Try another price
Pay or build?
Month
Score
COTBase Premium at $14 a month, estimated from its public pages on 13 November 2026. Each dot is one of 100,000 possible futures from the model. A dot leaves PAY in the month when, in its future, rebuilding becomes cheaper than three more years of the plan.
AI makes software cheaper to rebuild every month. PayOrBuild prices each subscription against its own rebuild, and finds the month the answer flips.
The ranking
Products in order of their score. Those whose owners had the code inspected come first, because their numbers are measured, not guessed. Hover a row to read its next five years.
This list is a prototype. COTBase Premium is the only real product in it, estimated from its public pages. Every row marked Sample is an invented product run through the same model, to show how the list works.
Check a product
A label covers one plan and one user over three years. It prices a lawful rebuild from what the product shows in public, sets the plan against it, and dates the result.
What the estimate is built from
What one subscriber uses
Each capability is sized in function points. Seen O1 or O2 means it can be read or observed from outside, and is priced. O3 and O4 are hidden, and are reported beside the score.
| Capability | Points | Seen | |
|---|---|---|---|
Hidden work, % of the counted size
to
Attention to run a copy, hours a month
to
Data licences, compute or fees that anyone running a copy would also pay. They put a floor under the line.
Unit costs and constants
Tokens per point, $
to
Oversight per point, hours
to
Halving time of AI cost, months
to
Attention at the floor, hours a month
to
No measured index exists yet. These are the declared ranges of the method, read as 10th and 90th percentiles. Horizon 36 months, 100,000 outcomes, seed 20261113.
Price against rebuild over time
P/R is the price divided by the cost of a rebuild that starts in that month. Below the line, paying is cheaper. The band is the range of the public estimate.
Chance that paying stays cheaper
Share of 100,000 outcomes in which the plan still costs less than a rebuild. The stamp is valid while this stays above 90%.
Every stamp expires
A rebuild gets cheaper as AI gets better, so the line climbs toward every price. A stamp is issued for a period, like a ration coupon, and lapses unless the product earns the next one.
Ship
Each month of new work adds rebuild cost, and each month the old work loses some. On these numbers the AI-priced part of a rebuild loses of its cost a month.
Take time to prove
If a copy needs months of evidence before anyone can trust it, that wait is a cost AI does not remove. It raises the floor under the line.
Carry real costs
Data licences, compute and fees are paid by every copy too. A price that sits close to them cannot be undercut by a rebuild.
Lower the price
The line is a price as well. For this product it stands at a month today and in three years.
The private inspection
From outside, an estimate has to guess two things: how much work it cannot see, and how much craft the code holds. The owner can replace both guesses with measurements. An open instrument runs on the real code, on the owner’s machines. Only ranges and a fingerprint leave the building.
What an inspection could change
Three ways to a stamp that lasts three years, for the product in the checker.
Months until the stamp expires, by what the inspection finds. An amber edge marks a stamp that lasts three years or more. Choose a cell to load it as a result.
Enter a result
Measured size, function points
to
Hidden work, %
to
Craft factor
to
Paste an inspection summary
Before and after
Example result, not a real inspectionAn inspection can lower a score as well as raise it. Either way the product moves ahead of uninspected ones in the ranking.
How a score is made
Five steps, shown with the numbers of COTBase Premium. The model runs in your browser with a fixed seed, so anyone can recompute a label and get the same result.
Read the plan
Take one plan and one user, and list what that user gets. Here: 14 capabilities, from the weekly report ingest to the charts.
Count the work
Size each capability in function points. Only what can be seen from outside is counted: 100 points.
Price a rebuild
Price a lawful rebuild at today’s AI and labour rates, with setup, upkeep and three years of attention. 100,000 outcomes give $3,000 to $7,900.
Set it against the price
Three years of the plan cost $504. The score is how far that sits below the cautious end of the rebuild: 83.
Date it
AI cost is taken to halve every 6 to 12 months. The line reaches this price in February 2029, so the stamp is valid until then.
What this version does, and what it does not
- It runs the whole model in your browser. The method is the ORC Value Standard, version 0.2, with the expiry rule proposed for 0.3.
- COTBase Premium is estimated from its public pages, read on 13 November 2026. Nobody has inspected its code.
- Every other row in the ranking is an invented sample, or a product you checked yourself. Your checks stay in your browser.
- For any other product it reads the public home, plans and features pages, as a polite robot that obeys robots.txt, and drafts the list of capabilities from them. The draft is a starting point: check the price and the count.
- The inspection instrument does not exist yet, so its result is typed in. The lettered results are invented to show the mechanics.
- The unit costs and the halving time are declared ranges. Nobody has measured them yet.
If you pay for subscriptions
See which plans are a bargain against their own rebuild, and for how long. The label never tells you to copy a product. Where a price is above the line it reports the premium and stops.
See the rankingIf you sell one
Know your date before your customers do. Have the code inspected, show the stamp on your pricing page, and keep it by shipping faster than the line climbs.
If you build agents
A buying agent can recompute any label before a renewal. Each label is a small JSON file with a fixed seed, so two parties always arrive at the same number.
Questions
Short answers. Each number on this page is an estimate with a stated range.
Does PayOrBuild tell people to copy products?
No. The rebuild is a yardstick: what a lawful rebuild from public evidence would cost one user. The advice runs one way. Above the line the label says pay. Below it the label reports the premium and says nothing more.
What is the line?
The point where three years of the plan cost the same as rebuilding the product, at the cautious end of the estimate. That is a score of zero. Rebuilds get cheaper, so the line climbs toward every price.
What does each dot at the top mean?
The model draws 100,000 possible futures for the inputs it is unsure about: the size of the hidden work, the craft, the attention a copy needs, and how fast AI gets cheaper. Each dot stands for an equal share of them. A dot moves to BUILD in the month when, in its future, a rebuild becomes cheaper than three more years of the plan. The stamp expires when one dot in ten has moved.
Why does a stamp expire when the product has not changed?
Because the alternative has. The model takes the AI-priced part of a rebuild to halve in cost every 6 to 12 months. A product that stands still drifts toward the line, and a stamp is only issued for the time before it gets there.
What leaves my company in an inspection?
A summary with four ranges (size, hidden work, craft factor, proof time), a pass or fail on whether the measured code is the live service, and a fingerprint of the full record. The code and the record stay with you.
Why do inspected products rank first?
Their numbers rest on measurements. A public estimate has to guess the hidden work, here at 5% to 40%, and reads craft from the outside. An inspection can also lower a score, as the flashcards sample shows, and that product still ranks ahead of uninspected ones.
What do the score and P/R mean?
P/R is the price divided by the cost of a rebuild. The score is 100 × (1 − P/R), taken at the cautious end of the rebuild. A score of 83 means that paying costs 83% less than building. Zero is the line.
How reliable are the numbers?
They are estimates at the lowest level of assurance. The unit costs are declared and not measured, and the halving time is an assumption. What is fixed is the arithmetic: two people who enter the same evidence get the same label.