Sentioscope

Speech and voice controls, as each vendor documents them

How to read a cell that records nothing

A blank here means the documentation read for this register does not address the point. Several products plainly do the thing anyway. The blank is a statement about a document, and the useful response depends on why the document is silent. As of 2026-09-12.

Four reasons a cell ends up empty, four responsesAn absence on a hosted page is a refusal to commit. On published weights it is an unmade measurement. Delegated, it is a supply-chain audit. Where a short list would have been easy, it is a risk.Why is this cell emptyThe weights are publishedMeasure itRun the model, record theresult and the dateSomebody else publishes itAudit the supplierThe answer exists, onelevel down the supplychainThe list would be shortTreat it as a riskLikeliest reading is acapability still movingA blank is a statement about a document, not a verdict
Fig. 1 Filling a cell from plausibility would reward whichever product a reader already trusted, which is the failure this approach exists to avoid.
Why a cell ends up empty, and what each reason implies. Recorded 2026-09-12.
Why it is blankWhat it impliesWhat to do about it
Answerable by running itThe weights or a free tier are availableMeasure it, and record the date
Delegated elsewhereAnother company publishes the answerAudit the supplier, not the vendor
Outside the documentA reference lists parameters, not resultsAsk, or test one short shot
Probably still settlingA short list would have been easy to printTreat it as a risk, not a gap

Inclusion rule. Reasons a documented field goes unfilled, as observed across the pages read for this register. Order. Alphabetical by reason.

1A blank is not a verdict

Filling a cell from plausibility would reward whichever product a reader already trusted, which is the failure this whole approach exists to avoid. So a field stays empty until a vendor writes a sentence, however obvious the answer seems.

The cost of that discipline is that a strong product with a thin page looks sparse. The benefit is that every row can be compared with every other, because they were all read the same way.

2The reason behind a blank changes the remedy

An absence on a hosted commercial page is a refusal to commit, and it stays absent until the vendor changes its mind. An absence on a model card with published weights is a measurement nobody has taken, and anybody can take it.

An absence that has been delegated is different again: somebody publishes the answer, just not the company being evaluated. That turns a gap into a supply-chain audit, which is more work and is at least possible.

3Treat a suspicious blank as a risk with a date on it

Where a short list would have been trivial to print and was not printed, the likeliest reading is a capability still moving. Nothing can be concluded from that, and it belongs in a risk register rather than a comparison table.

The practical step is to note what sentence would settle the question and ask for it directly. A vendor unwilling to put one line in writing has told a production something that no amount of testing will.

4Where the published detail is logged

What is described above is how the techniques work in general. Which models document which of them, in whose words, is kept in the speech controls table with a date on every field.

A mechanism note rather than a documented field. Nothing here is attributed to a model, and nothing here is a claim about one. The sourced material is on the speech controls table. Related: Counts and lists, Two speakers.