Sources and methodology
Where every figure on this site comes from
Nothing here asks to be believed. This page lists each primary document, how it was handled, what was measured from it, and what was excluded. Where an error was made building this archive, it is recorded below rather than quietly fixed.
What the primary documents are
219 screenshots of the Trustpilot profile at trustpilot.com/review/www.theinterstellarplan.com, captured before Trustpilot removed it in January 2023. Every one is published on this site, cropped to the review card. Each review page names the screenshot file it came from.
Two letters from Trustpilot, dated 5 January 2023 and 9 January 2023. Both are quoted on the record page in full rather than shown only as images.
One Internet Archive capture, web.archive.org/web/20220503012022/, taken 3 May 2022. A third-party record outside this company’s control.
One customer message dated 16 August 2022, published as an image and transcribed in full beside it.
Two WooCommerce exports of the company’s own customer review database, 2,714 rows, dated 24 October 2026. Used only for cross-matching. Not published, because every row carries a customer email address and IP address.
How the reviews were transcribed
Each of the 219 screenshots was read and transcribed word for word. Spelling, punctuation, capitalisation and emoji are the reviewer’s own and were not corrected. Where a line was clipped by the edge of a capture, it was not reconstructed from guesswork; it was recovered from an adjacent capture of the same review, or left out and marked.
The 219 screenshots do not contain 219 reviews. Long reviews were photographed across two or three overlapping scroll positions, and a few reviews were photographed twice. Overlapping captures were joined by finding the longest run of text common to both and splicing exactly there, so no passage appears twice. Duplicate captures of the same review were collapsed to one. That is how 219 screenshots resolve to 196 distinct reviews.
Six reviews whose captures could not be joined by text alone were reconstructed by reading the images directly, in correct reading order, and are marked as such in the working record.
How the cross-match was done
The 196 recovered reviews were compared against 2,714 rows of the company’s own store database on two independent signals. Name matching ignores case and punctuation. Text matching requires a 60-character verbatim run in common. 70 matched by name, 77 by text, 102 by one signal or the other.
The remaining 94 are not called into question by this. A customer is under no obligation to review in two places.
What was excluded
One capture shows a greyed repeat review with no readable text. It is counted as a card recovered but excluded from the 196 published reviews rather than padded out.
No email address and no IP address appears anywhere on this site or in any of its published data files. The store exports that contain them were used for matching and are not published.
Relative timestamps such as “2 days ago” appear on six captures. Because the capture date is not recorded, those cannot be resolved to a calendar date and are omitted rather than estimated.
Errors made building this archive
Four, all caught before publication, recorded here because a methodology page that lists no mistakes is not describing a real process.
Duplicated passages. The first joining rule concatenated two captures whenever the first ended mid-sentence. Because captures overlap rather than abut, this duplicated whole paragraphs in eleven reviews — up to 22 repeated sentences in one. Found by auditing every joined review for repeated sentences, not by reading the join count. Replaced with the overlap method described above.
Discarded text. The replacement rule then tested whether the second capture’s first 120 characters already appeared in the first, which is true of every overlapping capture while its tail is still new. That silently dropped real review text from eighteen pairs. Corrected to require full containment.
A claim contradicted by the evidence. An earlier version of the record page said the unrecovered reviews were “the oldest ones, on the last pages.” The Internet Archive capture disproves it: all 95 reviews posted on or before 3 May 2022 are present here. The sentence was removed.
A wrong rating. An earlier version recorded the profile score as 5.0, taken from the TrustBox widget’s share card. The profile page itself reads 4.9 on all 219 captures and on the Internet Archive. Corrected to 4.9.
The data, in machine-readable form
Each contains all 196 reviews with reviewer, country, rating, both dates, formulas named, full text and source screenshot filename.