PhoneTracker

How this actually works

No person writes anything on this site. A program reads the news, weighs it up and publishes the result. Here is the whole thing in plain English, including the bits it gets wrong — because a tracker that only tells you its good side is not worth reading.

  1. We read the news, every day

    A program fetches the tech sites and manufacturer newsrooms, politely and on a schedule. Almost everything gets thrown away immediately — only articles that actually say something about timing go any further. Roughly one in eight survives.

  2. We pull out the actual claim

    An AI reads each surviving article and writes down what it says: which device, which event, which date. It must also copy the exact sentence it got that from. We then check that sentence really appears in the article — so if the AI invents something, the claim is thrown out before anyone sees it.

  3. We start from the calendar, not the gossip

    Before a single rumour is counted, we already have a decent guess: phone makers launch on rhythms. Roughly the same gap between models, roughly the same time of year. Recent launches count for more than old ones, so a genuine change of schedule gets picked up quickly.

  4. Rumours nudge that guess — by as much as they deserve

    A report is not a fact, it is a measurement with an error bar. If a site has historically been three weeks early, we shift its claims three weeks later before we believe them at all. Vague claims like "early next year" are treated as a stretch of time, not a specific day, so nobody is rewarded for being vague or punished for being honest.

  5. Twenty copies of one leak is still one leak

    When a rumour goes round, we group the articles that are clearly retelling the same story and count the group once. Without this step, a single viral rumour would produce a very confident and very wrong date, every single time.

  6. When the phone appears, everyone gets marked

    The moment a device is officially revealed, every claim ever made about it is scored against what really happened — and compared with what the calendar alone would have said. Beating the calendar is the only thing that earns points. That score is what decides how much we listen to that source next time.

  7. We cannot quietly change our mind

    Every prediction is written down once, fingerprinted, and never edited. Changing our mind means adding a new line, not rewriting the old one — so the full history on every device page is genuinely the history, and anyone can re-check the fingerprints.

What the words on this site mean

Everything here is written to be understood without a maths degree. Just in case:

The five bars

How sure we are. Five bars means we know the week. One bar means it could still move by months. It is the honest width of our guess, turned into something you can read at a glance.

"Reveal"

The day the phone is shown to the world at an event or in a press release. It is usually a week or two before you can actually buy one.

"On sale"

The day units really reach buyers. We predict this separately, and a phone can never be shown going on sale before it is revealed.

"Separate reports"

How many genuinely independent stories are behind a date. Fifteen sites rewriting one leak counts as one, not fifteen.

"Moved 4 days later"

Our date changed since we last checked, and by how much. Red means later than we thought, green means sooner.

"Almost certainly between…"

Eight times out of ten the real date lands inside that stretch. It is deliberately not a single day, because a single day would be a lie.

"From £899"

A price somebody reported, in the currency they reported it in. We never convert between currencies — the moment we did, the figure on screen would be one no article ever printed. Where several sources disagree we show the range rather than pick a winner, and where a price is only known for a bigger storage model we say so instead of passing it off as the starting price.

"Worldwide" and "one region"

Two different questions, and we answer them separately. The list of countries is simply the places reports have actually named. The label above it — worldwide, one region, or not known yet — also says what it is based on: reports about this exact phone, or just how that product line normally launches. The second is a reasonable guess about a brand, not a finding about a phone, and we never dress it up as one.

"Already out"

Once a device has been revealed and has gone on sale, it stays on the front page for a fortnight and then moves to the history. Nothing is deleted — our old guess stays attached to it, so you can check how close we got.

Where we get it wrong

All of these are real. None of them are secret.

Is the tracker healthy today?

Checked automatically before anything is published. If one of these looks badly wrong, publishing stops on its own rather than putting out nonsense.

Articles read today Fine

870

How much the reader found to look at.

Date claims found Fine

144

Articles that actually said something about timing.

Sites being watched Fine

38

Feeds checked on every run.

Devices being tracked Fine

37

Products with a live prediction.

Reading errors Fine

5.3%

Articles the reader choked on. Lower is better.

AI errors Fine

1.2%

Calls that failed and were retried. Lower is better.

Made-up quotes caught Fine

12.5%

Claims thrown away because the quote was not really in the article.

Typical uncertainty Fine

53 days

How wide the average prediction is, in days. Lower means sharper.

device market rows Fine

8

devices with market scope Fine

43

region vocabulary miss rate Fine

0.0%

spec catalog pages Fine

0