost timing, account local time, and platform-side caching all shift the apparent moment of observation.

  3. Platform-native noise

    Every venue has incentives that produce surface activity unrelated to the topic. Karma farming, engagement farming, algorithmic boost, and recommendation cycles all produce observation-shaped noise.

  4. First-party silence

    A long absence from the publisher is itself a pattern. The desk observes the absence. Why the absence exists is outside what the data says.

  5. Search-interest ambiguity

    Public query interest reflects attention, not intent. A spike says people are searching for the term; it does not say what they expect to find.

  6. Press echo

    When multiple outlets republish the same source within a short window, the corpus of coverage looks larger than the underlying evidence. The desk records republication, not corroboration.

  7. Creator-cohort bias

    The watched cohort on every platform is small and visible. Cohort behavior does not generalize to the rest of the platform's audience.