Reference
Detectors
Every pattern the engine watches for, shown against one example channel’s tracked data — what each one looks for, when it can fire, and whether it has matched yet. Described, never prescribed.
Recently active
Matched a pattern within the last 28 days.
Milestones
A milestone reached
Notes the first time the channel crosses a round-number views, watch-time, or subscriber threshold. Useful early on, when each milestone is itself a signal that the catalog is building.
- Can fire from
- Sample threshold
An emerging signal
A catch-all slot for early-channel signals — first views, first impressions, first non-zero traffic from a given source, and other directional findings that often precede a clearer pattern.
- Can fire from
- Sample threshold
Discovery and traffic sources
A traffic source proportion shifting
Tracks per-video changes in the share of views coming from each traffic source. A shift usually shows up when YouTube starts surfacing a video through a different surface than it did at first.
- Can fire from
- Sample threshold
Single-source-dominated day
Names the most recent day inside the trailing fourteen on which one traffic source delivered eighty percent or more of channel-wide views. Channel-scope. Marks a single-surface day — useful for distinguishing a short-lived pulse from a steady mix and for matching it with what was happening on the catalog that day.
- Can fire from
- Sample threshold
Channel-wide top traffic source flipping
Compares the channel's top traffic source in two consecutive seven-day windows. Fires when the rank-one source changes — for example, Browse moving ahead of Suggested across the channel. A genuine shift in how the catalog as a whole is reaching new viewers, not a per-video signal.
- Can fire from
- Sample threshold
Per-source typical days-to-activation
Channel-scope summary of the per-source p25/median/p75 days-to-activation band — the typical age range a video must reach before each traffic source first sends views. Surfaces the gap between fast-activating surfaces (Suggested usually shows up within days) and slow-activating ones (Search and External commonly lag by weeks or months) on this channel's own history.
- Can fire from
- Sample threshold
A specific video's source activation outside the channel's band
Per-video flag — fires when a video's days-to-activation on a given source land outside the channel's p25-to-p75 band (with a two-day buffer). Often the most concrete way to call out a video that reached a surface unusually early or unusually late compared with the channel's typical pace on that surface.
- Can fire from
- Sample threshold
Source mix widening
Trailing 14-day slope of the channel's source-diversity score is rising. Typically a sign that views are spreading across more surfaces over time rather than concentrating on one or two.
- Can fire from
- Sample threshold
Source mix narrowing
Trailing 14-day slope of the channel's source-diversity score is falling. Usually means one or two surfaces are carrying a larger share of recent views than they did two weeks earlier.
- Can fire from
- Sample threshold
A source going quiet on a video
A specific video-and-source pair where the source sent at least 20 views in a 14-day prior window then sent zero across the last seven days. Often shows up when a feed-driven surface stops surfacing the video; can also reflect the natural tail of a short-lived spike.
- Can fire from
- Sample threshold
A gap in watch time per view across sources
Ranks the channel's traffic sources by average view duration and names the best-vs-worst gap. Viewers arriving from different surfaces often watch for very different lengths — this names which surface tends to send the longest-watching audience on this channel.
- Can fire from
- Sample threshold
Trajectory and outliers
A video tracking off the channel norm
Compares a video's view trajectory against the channel's percentile bands at the same days-since-publish. Marks videos sitting clearly above or below where similarly-aged videos usually land.
- Can fire from
- Sample threshold
Views accelerating
Picks up videos whose recent daily views are accelerating against their own decay curve. Acceleration is rare and tends to be short-lived; the detector flags the window, not a forecast.
- Can fire from
- Sample threshold
A week-over-week step up
A clear step up in weekly views compared to the prior week, beyond what the channel's own variability usually produces.
- Can fire from
- Sample threshold
Day-one numbers predicting longer-run reach
Tests how strongly day-one views correlate with day-30 totals across the channel. The strength of that correlation reflects how much the channel's reach is set early versus built over time.
- Can fire from
- Sample threshold
Cross-video outlier
Flags videos sitting far from the channel's median on any of five rate measures — CTR, retention, engagement, subscriber conversion, or views-per-day — using a robust, median-based outlier score (median and median-absolute-deviation, not a mean). Distinct from the trajectory outlier, which reads a video's view curve against same-age peers; this one reads where a video lands on those rates against the rest of the catalog.
- Can fire from
- Sample threshold
A live video sitting outside the channel's day-N cone for its age
Per-video reading on videos ≤ 30 days old. Compares the day-N reading against the channel's per-(age, peer) trajectory cone; fires when the day's views sit above p75 or below p25 of the cone for that age. The percentile is observational and rests on the channel's own history at that age — not a forecast of where the video lands. Decay-fit annotations attach when an R² ≥ 0.7 model is available.
- Can fire from
- Sample threshold
Per-video lifecycle
This video's arc
Reads the shape of a video's view curve and labels its current phase: early-burst, slow-burn, second-wind, or decline. Most videos pass through several phases over their life.
- Can fire from
- Sample threshold
Days from publish to peak
Records the day on which each video reached its highest daily-view total, and compares the spread across the catalog.
- Can fire from
- Sample threshold
Trend across recent debuts
Looks across the most recent batch of newly-published videos for a shared early-life pattern — a coarse marker for whether new videos are landing differently than older ones.
- Can fire from
- Sample threshold
Channel trend, week over week
Compares the channel's last seven days against the seven before — and against the channel's first seven days — across views, impressions, engagement, and traffic-source spread. Names the window-over-window movement as an observation; it does not grade the direction.
- Can fire from
- Sample threshold
Trend across recent videos
Compares the last five videos against the previous five — and against the channel's first five — on day-1 and day-7 views, plus the newest video against the one published before it. A coarse read of whether recent videos are landing differently than earlier ones.
- Can fire from
- Sample threshold
Engagement
Two metrics that usually move together pulling apart
Per-video cross-axis divergence. Names videos where two correlated engagement readings sit far apart — most often a high CTR paired with low retention, or strong retention paired with a quiet CTR. The V13 build runs this against the channel-internal CTR / retention medians; pre-V13 fixtures fall back to a percentile-rank pass that also covers views-per-day vs engagement-rate.
- Can fire from
- Sample threshold
A video drawing more engagement per view than the rest of the channel
Combined likes + comments + shares divided by views, per video. Fires when a video's ratio runs at least 2x the channel's median ratio AND lands in the top quartile of the channel's distribution. Names the single metric — comments, likes, or shares — whose per-view rate ran furthest above the channel's own typical pace, so the card stays concrete.
- Can fire from
- Sample threshold
A trailing-3-day jump in shares on a video
Sums a video's shares across the last three reporting days and compares the total against three days of the channel's typical per-video share rate. Fires when the recent total clears both an absolute floor of three shares and twice the channel's typical pace. Sharing tends to be lumpy — one external mention can show up as a clear pulse — so the multiplier gate sits well above day-to-day drift.
- Can fire from
- Sample threshold
Subscribe rate and reach moving in opposite directions
Channel-scope. Compares the trailing-28-day slope of the per-day subscribe rate (subs gained / trailing-7-day views) to the slope of trailing-7-day views itself. Two shapes can fire: audience quality improving (sub-rate climbing while reach holds roughly flat) and volume without depth (sub-rate easing while reach climbs). Both are framed as observation in opposite directions — neither is graded as good or bad.
- Can fire from
- Sample threshold
A sharp drop-off in the retention curve
Per-video reading — fires when a video's within-video retention curve shows a steep localized drop: the share of viewers still watching falls by at least 20 points across a narrow stretch of the runtime. Names where along the video the steepest decline sits. Descriptive only; the curve shows where viewers left, not why.
- Can fire from
- Sample threshold
Channel shape
Lifetime distribution of views across the catalog
Classifies the catalog's lifetime view distribution into hit-driven, balanced, long-tail, or flat. Hit-driven means the top video alone carries at least 40% of channel views; balanced is between 20% and 40%; long-tail is below 20% on the top but at least 70% on the top five combined; flat is below 20% on the top and below 50% on the top five. The detector stays quiet in the gap between long-tail and flat where neither label fits cleanly.
- Can fire from
- Sample threshold
Recent publishing gap notably longer than the channel's median
Finds the longest gap between two consecutive public-release dates inside the last 30 days. Fires when that gap is at least two times the channel's lifetime median inter-publish gap and at least seven absolute days. Prefers first_public_at when populated and falls back to published_at — the fallback is disclosed on the evidence row so the prose can hedge accordingly.
- Can fire from
- Sample threshold
How often this channel publishes
Classifies the publishing rhythm into a median-gap bucket — daily, near-daily, twice-weekly, weekly, biweekly, monthly, or irregular. Two special cases override the buckets: when five or more videos share one public-release date and every one of them carries first_public_at, the pattern reads as burst-launch; when first_public_at is missing on the cluster, the pattern reads as schema-ambiguous because batch-launch can't be told apart from an upload-date artifact.
- Can fire from
- Sample threshold
Recent 7-day publish count diverging from the prior 7-day count
Compares public-release counts in the trailing 7 days against the 7 days immediately before. Fires when the ratio is at least 2 (speedup) or at most 0.5 (slowdown) with an absolute difference of at least one video. A two-video floor on the prior window keeps 0-to-N or 1-to-N transitions from reading as infinite speedups — channels routinely sit quiet for a week and post once, and that is not a cadence shift.
- Can fire from
- Sample threshold
The catalog shape shifting
Detects when channel-wide patterns (concentration, cadence, source mix) moved enough to read as a phase change rather than week-to-week noise.
- Can fire from
- Sample threshold
How retention distributes across the catalog
Splits the catalog into retention tiers and notes how the distribution shifted over time — a slower-moving signal than per-video retention.
- Can fire from
- Sample threshold
Share of the catalog that went the past week without views
Reads the trailing-7-day share of public videos that received no views, against a maturity-scaled threshold (sprouting 0.50, growing 0.30, established 0.15). Quiet weeks are common as a catalog grows — recent videos and a few favorites usually carry most of the surface — so the gate is set wider for younger catalogs and tightens as the channel matures.
- Can fire from
- Sample threshold
Top-video and quiet-inventory shares trending up together
Reads the trailing 14-day OLS slopes of two channel-shape axes: the share of views carried by the single top video, and the share of public videos that received no views. Fires when both 10-day mean slopes clear 0.005 share-points per day upward AND the latest top-video 7-day share is at least 0.30. The dual gate keeps a brief concentration spike on a single upload from reading as a trend on its own — the catalog has to be meaningfully concentrated at the latest reading too.
- Can fire from
- Sample threshold
Share of videos below a view threshold
Counts how many public videos sit below a maturity-scaled view threshold (seedling 10, sprouting 100, growing and established 500). Stays quiet when every video clears the threshold and when no video clears it — the in-between case is the descriptive one. The framing is observational context rather than a high-signal alert, and the magnitude caps low so it rarely outranks more specific per-video findings.
- Can fire from
- Sample threshold
How often a video was iterated on
Counts the number of title or thumbnail changes a video has received and notes outliers against the catalog's typical iteration intensity.
- Can fire from
- Sample threshold
How fast YouTube is testing a video
Notes per-video impression rates that read as YouTube's ongoing impression-testing pattern — high impressions paired with CTRs that haven't settled yet.
- Can fire from
- Sample threshold
Quiet for now
Matched earlier in the channel's history, quiet in the recent window.
Discovery and traffic sources
First External-source views
Per-video milestone — emits one insight per video that received its first External-source views on the channel's earliest first-external date, when that date falls within the trailing seven days. External views typically arrive from links off YouTube (social posts, embeds, blog mentions); the matching timeline event is generated separately by the public-DB build.
- Can fire from
- Sample threshold
Not yet on its own line
Watching, but not yet reported on its own line — some patterns report through a shared slot.
- Discovery and traffic sources — 11 watching
- Trajectory and outliers — 4 watching
- Per-video lifecycle — 6 watching
- Engagement — 8 watching
- Channel shape — 8 watching