· The Typestream team
Answers that show their work
Why every number in Typestream opens to the articles behind it, and why we compile each question instead of scoring each article once.
Most media monitoring gives every article one fixed score: positive, negative or neutral, decided once, when the article arrives. That score can't tell you whether an article is critical of your product or of the whole industry, whether it blames your CEO or your supplier, or whether it's about you at all. For anything else, you need an analyst.
We built Typestream around a different idea: compile each question on demand, and show the work.
Questions, not scores
When you ask Typestream something, it doesn't look up a score. It works out what needs judging for each article to answer your question, and what can simply be counted.
"Which outlets are most hostile to us?" needs nothing new: tone toward you is judged for every article as it arrives, so the answer is a grouping and a sort. It's instant and free.
"Of the critical articles, do they blame the design, its durability or the company?" is different. That's a new question, asked of each critical article by JEV, TypeSafe's decision model, with a probability for each answer. Code then adds those up, with the uncertainty shown. Dates, counts and amounts never go to a model; they come from the articles' own data.
Every number opens
An answer you can't check is an answer you can't use in front of a client or a chief executive. So in Typestream, every number, point and bar opens to the articles behind it, each with what was judged about it. The count at the top of that list always matches the number you clicked, because both come from the same articles and the same filters.
Under every answer is a quiet line, How was this made?, that says in plain words what was asked, how many articles were in scope at each step, what was reused, and what it cost. One layer further down, Details has the exact questions and every article's probabilities. Nobody needs that layer. Anyone can reach it.
Calm until it matters
Showing the work doesn't mean showing everything. Typestream stays quiet until something changes, and then says what, by how much, and where to look. The AI does the setup, sorts the borderline articles itself, and asks you only when a name could mean two things, or before it spends more than you've agreed.
If you'd like to see it with your own subjects, start a trial or book a demo.