Why your codebase size doesn’t tell you what buyers will pay

Decide what you want a buyer to evaluate

For a repository-and-history license, your starting package might be the code, its development history, an inventory and the relevant permissions. For a coding-task proposal, you would also need to identify the problem, starting revision, expected change and tests. A proposal for reproducible tasks adds the environment, commands and execution evidence needed to attempt the work.

These are different things to appraise. You do not need to turn every repository into a bug-fix task before discussing a repository license. Equally, counting a candidate fix does not establish a finished task that a buyer has accepted.

Write a short description of the deliverable you can provide today, then separate the work that would still be required. That makes the next conversation concrete and helps you avoid comparing a per-task rate with an estimate for repository access. If you do not know which use fits, bring the inventory and available evidence to the appraisal and ask which licensing route is worth investigating.

Find the useful preparation scope inside your archive

The listed company has a training-task selection of 14 repositories. That is about 5% of its repository count, but it contains approximately 85% of the collection’s test files, 70% of its original code and half its commits.

One company’s selected portfolio. Measurements dated October 20, 2026; source checked October 23. Percentages use rounded totals. The selected repositories are included in the whole collection.

For this owner, the 14 repositories provide a focused place to investigate a task-oriented package. Starting there would cover most of the recorded test material without first preparing every repository in the archive.

For your own collection, look for a similarly coherent starting scope: repositories relevant to the proposed use, with history, tests and documentation you can actually supply. The right scope may be much smaller than your total inventory. There is no reason to copy this company’s count of 14.

This selection was not random. Its concentration does not show that 85% of the archive’s value sits in those repositories, or that the remainder cannot be licensed for another use. It tells this owner where much of the evidence for this particular review is located.

Turn “we have tests” into evidence someone can inspect

The selected repositories contain 233 bug fixes that landed with a test, and all 14 have CI configuration. The owner reports test runs with no failures, but Rightmark had not seen the logs at the source check.

If you have similar material, a useful next step is to document representative candidate fixes. Record the starting revision, the behavior being fixed, the relevant test, the dependencies and commands, and the execution output. If you cannot reproduce an example because it needs a private service, record that dependency as a preparation gap.

The output is a documented example that can be reviewed against a proposed use. It is not yet an accepted task count or a revenue forecast. The listing’s 233 recorded fixes remain candidate material until that work and the relevant acceptance review are established.

Where the money is: code licenses and finished tasks

The Rightmark Price Index, Q3 2026, updated October 10 and rechecked October 23, gives owners two useful price comparisons. All amounts below are US dollars. Each row prices a different deliverable; they are not successive rewards for improving the same repository.

Indicative estimate, not an offer. Final offers depend on the systems you connect, data size and quality, access terms and due diligence.

The first row concerns access to a repository and its history. The second concerns a finished task product, which may involve additional engineering by a task builder. The ranges cannot be added together, and the task price need not all go to the original code owner.

A separate estimate, Low confidence: the index places premium or task-grade repositories at $10,000 to $100,000+ per repository. This is Rightmark’s own estimate for repositories that build, pass tests and contain real bug-fix history. No published buyer per-repository price supports the band. It is a route to investigate through appraisal, not a premium earned by this portfolio.

A public licensing program makes the volume-based route more concrete. Its advertised non-exclusive terms, checked October 23, value source code at $100 per million tokens and Git history at $40 per million tokens. It requires original work, authority to license it for AI training and compliance with its technical requirements. These are advertised rates, not evidence that an owner has been paid; the posting does not establish a complete net-payout calculation.

For illustration only, suppose that program qualifies a repository with 1 million source-code tokens and 5 million history tokens. The advertised-rate calculation is (1 × $100) + (5 × $40) = $300. That is illustrative arithmetic at the advertised rates, before any adjustments or terms not stated in the posting. It is not this company’s estimate. Its 5 million lines of code cannot be substituted for token counts.

What buyers require before the task price applies

The research behind the task benchmark reports a $200 to $2,000 ballpark from multiple industry interviewees. The original research covers reinforcement-learning tasks across domains; Rightmark uses it as the reference for its code-task index row. It also explains the work behind that price: a task needs an environment in which it can be attempted, a grader that rewards the intended solution rather than a shortcut, and difficulty suited to the model being trained. These are reported purchasing and quality criteria, not a live order for this portfolio.

That makes the 233 fixes useful leads for investigation. To turn one into a task proposal, demonstrate the starting state, the requested change, the execution environment and how the tests distinguish a correct solution. CI configuration or an owner-reported green test run alone does not establish those properties or automatically raise an estimate.

Here is the money calculation for a hypothetical batch of 10 completed tasks accepted at the reference rates: 10 × $200 = $2,000, or 10 × $2,000 = $20,000 in gross task-level revenue. Neither the task count nor acceptance is observed in this portfolio. The source does not establish its conversion rate from candidate fixes to accepted tasks, so multiplying 233 by the task price would not produce a defensible estimate.

Before paying to prepare a batch, establish the proposed rate, acceptance test, payment trigger and who funds the work. Then calculate your own proceeds: the amount payable to you under the agreement, less your preparation costs and any applicable fees or deductions. If another business turns your licensed code into tasks, its task-sale revenue and your licensing income are different amounts. No owner share or preparation budget is established for this case.

Find your next step from the evidence you have

Use the following guide to locate the work that would make your own appraisal more specific. It is a preparation guide, not a scoring model or a prediction of price.

Rights can change the scope as much as technical preparation. One repository in the example’s task selection was built for a client. The owner says it may license it; Rightmark had not seen the contract. If your archive includes client work, identify it in the appraisal instead of treating ownership of the repository as proof that every licensing permission is established.

The insight: match your evidence to the dollars you want to pursue

For an owner with an inventory and full history, the $150 to $1,500 per-repository benchmark is a starting reference for discussing a standard license. Ask for the code and history volumes, eligible scope and rights behind any actual estimate. Hundreds of folders or years of development spend do not tell you those inputs.

For an owner with reproducible, well-graded problems, the $200 to $2,000 per-task reference makes a different commercial conversation worth having: who will build the tasks, what the buyer will accept, and what the code owner receives. The 14-repository selection points to where this company could investigate that route. The next commercial question is how many usable tasks a sample can support, at what preparation cost and under whose agreement.

If your appraisal seems low, ask which of these products it evaluates before spending money on improvements. Preserve the history for a volume-based license; document execution and grading for a task proposal. A change that helps one route may not change the price under another. Resolve the proposed unit and acceptance criteria before preparing the whole archive.

The outcome to seek is a scoped commercial proposal: what can be licensed, the dollars per qualifying unit, the evidence still needed, your preparation costs and the amount you would receive under the agreement.