How it works
From an official page to a closed action
The same four steps run for every source. Here is what happens inside each, what you see, and what we refuse to do.
1. Monitor
Each source in the registry is an official index page: for example the NBR VAT SRO list or the Bangladesh Bank circular archive. On a schedule we fetch the page once, parse the rows and compare them to what we already hold.
What you see
- Last checked and last successful check per source
- Health state with the plain reason when something fails
- The listing URL, so you can open the same page we read
What we never do
- Bypass captchas, challenges or TLS errors
- Log in, scrape private areas or call undocumented APIs
- Hammer a site: one request per page, backoff on errors
2. Compare versions
A new PDF or HTML document is stored with its SHA-256. If a document we already know is reissued with different bytes, it becomes a new numbered version and a diff is computed on the extracted text.
What you see
- Version list with fetch times
- Side-by-side diff with added and removed passages
- Extraction quality so you know when the text layer is weak
What we never do
- Overwrite an earlier version
- Guess a publication or effective date that the page does not state
- Hide a document because it failed to parse
3. Cited brief
Rules read the extracted text for sentences with obligation language, dates and scope. Each finding carries the quoted sentence and its anchor. Anything the text does not settle is marked unknown and routed to review.
What you see
- What changed, who is affected, obligations, effective dates, uncertainty
- A citation chip on every material statement
- “Pending human review” on low-confidence items
What we never do
- Present a draft as enacted or a notice as a binding rule
- Let document text act as an instruction to the system
- Upgrade a legal status by AI output
4. Act
Relevance rules match each document to your entity profile and raise an alert. From there you create tasks, assign owners and attach evidence. The audit log records every step.
What you see
- Alerts with the rules that matched
- Tasks with owners, due dates, evidence and comments
- Exportable weekly and monthly summaries
What we never do
- Send one email per document
- Share one organisation's data with another
- Delete history on plan changes
Timing and cadence
Listing pages are polled on a schedule during Bangladesh business hours, typically hourly for high-volume sources and less often for slow ones. PDFs are fetched once and then checked with conditional requests. Publication is bursty, especially in budget week, so a quiet month on a source is normal and is shown as such.
The status vocabulary
Every document carries three independent signals so that you never have to infer them:
- Legal status
- draft, enacted, in force, amended, superseded, withdrawn, repealed, guidance, informational, or unknown
- Date confidence
- stated in the document, inferred from the listing, or unknown
- Extraction quality
- text, partial, OCR needed, manual review, failed
See it on your own regulators
A 30-minute walkthrough with live documents from the sources you care about, or start with the free sample and judge for yourself.