Platform · J.1 How tools get made
How we decide what to build
DesignThe five knockout gates an idea has to pass, the 100-point rubric it is scored on, how ties are broken, and who makes the final decision.
J.1.45 min readLast revised November 2026
Status: Design. A specification. It describes how the component is meant to work, not something you can use on this site today. What the labels mean.
On this page
Section M of a solicitation tells bidders how their proposals will be judged. This page is the same thing for ideas: the complete set of rules by which a candidate tool is screened, scored and decided. The rules are public so that anyone can see why one tool was built and another was not.
The screening and scoring are done by fixed rules with no AI, so the same facts always give the same score. The decision is made by a person.
The five knockout gates#
Before any points are counted, a candidate has to pass five yes-or-no tests. Failing one ends the evaluation, and the reason is recorded.
- Public source. Its rules, formulas and thresholds come from officially released material: statutes, the FAR, the DFARS, NIST publications. Not a paywalled manual, and not industry folklore.
- Runs in the browser. It can work entirely on the user’s machine, with no server and no outside service.
- Private by design. It works offline, needs no account, tracks nothing and can enforce a content security policy that blocks outside requests.
- One job. It does a single task that can be described in one sentence starting with a verb. A suite fails, unless it is split into smaller ideas.
- Arithmetic, not advice. It computes, structures or looks things up. It does not give legal advice or stand in for a contracting officer’s judgment.
The 100-point rubric#
A candidate that passes the gates is rated from 1 to 5 on seven factors. Each rating is multiplied by the factor’s weight, giving a total between 20 and 100.
| Factor | Weight | A rating of 1 | A rating of 3 | A rating of 5 |
|---|---|---|---|---|
| Need for small businesses | ×4, up to 20 | A rare edge case | Comes up now and then, such as once a year | A recurring obstacle for most small contractors, costly when done wrong |
| Gap: no free, private option exists | ×4, up to 20 | Good free tools that run on the user’s machine already exist | A clumsy government portal, or open tools that are hard to use | Only paid products or lead-capture forms exist |
| Strength of the public source | ×3, up to 15 | An informal slide deck or an unclear draft | A policy memo or an interim rule | A statute, a codified regulation or a published standard |
| Fits in a browser | ×3, up to 15 | Needs very large data or heavy computing | Needs moderate bundled tables or a special parser | Simple arithmetic or text handling with very little data |
| Small enough to read | ×2, up to 10 | A sprawling program with many settings | Moderate, with clear inputs and outputs | Short enough to read in one sitting |
| Can be checked | ×2, up to 10 | Rests on interpretation | A formula with edge cases that need warnings | Plain arithmetic with official worked examples to test against |
| Low upkeep | ×2, up to 10 | Needs frequent data updates or fragile page reading | Updated once a year | Rests on text that changes only when the rule is amended |
What the score recommends#
| Result | Recommendation |
|---|---|
| 75 or more, and a Gap rating above 1 | Build it from the public rule |
| 75 or more, and a Gap rating of 1 | Link to the existing free, private tool instead of rebuilding it |
| 55 to 74 | Park it: a sound idea held back by missing data, a pending rule or capacity |
| Below 55 | Reject it, with the breakdown of points published |
One more rule: a candidate rated 4 or 5 on Need and 1 or 2 on Small enough to read is flagged for the owner to consider splitting into smaller tools.
ClearedStack does not exist to duplicate good work. The Link outcome is there so that a tool which is already free, private and built on public sources gets pointed to, not rebuilt.
Breaking a tie#
When two candidates have the same total, they are ordered by these tests, in this order:
- Fit with the current phase of the roadmap.
- A higher Gap rating.
- A higher Low upkeep rating.
- A higher Small enough to read rating.
- More independent requests from the community.
The decision is a person’s#
The score is a recommendation. The owner reviews each scored candidate and chooses one of the outcomes: build, link, merge into another tool, park or reject. The owner can also override the recommendation, or reopen a candidate that was screened out. An override needs a written reason, and the reason is recorded in the public decision log.
- 1 · The publicSuggest a tool or a correctionBy issue, by email or by proposing an edit.
- 2 · AI assistants (while writing)Help draft the rulesHelp write the scoring rules and their tests. They do not run the scoring.
- 3 · Human ownerApprove or overrideMakes the decision and writes down the reason for any override.
- 4 · AI assistants (while writing)Help draft the toolHelp write code and tests from the official rule.
- 5 · Human ownerCheck the math and the citationsVerifies each formula and citation against the official text before anything is merged.
- 6 · The publicCheck the resultAnyone can test a live tool and report a wrong answer.
Where the rules come from#
Each area of the catalog has its own profile of thresholds for the rubric: pricing, proposals, cyber, market research and DevSecOps. A candidate in an area with no profile is not scored by guesswork. It is paused until a profile is written, the same way any worker handles a case it has no rule for. See The pipeline and its ten workers.
To suggest a tool, see Request a tool or add your own. How often a tool is asked for is the fifth tie-breaker above.
Found a mistake, or something that has gone out of date? Open an issue or write to [email protected]. This page is a text file you can read and correct: Source of this page.