· SourceSignal
Reading public YouTube signals with their sources attached
A historical Power Tools example separates lifetime view averages, channel concentration and query coverage without treating them as search volume or a prediction of success.
Originally published 15 June 2026. Measurement language reviewed 10 October 2026.
A research number is easier to assess when its source, denominator and collection date travel with it. A collected YouTube result, a public view-count snapshot and an autocomplete suggestion are different observations. Combining them does not make them interchangeable.
This note retains selected figures from the original analysis as historical illustrations. They have not been recomputed or independently verified in this review. The article did not include reproducible queries, an exact snapshot window or a source export. These figures should not be read as the current state of a category.
Similar historical averages, different channel shares
The original Power Tools example compared two collected query-result sets:
| Historical field | bauer reciprocating saw at harbor freight |
circular saw uses |
|---|---|---|
| Reported aggregate lifetime view average | 43,084 views/day | 44,971 views/day |
| Reported channels in collected results | 10 | 9 |
| Reported leading channel's share of the view-average signal | 99.9% — Harbor Freight | 42.5% — 731 Woodworks |
Here “views/day” means a lifetime view count divided by video age at collection, then aggregated for the result set under the original calculation. It is a historical average. The article did not establish the exact deduplication or filtering rules used for these two rows.
It is not a measurement of current daily traffic, query search volume or views attributable to that query. Videos receive views from recommendations, direct links and other sources. The aggregates were within roughly 4% of each other, while the reported leading-channel shares differed. That supports a descriptive comparison of the collected result sets.
It does not establish that one query is “beatable” or a new video would be invisible in the other. Concentration describes the chosen signal's distribution; it does not predict ranking changes, audience response or commercial returns.
Keep observations and interpretations separate
The Research dataset uses public observations with distinct limits:
- Autocomplete suggestions record phrases returned for a seed at collection. They do not disclose query volume, individual users or proven purchase intent.
- YouTube results and metadata record collected rankings, lifetime views and related fields. A lifetime view average differs from the change between two snapshots.
- Detected description links record identifiable URLs. They do not measure clicks, sales or affiliate income.
The earlier version described web results as a current input. That collector and its live metric surfaces were retired in October 2026. This note does not advertise a current web-search dataset. These public Research signals are separate from the product-review evidence used in SourceSignal's main commercial reports.
An exclusion rule does not prove a spike
Dividing lifetime views by video age can give large averages for young, high-view videos. Two million views over nine days is roughly 222,000 views per day on average over that interval. It does not mean the video receives that many views each day afterwards.
The original analysis reported this filtering example:
| Historical reported field | Value |
|---|---|
| Rows excluded under the rule used then | 585, approximately 0.005% of 12.4 million rows |
| Share of the unfiltered aggregate view-average signal attributed to those rows | 3.8% |
| Mean unfiltered lifetime view average of those rows | 392,325 views/day |
These figures illustrate how a small row count can carry a large weight. They are not a current exclusion inventory. The original implementation and denominator have not been reverified, and the rows were ranking records rather than necessarily distinct videos.
The current Research lifetime-average calculation applies an explicit exclusion rule to certain high-view young or incompletely enriched videos. That reduces a known source of instability; it does not establish that excluded views were artificial, a video was irrelevant or the remaining signal measures buyer demand. See methodology for the definition.
Deduplication depends on the question
The original corpus example reported 12,371,436 ranking rows for 1,475,909 distinct videos: about 8.38 rows per video. These are historical reported counts, not today's corpus size.
A video can appear under several queries. Summing its view signal for each appearance would repeat its contribution in a corpus-wide video total. But 8.38 appearances per video does not imply the aggregate is inflated by exactly 8.38 times: repeat counts and view weights can differ between videos.
The aggregation grain depends on the question. A distinct-video summary and a query coverage summary should have different denominators. Attribution makes that choice inspectable; it does not establish that every historical number used the appropriate grain.
Different proxies do not reveal purchasing
The earlier analysis reported this comparison:
| Category | Historical keyword-wording proxy (%) | Historical detected-link proxy (%) |
|---|---|---|
| Running | 80 | 26 |
| Power Tools | 18 | 53 |
| Gaming | 46 | 48 |
Exact denominators and weighting were not specified. These are examples of different classifications, not estimates of viewers wanting to buy or creators converting that interest.
Query wording can omit comparison terms during purchase research; a phrase containing “review” need not indicate a purchase. Links serve several purposes and may be missed by detection. A difference suggests a question for investigation. Neither proxy proves the other wrong; similar percentages do not validate a commercial conclusion.
Source attribution supports review
Ambiguous product words can bring irrelevant videos into results. Collection coverage, keyword selection, classification and exclusions can change comparisons. Keeping source records makes these issues inspectable. It does not remove them or establish that a sample represents all YouTube activity.
The duration-weighting research note gives another example: a retention assumption changes a modelled ranking. The proxy is not observed watch time, and a different assumption does not by itself correct a result to ground truth.
For a useful comparison, ask which records were included, when they were collected, whether videos repeat, what the denominator means and which steps are assumptions. Sales, entry opportunities and future performance need evidence beyond these public fields.
The available category momentum and creator momentum indices use matched-pair changes relative to the tracked market. They answer a different question from these lifetime averages. Research methodology explains the distinction.