Skip to content

Open renewal decision

Will XO, Kitty be renewed for Season 4?

Episode Club puts XO, Kitty’s renewal odds at 43% as of 19 September 2026. Netflix has announced no decision.

That figure is arithmetic, not a vote and not a tip. Episode Club computes it from four public signals; no members predicted it and nobody inside Netflix was asked. Every input, its weight and its source are listed below so you can disagree with a specific one.

All shows on the renewal tracker · How accurate have these predictions been?

43%

Leans cancel

550 · even95

Modelled probability of renewal

Evidence coverage 1.00 · 4 of 4 signals available · display clamped 5–95%

Evidence coveragefloor 0.55
Recalculated 19 September 2026Sources verified 11 August 2026Algorithm v3

No members have added a forecast yet.

How this number is made

Episode Club’s renewal model puts this at 43%. No member has added a forecast, so the published figure is the model on its own.

How we calculate this →

Model

43%

Make your call

No member forecast yet

Results in

43%

Forecast on this yourself

Episode Club members can add a forecast to this question. It is free, it takes one drag of a slider, and your answer is never shown to anyone else.

Create a free account

Free game · points have no cash value

— 39 days ago

What is on record

Four signals, in the order they carry weight. Each one links to the page the fact was read from.

  1. 01

    Network silence

    Netflix has said nothing about another season in the 170 days since the finale on 2 April 2026. That is measured against a 90-day median wait for a decision at Netflix.

    0.20

    40% weight used

    +8.1 pts

  2. 02

    Audience performance

    Netflix global Top 10 (TV, English): 3 weeks in the published chart, peaked at #1, opened 9% below the previous season.

    0.62

    30% weight used

    +18.6 pts

  3. 03

    On-record statements

    3 on-record remarks are counted, weighted by who said them and faded as they get older.

    0.50

    20% weight used

    +10.0 pts

  4. 04

    Critic score

    Rotten Tomatoes Tomatometer (season 3): 59 out of 100, from 17 reviews.

    0.59

    10% weight used

    +5.9 pts

Source ledger

The receipts, in the open.

Every figure above traces to one of these. Each entry says which signal it backs, so a citation under the wrong claim reads as wrong rather than merely vague.

  1. whats-on-netflix.com · checked 9 August 2026

    ‘XO, Kitty’ Season 4 Finally Gets Long Awaited (But Disappointing) Netflix Update

    Backs Network silence.

  2. netflix.com · checked 11 August 2026

    Netflix Top 10 — TV (English), global weekly rankings (official Tudum data)

    Backs Audience performance.

  3. yahoo.com · checked 9 August 2026

    Anna Cathcart breaks her silence on XO, Kitty season 4 renewal status

    Backs On-record statements.

  4. deadline.com · checked 9 August 2026

    ‘XO, Kitty’ & ‘Man On Fire’ Still In Limbo After Making Top 20 In Netflix’s Semi-Annual Report

    Backs On-record statements.

  5. rottentomatoes.com · checked 9 August 2026

    XO, Kitty: Season 3 — Rotten Tomatoes

    Backs Critic score.

The silence clock

A network that has said nothing for less time than it usually takes to decide has not told you anything yet. This figure falls on its own as the gap widens, with no new fact — that is the clock, not the show.

How the model’s 43% was calculated

Every row, including the ones that contributed nothing. A signal with no data is left out of the average rather than scored as a coin flip, and the weights of the remaining signals are rescaled to fill the gap — that rescaling is the "weight used" column.

Weight available — 1.00 of 1.00

The hatched gaps are signals with nothing on record. The remaining weights are renormalised over what is present — a missing input is never quietly scored as zero, which would read as a bad result rather than an absent one.

Contribution to the published figure

The inputs behind the modelled probability, with their weights and points
SignalScoreWeightWeight usedPoints
Network silence0.200.400.400.081
Audience performance0.620.300.300.186
On-record statements0.500.200.200.100
Critic score0.590.100.100.059
Total0.426

The points column adds up to 0.426, which is the model’s 43% after rounding. Coverage was 1.00 of 1.00, so the weights above were rescaled by dividing each one by 1.00.

The method, in full

Opacity is the moat for everyone else in this field; it cannot be ours. Every weight, every curve and every refusal rule is printed below, so a reader can disagree with one specific thing rather than with the vibe of a number.

The four signals and what each is worth
SignalWeightWhy that much
Network silence0.40The heaviest signal, for two reasons: every show has a finale date, so it is never missing, and it is the only one that moves on its own. Silence starts as mild good news — a network that has decided to cancel usually says so quickly, because that frees a slot — and decays from there.
Audience performance0.30The closest public proxy to what the network actually decides on. Not the heaviest, because the chart is censored — a Top 10 hides everything below tenth, and Apple TV+ and HBO publish nothing at all.
On-record statements0.20The only forward-looking signal — everything else describes the past. Held down because the same field has to hold a network president’s answer and an actor saying "we would love to do more", which is what every actor says. Who spoke is weighted separately: an executive counts 1.00, a creator 0.60, trade reporting 0.50, a cast member 0.30. Remarks lose half their force every 120 days.
Critic score0.10Deliberately small — a tiebreaker. Acclaim and renewal correlate weakly in both directions: the list of 90%-rated shows cancelled after one season is long, and the list of critically dismissed procedurals renewed for a decade is longer.

How a raw fact becomes a score

Each signal is turned into a number between 0 and 1 before the weights above are applied. These are the four transformations, in full. They are judgements about shape, not measurements, and they are the part most worth arguing with.

Network silence
score = 0.05 + 0.57 × 2^(−days since finale ÷ median decision days). Starts at 0.62 the day the finale airs and halves its distance to a floor of 0.05 for every median-length wait that passes. The halving is the one piece of arithmetic here with a derivation rather than a judgement behind it: the median is by definition the point at which half of that network’s decisions have landed, so a show still waiting there has fallen out of the faster half — one bit of bad news, one halving.
Audience performance
The mean of whichever of three measures we have. Weeks in the chart: 1 − 2^(−weeks ÷ 3), so 3 weeks scores 0.50, 6 scores 0.75, 12 scores 0.94. Peak rank: 1 − (rank − 1) ÷ 10, so #1 scores 1.00 and #10 scores 0.10. Season-over-season change: 0.5 + change × 1.5, so flat is 0.50 and a one-third collapse reaches 0.00. Measures we do not have are left out of the mean rather than filled in at 0.5.
On-record statements
Each remark scores 0.85 if it points towards another season, 0.50 if it says the question is open, 0.15 if it points against — never 1 or 0, because people are optimistic on the record about shows that then die. Each is weighted by who spoke and halved every 120 days. The weighted average is then pulled towards 0.50 by one unit of doubt — a pseudo-remark of weight 1.0 sitting in every sum — so a single remark moves the score part of the way rather than setting it outright. One network president speaking today carries weight 1.0, so even that moves it only halfway. Once the total weight of a show’s remarks decays below 0.1 — a tenth of that — they stop counting as a signal at all, rather than lingering as a neutral score that no longer says anything.
Critic score
The published score divided by 100, untransformed. Any curve here would be invented, and the weight is 0.10 — the honest move is the cheapest possible mapping, one a reader can check in their head, and to argue about the weight instead.

The rules that stop a number being published

  1. Missing evidence is missing, never neutral

    A signal with no data contributes nothing at all — not weight, not points. If the signals we do have account for less than 0.55 of the total weight, no percentage is published. Scoring an unknown as 0.5 would let a show with one fact on file print a confident-looking figure.

  2. A figure nobody has checked comes down by itself

    Past 45 days without a check, the page says so in the open. Past 90 days the percentage is removed automatically. The failure this guards against is not a stale page — it is a confidently wrong one, still saying 62% for a show that was renewed a week ago.

  3. An announcement is a fact, not a forecast

    Once a decision is public the percentage disappears for good and the page carries the announcement and its source instead. A decided show is never given odds again.

  4. No certainty in either direction

    Modelled figures are held inside 5%–95%. Four public signals cannot justify 0% or 100%.

Who maintains this, and how to correct it

The renewal tracker is built and maintained by Episode Club. Facts are read from their sources by hand; the arithmetic re-runs on every page load, so the figures move daily even when no new fact arrives. There is no editorial board, no expert panel and nobody inside any network — it is a published formula over public facts. Every page prints the date its sources were last opened, and a figure is removed automatically once that date is more than 90 days old.

If a fact here is wrong, or a decision has been announced and this page has not caught up, tell us. We will fix it and record what changed in the changelog below: corrections@episodeclub.com

Method changelog

A percentage is only interpretable next to the method that produced it, so every change to the arithmetic gets a version and a line here. Data corrections are listed too, because a figure can move without the model changing at all — and on this tracker, most of them have.

  1. 9 August 2026

    Version 3 — evidence that has decayed to nothing stops counting

    Remarks lose half their force every 120 days, but a remark still counted as a whole signal no matter how far it had decayed — so an old quote that could no longer move a figure was still enough to let one be published. Nine Perfect Strangers was carried over the evidence floor by a single non-committal line from July 2025 worth about 3% of its original weight. Statements now stop counting once their combined weight falls below a tenth of one network president speaking today. One show left the board: Nine Perfect Strangers is a facts page again, with no percentage.

  2. 9 August 2026

    Audience figures moved to Netflix’s own published data

    Every chart figure had been typed by hand from trade coverage. Checked against Netflix’s official weekly Top 10 dataset for the first time, five of the six Netflix titles were wrong: Man on Fire had 5 weeks on the chart, not 2; Sweet Magnolias’ whole run and XO, Kitty’s peak position were on record as unknown; Ransom Canyon was recorded as #1 in the US when it never went above #2; and BEEF had a chart run we had missed entirely, which had been keeping its page below the evidence floor. Published figures moved by up to 13 points. The figures are now read from the dataset, and the test suite fails if this page and the dataset ever disagree. No change to the algorithm, so the version did not move.

  3. 9 August 2026

    Facts corrected and found away from the chart data

    The same audit re-read every non-chart fact. BEEF’s finale date was ten days wrong — a chart week-ending date had been recorded as a release date. A trade report was dated two days late, because we had stored the day an aggregator repeated it rather than the day the original ran. Critic scores were found for four shows we had recorded as having none, and a dated on-record quote was found for a fifth; between them those moved Sterling Point from 62% to 66% and took Nine Perfect Strangers from no figure at all to 28%. Our estimate of how long Apple TV+ takes to decide was corrected from a rounded 150 days to the 138 its single source case actually measures.

  4. 9 August 2026

    Source citations, publication dates and a correction address

    Sources used to be listed as bare domain names — six identical rows for six different articles. Every fact now carries its source’s headline, the date that source was published and the date we last opened it. Two dead citations were replaced with primary ones, and two dates were corrected after we found we had recorded a republication date rather than the original.

  5. 8 August 2026

    Version 2 — statements are pulled towards neutral

    A single on-record remark used to score at full strength no matter who said it or how old it was, because with one statement the speaker weight and the age decay divided themselves out. A network president speaking yesterday and a cast member speaking two years ago both scored 0.85. Statements are now shrunk towards neutral by a fixed unit of doubt. Two pages moved: How to Get to Heaven from Belfast fell from 49% to 41%, XO, Kitty rose from 32% to 35%.

  6. 8 August 2026

    Version 1 — first published

    Four signals, the weights above, the coverage floor and the two staleness horizons.

Algorithm version 3. The version is printed next to every figure so an old number stays interpretable: bump it and the numbers move, which is exactly why it is written down. A data correction does not bump it — the model did not change — so the changelog above records both kinds.

Questions people arrive with

Has XO, Kitty been renewed?

No announcement is on record as of 19 September 2026. If one lands, this page replaces the percentage with the announcement and a link to it — a decided show never gets odds again.

Who is predicting this?

Nobody is. There is no expert panel, no member vote and no insider. It is a published formula run over public facts, and it will be wrong sometimes — which is why the method is on the page instead of behind it.

What would move this number?

Time, first: every day the network stays silent, the silence signal decays, so the figure drifts down on its own. Then new evidence — another week on a published chart, a dated remark from someone involved, a critic score. And an actual announcement, which removes the number altogether.

Why is it never 0% or 100%?

Modelled figures are held inside 5%–95%. Four public signals are not enough to justify certainty in either direction, and a model that prints 0% is claiming to know something it cannot.

How often is this updated?

The facts are re-read by hand and the date of the last check is printed above. The figure itself recomputes on every visit, because part of it is simply how long the silence has run.

Sources

Every page a fact above was read from, with its own headline, the day it was published and the day we last opened it. Figures — dates, ranks, counts, scores — are restated here with attribution. Nothing is reproduced at length from any of them.

Episode Club is not affiliated with, endorsed by or sponsored by Netflix, Apple TV+, Rotten Tomatoes or any other company named on this page. Those names appear only to say whose show it is and where a figure came from.

Waiting on this one?

Episode Club tracks what you have watched and opens a discussion room behind each episode once you have seen it. Free, and this page needs no account.

Get started free