The free window has four shows in it, and CBS Studios made all four

The decision this is for

A network entertainment president setting next season's window policy, with the subscription P&L on the other side of the table. If the window is a discovery instrument, it belongs anywhere discovery is the problem — including a returning show that has lost audience and needs resampling.

Tracker is why that question has two answers and only one of them is a choice. It is the most-watched series on the network. It is also produced by Disney's 20th Television, which puts it outside the one condition that holds without exception across the slate. A policy debate about whether to give Tracker away is a debate about something the network cannot do unilaterally.

Eleven of the fifteen titles are returning. Ten of those eleven are out of the window, and the eleventh is a prequel launch.

Every title on the slate

Season numbers and the slate from TV Guide and CBS's own 2026–2027 schedule release. Window status is verified by channel id, not by title match. A title counts as in only on an official full episode on a CBS-owned channel — trailers, clips, sneak peeks and third-party uploads are not the window, and neither is an affiliate.

Four in, ten out, one unexamined. The in-column and the CBS-Studios column are the same four titles.

Each condition, scored on its own

A two-condition rule that is only ever scored as one rule hides which half is carrying it.

The clean eight-for-eight on the rights condition carries almost no information on its own. All eight outside-produced titles are established, and the launch condition already predicts every one of them is out. The rights condition is consistent with every observation here and tested by none of them.

The one cell where the two conditions do disagree is small and split: among CBS Studios titles at season three or later, Origins is in while NCIS and Elsbeth are out. A prequel-launch reading explains that. So does a franchise-channel reading. Three titles is not enough to choose.

What the window reaches: unresolved, and that is the result

The first question registered was not which titles are in the window but what the window reaches — an audience the paid tiers do not, or the same audience arriving free. Four passes could not answer it, and the registered stopping rule is why this page says so instead of estimating.

The two explanations differ on who is watching. Every instrument available measured how many. The test needed commenters to describe how they reached the show, and across 782 comments in three franchises not one access term reached any top 20. The registered floor was 30 per franchise row. The arithmetic ceiling on any unlisted term was 8 of 300, 9 of 264 and 8 of 218.

What the corpus was about instead:

Renewal anxiety and show quality, not access. That is a real finding about a different question, and it is registered separately rather than folded in here as though it were the answer to this one.

The sizing, and the figure that is not a finding

Across four titles, official full-episode views ran 2.9% to 4.0% of the linear audience. On CBS's own channel, promotion outran product by roughly twenty to one: a teaser at 6.95 million views against 342,000 for all three episodes combined.

The raw column looks like a 59% collapse and is not one. Episode 1 had been up three times as long as episode 3 when the rows were read. Divided by days since upload, the rate is flat. The decay figure was caught before it was published and is kept here because a figure that nearly shipped is worth more on the page than off it.

What we registered before the data was pulled

Three falsifiers across two briefs. One was withdrawn as unusable, one could not be resolved by any available instrument, and one resolved on half of itself.

Not mutually exclusive. The live reading is launch need and the rights boundary together, with the boundary as a constraint rather than a policy.

Nothing in the first brief's priors moved at all. That is the correct outcome of a study whose falsifier turned out to be unresolvable on the available instrument, and it is printed rather than smoothed over.

Three things this study got wrong first

All three are in the failure register. They are here because a study that publishes only its final reading is asking to be trusted on the part nobody can check.

1. A falsifier both candidates predicted. A magnitude test was registered to separate two explanations that differ on composition. It ran, both ends matched, and the result was written up as a verdict. Caught by the client on reading the conclusion, not by the instrument. The prior that should have prevented it was available before the brief was written: CBS primetime carries a median viewer age of 67.8, the oldest of the five broadcast networks.
2. A census one unit short of the census. The stopping rule was completeness and the population was written as “roughly fifteen titles.” Three titles had been examined, four were known to be in the window, and the spec built to finish the job carried seven rows. Seven plus seven is fourteen. The spec validated, 37 tests passed and every row returned: completeness is the one property the test suite cannot check, because it has no model of the population.
3. A reading stated as a finding. “The window is a season-one instrument” was reported before the producer column had been totalled and while one season-two title was still unexamined. It was an unregistered sharpening of a registered hypothesis, and the producer cross-tab changed it within the hour.

All three were design-stage errors, not execution errors. Every spec validated, every test passed, every row returned and the arithmetic was right each time. The register now carries more failures in research design than in execution, which says the instruments are ahead of the questions pointed at them.

What this study cannot tell you

• One title of fifteen is unexamined. Harlan Coben's Final Twist, season 2. It cannot move the registered falsifier, whose clauses are written on season one and season three-or-later. It can break the rule above, which claims to fit every title.
• The two conditions of the rule have never disagreed. Every first-season title on this slate is a CBS Studios title, so the cell that would separate “CBS gives away new shows” from “CBS gives away shows it owns” does not exist in the population. Separating them needs another season or another network.
• Shortlists, not censuses. Twenty candidates considered and ten returned per row. A dedicated per-title query is strong evidence of absence and is not proof.
• Two studio credits are inferred and two are unverified, and they are marked as such in the table. Producing credits visible from outside are coarse, which the brief flagged before the pass ran.
• Inclusion is intent, not effect. Everything here says what CBS believes the window is for. Nothing here says whether it works, and four passes have now established that the comment corpus does not carry the audience composition that question needs.
• Views are not viewers and comments are not either. Commenters are a self-selected fraction of whoever watched, and the fraction is unknown.
• A snapshot that dates itself. A title absent on November 5 may be added in December, and the window may close.
• One network, one season, one country. Nothing here says what ABC or NBC do, or what CBS did last year.

How this page changes

A standing study, not an article. Corrections are published in place and dated, never applied quietly.

Related reading, free and in full

• Streaming research — opening demand and the decay curve that follows it
• A 94% audience score, and they were talking about Batman
• Every call we registered, including the ones we got wrong