Methodology
A registry that asks you to trust it owes you the method. This is what the crawler does every day, what gets written down, and what is still missing.
As of
The four steps
Step 1: Read the page
Each program's own affiliate page is fetched daily and converted to text. The page body is stored by content hash, so an unchanged page costs nothing and every historical reading stays re-parsable when the extractor improves.
Step 2: Extract only what is stated
Commission, cookie window, payout threshold, schedule and methods are pulled from what the page says. Nothing is inferred from industry norms. A guess would corrupt a record meant to be relied on for years; a blank is honest and can be filled later.
Step 3: Append, never overwrite
Every reading is a new row. Nothing is edited in place. The database rejects updates and deletes on the history tables outright, so the record cannot be quietly revised.
Step 4: Separate learning from change
Finding a value we did not have is a discovery. A program editing a value it had published is a change. Only the second kind reaches the change feed.
Where this is weakest
Publishing the gap is not modesty. A registry that overstates what it knows is worth less than one that is precise about what it does not, and these numbers are the honest state of a young extractor.
The rest publish nothing concrete on their affiliate page, or state it in a way the extractor cannot read without guessing. Those are listed with their source page and nothing else.
- First values recorded
- A value the registry learned for the first time.
- 1,053
- Actual term changes
- A program editing a value it had published.
- 39
- Values lost
- The extractor's own instability, published for the same reason as the rest.
- 2
- Independently verified
- Programs where three different affiliates attached a payout record that a machine read and matched to their report. The rest of the 286: public terms, or reports without a matched record.
- 0