A work sample. It takes the author’s own working paper, The Response Gap (doi:10.5281/zenodo.21752834), which is not peer-reviewed, and does what a governance analyst does with a policy proposal: states each requirement plainly, says how it would be tested, and runs one test. The minutes in the example are invented. The check has not been tested on a real system. This chapter does not re-check the paper’s claims about credit markets, and it is not data-privacy or data-protection experience.
Many lenders run automated systems that watch a borrower’s collateral and act when it falls below a threshold. The paper calls the interval between the system detecting a shortfall and taking an action that cannot be undone the response gap. Its argument is that this interval is often a design choice, not a physical limit, that it can be set shorter than an informed response needs, and that it should therefore be treated as a disclosed and bounded parameter. It also predicts, as something not yet tested, that AI models trained only to minimize the institution’s loss will tend to shorten the interval unless response time is written into the model’s specification as a limit.
The paper tags each of its own claims as documented, modeled, prospective or open. A policy that cannot say how it would be checked is only a wish, so the next section does that for each proposal.
The requirements follow the paper’s section 5. The test column is this chapter’s proposal for how each could be checked, not something the paper specifies.
| Control | Requirement, in plain words | Test (proposed here) | Status | Paper |
|---|---|---|---|---|
| The trigger is written into the contract as a formula, not described in policy language. | Give an outside reviewer the contract and a price history. They should reach the same trigger moment by calculation. | Proposed. The paper cites disclosure already common in some derivatives markets. | § 5.1 | |
| A disclosed minimum time must pass between detection and the irreversible action. | For each event, compare the actual interval with the budget. Run below. | Proposed. The paper draws the idea from rule-based finality in distributed-ledger protocols. | § 5.2 | |
| Each parameter that governs liquidation risk is labeled by how it is known. | Every parameter carries exactly one label: validated on the institution’s own data, inherited default, or set by the model without a human-readable reason. | Proposed. The paper borrows the practice of tagging claims from formal verification. | § 5.3 | |
| A forced sale above a size or speed limit fixed in advance pauses automatically. | Submit a sale just over the limit and one just under. The first should pause with no staff decision, and the second should not. | Proposed. The paper points to exchange-level circuit breakers. | § 5.4 |
The response-time budget is the easiest to make exact. The check takes the minute a shortfall was detected, the minute the irreversible action happened, and the disclosed minimum in minutes. It says pass if the interval is at least the minimum (equal counts as a pass), fail with how many minutes short if it is not, invalid if the action is recorded before the detection, and ask if anything is blank. The example numbers are invented.
Nothing you type leaves this page.
A control that nobody maintains stops being one. For each row of the matrix a register would keep the same few fields, shown here as a template with no real entries: an identifier, a named owner, a version, the date last reviewed, the date the next review is due, and the condition under which the control is retired. When the paper’s own text changes, the matrix changes with it and the old version is kept, which is how the author’s other pages are handled: a dated audit log records each change.
This is one worked example of reading a policy as controls. It is not data-privacy or data-protection experience, and the paper is about credit systems, not enterprise data. The paper is a working paper and is not peer-reviewed. Its claims about credit markets are tagged in the paper by status and are not re-verified here. The four tests in the matrix are proposals made in this chapter. The response-time check has not been tested on a real system, and its example minutes are invented.