Weekly report
What real users' questions did to Tattva's reading over the last week, per project, and the capability gaps for the engineering team. Platform admins only.
The weekly report shows how Tattva read the questions people asked: where its reading needed fixing, what it asked back, what it turned away, and which words it could not place.
Where to find it
Open Settings › Weekly report, or go to /settings/weekly-report. Only platform admins (superadmins) see the tile and can open the page; anyone else gets "page not found".
Pick 7 days (the default), 14 days or 30 days. Questions asked by the automated style test's account are left out and only counted.
Capability gaps
The first table lists words people used more than once that no definition, key, input, convention or skill could cover, for example a kind of analysis Tattva does not do. They are engineering work, most asked first. Each row shows the words, the project, how many questions and people used them, why no entry could cover them, the reading problems those questions hit, and an example question.
Words an analyst can cover are not listed here: they go to the project's analysts under Ontology › Maintain › Feedback › Asked but not answered › Words not understood (see Work answer feedback). A model sorts the words, given the project's definitions, columns, inputs, conventions and skills; words go to the analysts only when the entry the model names exists in the project.
The report, per project
Click a project to open its report:
| Section | What it shows |
|---|---|
| Slot × problem cells | Each kind of reading problem (a period not in the question's words, a period missing, a filter value the column does not hold…): how many questions it was found in, how many the automatic repair fixed, how many still had it, how many were asked about and answered |
| Repairs | Each question whose reading was repaired: the parts changed, and the problems before and after |
| Asks | Questions Tattva asked back, how many were answered, asked again without an answer, or left; per kind of question asked |
| Refusals | Questions turned away, by reason and subject |
| Words | Words Tattva could not place, words still not covered after the repair, and filler words, grouped however they were written |
| Dropped | Breakdowns, filters and metrics read from a question but not applied |
| Per definition / Per user | The same counts for each definition and each person |
An ask counts as answered when the person picked or typed an answer in the clarification card.
The same report as text for a review: npx tsx scripts/weekly-report.ts --days 7 in apps/web (add --classify to sort the recurring words; it writes nothing unless --store is given).