What agencies ask before they bring a file.

An SPSS .sav file, which carries the question wording, the answer labels and any user-missing codes, so nothing has to be typed in. Or a CSV with one row per person and one column per question. A CSV with no labels of its own can be joined by a two-column labels file, column name and question wording, and by a pairs file naming the look-alike questions.

A weight column in the file is applied and named on the card. The weighted figure leads and the unweighted one stands beside it. Significance tests on weighted figures run on the effective base. How weights are counted.

On the Waves tab, name the wave and drop the next file. Each wave stays its own file. A question with no column on a wave has no figure on that wave, and the Waves tab lists which questions were asked on both, dropped, or added.

No. You upload in the desk, over HTTPS, to the EU. We do not accept files by email, and we ask you not to attach one when you write to us.

The way your client says it on the phone. “How many switched supplier last season?” binds to the question about switching supplier, in the past tense. Add “by region” to cut it. Add “among those who changed supplier” to narrow it. Ask for “the mean” or “the average” on a scale and you get the mean.

One of four reasons, and the card says which. The question names something the study never asked about. No field in the study is a clear match. Two fields match too closely, in which case you are asked to pin one. Or the question is in the study but was not asked on that wave, in which case the card names the look-alike that was asked and the percentage it would have given. The refusal rules.

Your decision, remembered. When two questions score too close together, the desk shows both and asks you to pick. Pin one and every later ask with that wording goes to it. The card says it was pinned, and pins are listed on their own tab.

Yes, from the studies page. Each study answers on its own tile, with its own waves and its own receipt. Two studies are never blended into one percentage.

That the figure on the card came from that study, that wave, that question, that field and those rows, counted without a language model. Run the same question on the same file later and the hash matches. Change a row and it does not. Copy the receipt from the card and anyone can re-hash it on the desk, without a login. What is in a receipt.

Every column in a cut gets a letter. A cell lists the letters of the columns it is higher than at the 95% level, tested with a two-proportion z for shares and a Welch z for means, on effective bases when the figures are weighted. The footnote under the table says which bases were used.

A 1–5 or 0–10 question is read as a scale. The card gives the favourable box, top 2 on five points and top 3 on 0–10, and the mean, both on the people who answered on the scale. A “don’t know” sits outside the scale and is left out. The reading band then tells you what the figure would be with the don’t-knows in the base, and whether that changes the picture.

The desk read the routing back from the file. When a question was put only to people who gave a particular answer earlier, the pattern of blanks shows it, and the card says who was asked, in a sentence. How routing is read.

No. Every figure is arithmetic on the rows of your file, and every sentence on a card is built from the file’s own wording. Every receipt records model=none. Nothing you upload goes to ChatGPT or any other model provider.

One page with your agency’s name on it. On it, the cards you chose to send, each with its figure, wording, base and receipt. If you open a study to them, a box where they can type their own question and get an answer under the same rules. Never the file, never the codebook. See a client page.

Every question your client types is listed on your client page desk, with what came back and when. If one came back too close to call, you can pin it from there.

By the link you send them, the way you already talk to them. Or, if you list their work email, they sign in once on the brand side and find every page they have been given, from every agency that uses Fieldlock. A new link closes the old one, and you can close the page at any time.

It runs the real engine in your browser on Harborline, a made-up food packing study with two seasons. Ask it anything. You will see a counted card, a refused card and a receipt, exactly as they appear on the desk.

Yes. Drop a .sav or a CSV into the “try it on your own file” desk on the homepage. Your browser reads it and counts it in the tab. Nothing is uploaded. Drop a second file and you have two waves.

It does not keep your studies between visits, it does not remember pins, and it has no client page. Those live on the desk.

The desk is €2,400 a year for one login with every study you run. With a page for one client it is €3,600 a year. Each further client is €1,200 a year. Pricing.

Write to us from your work email. The desk is invite only, so we open a login for you and set up your first study with you on a call, in an afternoon. Do not send the file by email; you upload it in the desk.

Database, sign-in and file storage are on Supabase in Frankfurt. The application is hosted on Vercel. Nobody else processes your study data, and no language model provider is involved. Security and data.