Buyer-workbench methodology

The proposed use is an input, not permission

The buyer brief starts with four use categories: internal research, client reporting, external redistribution and execution integration. These are descriptions of intended activity, not assertions that a particular agreement permits it. An actual review must identify the legal entity, recipients, jurisdictions and specific actions contemplated. The same output can raise different questions when viewed internally, quoted to a client or incorporated into an automated process.

Delivery is separately described as files, a research portal, an API or source code. Choosing source code does not establish ownership of the code, data or modifications. Choosing an API does not establish that a provider operates a suitable endpoint. The workbench turns each choice into evidence requests so the unknowns remain visible before a commercial decision.

How the checklist is assembled

Every brief includes eight base requests: intended use and recipients, model rights, underlying data rights, sample and evaluation criteria, delivery acceptance, information handling, support/change process, and retention/exit. The archive flag adds a historical-data request. Client reporting or redistribution adds an external-recipient request, while execution integration adds a responsibility-boundary request.

API delivery adds an API-acceptance request. Source-code delivery or proposed modification adds one source/modification request; selecting both does not duplicate it. Proposed training use adds a separate permission question. The default internal-research example has three users, 365 requested retention days and historical archive selected, producing nine requests. None is automatically marked approved.

User counts must be whole numbers from one to ten thousand, and requested retention must be a whole number from zero to 3,650 days. Zero retention is a request not to retain an archive, not a conclusion about legal recordkeeping obligations. Input bounds keep the worksheet usable and explicit; they do not express standard commercial terms or legal limits.

Evidence references are a record, not a score

A reader can attach a short document reference, version or question owner to each active request. The displayed count records how many active requests contain nonblank notes. It does not assess the quality, authenticity, completeness or legal effect of those notes. An empty reference means no note was entered; a completed reference does not prove the requirement is satisfied.

Changing the brief may add or remove questions. The current export includes only questions relevant to the current choices, with the supplied references attached. References for temporarily hidden questions remain only in the open page's memory and reappear if that question becomes relevant again. Reset clears the brief and all references; refreshing the page restores the default example. No persistent browser storage or server-side case record is created.

The cost equation and common boundary

For each quote, monthly recurring cost equals the monthly base fee plus the number of named users multiplied by the per-user monthly fee, plus the monthly data fee. Term recurring cost is that monthly sum multiplied by the selected number of months. Total cost adds the entered setup and exit charges once each. Effective monthly cost divides that total by the selected term.

All users are charged the entered per-user rate. The calculator does not assume a bundled free seat, a minimum purchase or a volume discount. If a real quote includes those features, convert it to a documented comparable assumption outside this simplified structure or use another model. The same user count and term apply to all three columns. Fees are interpreted as USD and remain constant throughout the comparison.

The default examples use three users and twelve months. Illustrative A has a 900-dollar monthly base, a 100-dollar per-user fee and 200 dollars of monthly data, plus 2,500 dollars of setup. Its monthly sum is 1,400 dollars and its term total is 19,300 dollars. Illustrative B totals 17,000 dollars and C totals 17,700 dollars under their separately displayed assumptions. These are invented teaching inputs, not market quotes.

Costs and judgments left outside the calculation

Taxes, currency conversion, inflation, fee escalation, internal implementation labor, usage overages, contingent discounts, financing and unentered charges are excluded. Exit cost is included only at the value entered, and the comparison assumes that amount is incurred within the selected boundary. An actual termination liability could follow different contractual triggers. The worksheet does not calculate damages, enforceability or a refund entitlement.

A lower numerical total is not a recommendation. Rights, output coverage, data history, update frequency, support, reliability and the adequacy of research evidence may differ. Establish comparable scope before interpreting a price difference. Keep material unknowns in the evaluation record rather than assigning a zero cost or an invented quality weight merely to complete a ranking.

Exports, implementation and revision

JSON exports retain the brief inputs, active requests, user-supplied references, engine version and export timestamp. CSV exports use the locally hosted Papa Parse library, including formula-like text escaping for spreadsheet safety. Cost exports preserve every quote input, shared assumption and calculated total. Files are created on the reader's device; they are not uploaded to a procurement service.

The initial implementation is version 1.0.0, published on 16 October 2026. Unit tests exercise the conditional request rules, input validation and independent cost totals. Browser checks exercise export, reset, navigation and mobile layouts. Those checks establish software behavior, not legal suitability, vendor quality or investment performance. A material change to the calculation or request rules should retain a dated explanation and a new versioned reference artifact.

Source notes

1. Papa Parse: structured CSV export and formula escaping
2. NIST: AI Risk Management Framework, for applicable AI systems rather than universal licensing requirements