Its shape
A report has atitle, an overview and five pages, in this order:
The overview has a
bottom_line, three stat cards and the first_question to ask. Each page has:
state:empty,written, orwaitswhen it can’t be written yet and its method says whyheadline, which states the finding, and alederead: what was read, as figures, such as reviews readquestions: two, each with itsask, itsanswerin a line and up to three statcardsexhibits: up to 12, each a saved question to a number toolquotes: up to 30 quotes, reviews, passages or records it cites, by idmethod: what was read, over which window (from,to), and what was left out and whynames: the companies, brands, places and sources it names, which the check reads as words
A figure and an exhibit
Copy never holds a typed number. It cites a figure in braces,{exhibit.figure}, and the figure comes from an exhibit on the page:
numbers is a number tool, company the target or a confirmed rival, and params the tool’s own arguments. Each time the report is read, PodData asks the tool again and puts its figure in place of the braces, so "Patients rate Northwind {customer-1.average} stars" reads "Patients rate Northwind 4.3 stars". See Number tools for each tool’s figures.
Write it
POST /diligence-reportswithcompanymakes one, with five empty pages. The answer gives itsidandrev.PUT /diligence-reports/{id}/pages/{page}saves one page, withrevandpage; for the overview,overviewis the page and the field. A page is saved whenever it fits the shape, whatever the check says, and the answer is the report filled, with what the check finds now.- A
409on a save means someone saved over the version you read. Read the report again withGET /diligence-reports/{id}, and change only what you were asked. POST /diligence-reports/{id}/checkruns the check and marks the reportcheckedwhen it refuses nothing.
start_report, save_report_page, check_report, get_report. get_report with page gives one page as saved, in JSON, to change only what you’re asked and keep the rest.
What the check refuses
The check refuses a report that would say more than PodData read. Among what it refuses:- a figure in braces that no number tool gave, or braces that don’t hold one
- an exhibit with no evidence, or whose evidence comes through a link no one confirmed
- a company compared that isn’t the target or a rival a person confirmed
- a page that doesn’t say what was read and over which window, or that waits without saying why
- an empty page, a page without its two questions, an overview without its three cards
pages.0.exhibits.1) and why.
Read it
report), the same with each figure in its place (filled), each exhibit’s number tool asked now (answers), each item a page cites looked up by its id, at most 80 words (cited), and what the check says (check). Since each read asks every tool again, a saved copy is that moment’s.
As files
format downloads the report as files, every figure filled as it reads now:
?format=md: one Markdown document, the overview and then each page, with its exhibits as tables, its quotes and how it was built.?format=zip:poddata-report-<company>-<day>.zip, holdingreport.json,report.md, a Markdown file a page underpages/, each exhibit’s rows and its figures as CSV and its answer as JSON underexhibits/,evidence.csv,cited.csv, and aREADME.txtof what each file is.
get_report answers the report as JSON.
GET /diligence-reports lists the team’s reports, newest change first, each with its status and where each page stands.
PodData’s agent
PodData’s own agent writes a first draft, or withmessage makes a change a person asks for: POST /diligence-reports/{id}/runs, from a key, an assistant signed in, or the workspace. PodData pays for its turns, up to the run’s cap (cap_usd), so an assistant asks the person before it starts one. GET /diligence-reports/{id}/runs lists a report’s runs, and GET /diligence-runs/{id} gives one with each agent’s turns and what it cost. Tools: start_report_run, list_report_runs, get_report_run. A run ends done, stopped, capped when it reached its cap with pages left, refused or failed.