Standing reference · Re-verified weekly
Read these. Don't run them.
Two reference tables the tools draw on. Nothing happens when you arrive — no inputs, no results. They exist so you can point at a number in an argument and have the other person click through to the document it came from.
All of these are open, and stay open. The thing worth subscribing to isn't the table — it's the line that changed in it this week.
AI Repricing Index
What happened to the price of software you already own. Twenty-two rows across four segments, each carrying its billing unit, the published rate where one exists, what it replaced, and the date it was last verified. Credits, conversations, resolutions, seats and tokens are five different denominators — the index refuses to convert between them, because any exchange rate would be invented.
Benchmarks
Retention, growth and margin from companies you can look up. Every comparison point is a named public filer and a figure taken out of its own filing. Samples are small and stated on every metric — and the disclosure rate is the finding most people miss. When half the covered companies publish no retention figure at all, that silence is information about the metric and about them.
Every company
One page per filer, and the dot tells you how much is on it. Ticker, CIK and legal name read straight off SEC’s own bulk index, so every link opens the right company rather than a similarly named registrant. Each page shows exactly how far it has been read — hand-read, XBRL facts held, or a live route into EDGAR.
The coverage ledger
Exactly how much has been read, including the parts that look bad. Every figure is computed from the data files when the page opens, so nothing can drift upward while the work stands still. It names the competitor numbers directly — 350 sources, 500 million documents, 83 companies — and explains why those count a different quantity than this does.
MCP endpoint
Query it from Claude, Cursor or any MCP client. Five read-only tools over the same filing-sourced data this site reads, so the two can never disagree about a figure. Every result carries its sourcing tier, the URL of the document, and an explicit instruction to the calling model not to fill gaps from its own training.
The table is free. The change is the point.
A pricing index tells you where things stand. What actually costs you a deal is the row that moved on a Tuesday while you weren't looking — one in three tracked vendors changes something on their pricing page in any given week.
So nothing here is walled. Both tables stay open, permanently, and the weekly email carries the delta: what moved in the index, what a filing disclosed, and what got corrected. That's the only part you can't get by refreshing the page.
Get the weekly change.
Ten graded signals, plus every row that moved in the index. One email, unsubscribe in a click.
How it's sourced, and where it was wrong
- The methodHow a filing becomes a graded, linked claim — and why nothing publishes without a human reading it first. Read it →
- CorrectionsEvery fix, dated, with the original claim left visible. Nothing is edited silently. This is the checkable version of the trust claim everything else here makes. Read it →