Research and Data

Most writing about the social media growth industry, this site’s own earlier work included, restates what the platforms already publish. There is a reason for that: the industry does not release figures, the platforms do not release the figures that would settle the arguments, and the people who do know what a service delivers are the people selling it. So the writing circles the same documentation, and the disagreements never resolve because nobody brings a number.

This page lists what we have measured ourselves, and the rules we measure under. Everything here is published with the material it was derived from — the raw capture as it arrived, the scripts that read it, and the aggregate figures those scripts emit — so that a reader who doubts a claim can rerun the arithmetic instead of weighing our credibility against somebody else’s.

The standing method

These are commitments, not aspirations, and they constrain what we are able to publish. Where a study departs from any of them, the departure is named in that study’s method.md rather than left for a reader to notice.

• Public pages only, unauthenticated. No accounts are created, no orders are placed, no forms are submitted, no logins are used. Every page we read is one any visitor can open. This is a real limit and it excludes the most interesting questions — what a service actually delivers after you pay for it cannot be answered this way, and we do not pretend otherwise.
• robots.txt is respected. A host that disallows the path we want is skipped, and the skip is recorded in the fetch log rather than worked around.
• Rate limits, and an honest user agent. At most one request per host every two seconds, identifying itself as Novamya-Research with a link back to this site. Collection takes as long as it takes.
• The raw capture is retained and published. Live pages move; several catalogues in the September survey had changed within days. The stored copy is the source of record, and it ships with the study.
• Every published figure is reproducible offline. The analysis scripts run against the retained files with no network access and rebuild every number in the article. If a figure cannot be produced that way, it does not go in the article.
• Failures and exclusions are recorded, not dropped. A host that timed out, a catalogue that could not be parsed, a row that failed a comparability filter — each is data about the market, and quietly discarding it is how a survey flatters itself.
• Our own commercial relationships are in the sample and labelled. Novamya is published by Utta, which builds and maintains the websites of two panels in this market. Those two panels appear in our surveys, named, on the same terms as everyone else. Leaving them out would remove the one relationship in the sample we can verify from the inside. The disclosure page sets out the arrangement in full.
• We do not rank, rate or recommend providers. Not our publisher’s, not anyone’s. A league table produced by a company paid by two entrants would be worthless to you.
• The finding is about the market, never about a named business’s honesty. Our data can show that two storefronts share a codebase, or that a price is an outlier. It cannot show intent, and we do not write as though it can.

The catalogue

Licence, and how to cite it

Everything under novamya.com/data/ is released under Creative Commons Attribution 4.0 International. Reuse it commercially, republish it, build on it, correct it. The one condition is attribution.

The licence is stated because data without one is data nobody responsible can reuse. A survey published so that others can check it should not then make checking it a legal question. Each study’s page carries its own citation line; the general form is the study title, the year, the word Dataset, and the URL of its data directory.

If you need a different cut

If you are a researcher, a journalist or a regulator and the published shape is not the shape you need — a different serialisation, an intermediate parse, an earlier or later capture, or a variable we collected but did not report — write to editor@novamya.com and we will produce it. There is no charge and no form. We would rather the numbers were used than admired.

The same address takes corrections. If a figure here is wrong, it will be fixed, dated and recorded on the corrections page, which already lists the four figures an internal review caught in the first survey after publication.

The questions this method cannot answer

Reading public pages tells you what a market advertises. It does not tell you what it delivers, and the gap between those two things is the whole subject. Being specific about the gap is more useful than implying it does not exist, so here is what we currently cannot settle and why.

Whether an order arrives, and what arrives. This is the question every reader actually has, and no amount of catalogue-reading touches it. Answering it honestly means placing orders, disclosing that money changed hands, and accepting that a handful of transactions is a handful of anecdotes rather than a measurement. A study designed to survive that objection needs a sample large enough to say something about a distribution, on services chosen before the money is spent rather than after, with the wallet disclosed in the method. We have not done it.

What happens after thirty days. Retention is the claim the industry sells hardest — “non-drop”, “lifetime guarantee”, “refill” — and the one no listing evidences. A listing’s refill window is a promise about a process, not a measurement of an outcome. Testing it requires the orders above, plus a month of patience and a way to distinguish a service’s losses from the platform’s own routine account removals, which run continuously and affect channels nobody has bought anything for.

Who operates whom. Public infrastructure can show that two sites share a codebase, a certificate or a subnet. Shared software shows a shared vendor, and nothing more: the same off-the-shelf panel script runs thousands of independent businesses, exactly as thousands of independent shops run the same shopping cart. Shared hosting narrows it. Neither establishes common ownership, and we do not write as though they do.

What any platform’s enforcement actually keys on. Telegram, Instagram and YouTube publish policies, not thresholds. Every confident number in circulation about how many purchased followers triggers what — and they circulate constantly — is somebody’s inference presented as a specification. Where this site describes enforcement, it describes the published policy and labels the rest as inference.

Cadence, and what happens to old figures

A price captured on one date is a fact about that date. It is not a fact about the market, and it stops being useful the moment it is quoted as one. Two consequences follow, and both are visible in how these pages are written.

First, every figure on this site that comes from a capture is stated with its capture date attached, in the sentence, not in a footnote. Second, a study is worth repeating only if the repetition is comparable — same basket, same filters, same window discipline — so the scripts are published partly so that the next capture is forced to use them. A second capture that quietly changed its own definitions would produce a change we could not attribute to the market.

Superseded figures are not deleted. When a later capture replaces an earlier one, the earlier study keeps its page, its data and its date, and says which study replaced it. A publication that overwrites its own history is asking to be trusted about the present on the strength of a record it has edited.

What we will not do

We will not buy access to a dataset and republish it as our own measurement. We will not place covert orders and write up the result without saying an order was placed — if that study is ever done, the transaction will be disclosed in the method before the findings. We will not name a business as fraudulent on evidence that shows only that it resembles another business. And we will not publish a number we cannot hand you the workings for.

Those constraints cost us the more sensational versions of several of these studies. They are the reason the ones we do publish can be checked.