Instrument 01
This is a beta. The method is published so it can be checked, not because it is settled. Two questions to Nielsen are outstanding and one of them is load-bearing. Figures may change.
Nielsen's Gauge shows the slice. This shows the pie it came out of, rebuilt entirely from figures Nielsen published itself.
As published
−0.1 pt
Streaming's share of total TV fell to 48.5%, down from 48.6% in May. Rendered out of focus because it cannot be checked. Nielsen distributes The Gauge as an image file.
Denominator restored
+3%
Streaming usage rose about 3% against May. Total TV usage rose 3.1%, so the share slipped anyway.
Both figures are Nielsen's, from the same release. The category grew and was reported as having slipped, because the denominator grew faster.
Share movement is Nielsen's output. The implied index is ours. The band is the honest version of the index. Check vs Nielsen shows the volume change Nielsen states outright in its own prose, and appears only on the rows where it published one.
Week normalisation is struck out because we will not compute it. Dividing by interval length assumes the stated usage change is a raw total. If Nielsen already normalises by day, week or average-minute audience, that correction is not conservative, it is wrong. Asked, unanswered, uncomputed.
Same arithmetic, same controls. This is where share framing does the most damage, because a company can post double-digit growth and appear in the chart as a rounding change. It is also where our method degrades fastest.
Correction to an earlier draft. We called Peacock a method failure because our point estimate missed Nielsen's stated +34%. It was not a failure. Nielsen's figure sits inside the band. The point estimate was never meaningful at that share size. Tubi, Roku Channel and Paramount Streaming are excluded entirely for the same reason.
Nielsen reports on the broadcast month, which ends on the last Sunday of the calendar month. Months run four or five weeks and the pattern shifts year to year, so same-month comparisons are not always like for like.
| Interval | Dates | Weeks |
|---|---|---|
| May 2025 | 04/28 – 05/25 | 4 |
| Jul 2025 | 06/30 – 07/27 | 4 |
| Nov 2025 | 10/27 – 11/30 | 5 |
| Jan 2026 | 12/29 – 01/25 | 4 |
| May 2026 | 04/27 – 05/31 | 5 |
| Jun 2026 | 06/01 – 06/28 | 4 |
May 2025 ran four weeks. May 2026 ran five. Every year-over-year May comparison published this summer set one against the other, and none of them said so.
Days between the close of the measurement interval and the date on the release.
The lag roughly tripled. Live+7 does not explain it: June's interval closed on the 28th and the last night's seven-day window shut around July 5, but the report appeared August 18. April 2026's interval end is derived from the broadcast calendar rather than read from the release, and must be confirmed before that row is published.
They call it a monthly snapshot. Snapshot. A snapshot is a thing that happens in a fraction of a second and then you have it. June closed on the twenty-eighth. The thing showed up August eighteenth.
Fifty-one days. That is not a snapshot, that is a portrait sitting. That is a guy with a plate camera telling everybody to hold still and not breathe while the light does its work.
Somebody will tell you it is Live plus seven. Live plus seven closes July fifth. I can count. I counted for a living for thirty-eight years, some of it for them.
Kowalczyk Ratio on this: 0.71. Boat's still in the garage.
index = 100 × (s₁ ÷ s₀) × (1 + g) · band uses s ± 0.05 (± 0.10 where derived) and g ± 0.0005
Where it comes from. Nielsen publishes share in the chart and the change in total TV usage in the release prose. Everything here is a public Nielsen figure. No licensed data, no Nielsen subscription. This is Rating = PUT × Share, Nielsen's own framework, run from outside because the inputs are no longer published.
Not a volume, not minutes. Nielsen builds streaming from a separately weighted subset of streaming-meter households while broadcast, cable and total usage come from the overall TV panel. Our arithmetic multiplies figures that may not share a denominator. The output is an implied change in Nielsen's Live+7, time-period-based measure. Nothing shorter is accurate.
Not an advertising number. The Gauge covers all TV-screen viewing, ad-supported and ad-free alike. Nothing here is impressions, commercial audience, GRPs, reach or monetisable inventory, and it must never be used as a proxy for any of them.
"Other" is not reported. Nielsen's other bucket is unmeasured tuning, unmeasured VOD, audio streaming, gaming and device playback. It is a residual, not a category. An earlier draft reported it rising 6.6%. That figure was arithmetically implied and substantively meaningless.
Derived cells. FOX's and Peacock's May shares are reverse-engineered by subtracting Nielsen's stated point change from the June figure, and carry double the rounding error. April 2026's interval end date is derived from the broadcast calendar rule.
Reliability by share size. Usable at 10% and above. Approximate between 5% and 10%. Directional only between 2% and 5%. Below 2%, publish the share-point movement and nothing else.
Structural breaks. vMVPD linear streaming was excluded from the streaming category from February 2023. Set-top-box viewing of streaming originals was reclassified into streaming, out of "other," from June 2023. Both moved viewing into the category whose growth is most reported. ARF DASH universe estimates land this fall.
Excluded. February and March 2026, both reported delayed by secondary sources amid a methodology dispute, unconfirmed against a primary source.
Whether Nielsen's stated usage change is period-normalised or aggregate, and how Live+7 viewing crossing a month boundary is credited. Both put to Nielsen, neither answered. Every number on this page depends on the first one.
