← Back to Can I trust what the AI suggests and builds?
What counts as proof that something works?
Short answer
In the Kai-Zen method only a run is proof: every Status and every finished piece of work says what was run, where and when.
Before you start
Words used here (1)
- Kai-Zen
- Kai-Zen is the planning agent: it finds out what you want, chooses the way, hands the work to other agents and reports back.
Deeper
Also asked:
What counts as proof that my app does what I asked? Only a run: a test or a measurement actually carried out, tagged REAL, MOCK or UNPROVEN. It is tested by an agent other than the builder, with a saved PASS or FAIL, and a pass on the builder's own computer does not count as working live.
Kai-Zen follows these rules when it reports to you, and so do Agent.Coder, which builds apps, Agent.Tester, which tests them, and Agent.Concept, the agent that finds and makes GilbVeda's terms (GilbVeda is the shared list of terms). An agent is your AI following one written procedure, its standing instructions, for one kind of work. A run is something that was actually carried out, such as a test or a measurement. Agents agreeing is not proof; only a run is.
Every finished piece of work is tagged REAL if it ran on the real thing, MOCK if it ran on a stand-in, or UNPROVEN if nothing was run. Every finished piece of work, and every Status (where a Value stands now), also says what was run, where, and when. A Value is a perceived benefit, measured as a number. A number that was not measured is not a Status.
A Status holds only at the condition its Scale, the ruler the Value is measured on, names: where the stakeholders use it, at their size of data. A stakeholder is a person, group or system that cares whether a Value goes up or down. A reading taken elsewhere, such as on a local copy, on a smaller set of data or with a stand-in for the program people use, is written beside the Status with its condition, and does not move it. A Value Decision Table (VDT), a table that ranks options by how much they move the level's Values for their cost, counts only the Status.
A Status is only as current as its date. If the product has changed since, it is measured again before a VDT or a report leans on it.
Before a new test, or a new Meter (how a person, an AI or a test checks it), is trusted, it is run once on a case whose answer is known. A result of zero, or of nothing found, says what was searched, so a blind search cannot pass as a clean one. A fact from memory or from an old note is a lead, not proof: it is given with where it came from and its date. When it disagrees with the specs, the requirements written in your plan, the code or what you say now, the current one wins, and the disagreement is said.
In a round of delivery, Kai-Zen has what was built tested by an agent other than the one that built it, with a saved PASS or FAIL; for an app on GilbPlatform that is Agent.Tester, and a verdict that is only in its reply does not count. If Agent.Coder cannot start a separate tester, it tests the work itself and marks the result self-QA, which may stand in for a PASS from Agent.Tester but is not equal to one. Agent.Coder also proves a change to an app's screen in a real browser, with a screenshot and a count of the errors the browser reports; a passing test run is not that proof. A pass on the builder's own computer does not count as working in production, the live system people use. Kai-Zen has each Value the step aims at measured by an agent other than the one that built it.
A letter to someone outside the team about a delivery is written only when the delivery is REAL: live, and read back where the reader uses it. The answer on getting a delivery letter written explains it.
In GilbApp, an Actual, how far a delivered step was measured to move a Value, has an Evidence tag with the same three words, as the answer on tracking estimates and actuals shows. GilbApp offers Set Status whatever the Evidence tag says. It also counts the Status with the newest date as the current one, whatever its condition says, so a reading taken elsewhere and typed in as a new Status row becomes the current one there.
This answer is written for a computer screen.
Source: Standards read 2026-11-10: Kai-Zen.EvidenceRules (K 06ge5vp5qdx7552qp6hzxe4gq0), all 9 rules: Measured (A number you did not measure is not a Status.), Reading (Every Status and every 'done' says what was run, where, and when.), Tag (REAL (ran on the real thing), MOCK (ran on a stand-in) or UNPROVEN (not run)), KnownAnswer, Zero (so a blind search cannot pass as a clean one), Agree (Agents agreeing is not proof. Only a run is.), Fresh, Memory (a lead, not proof ... the current one wins; say that they disagree), Condition ("where the stakeholder uses it, at their size of data ... (a local copy, a smaller set, a stand-in client) ... does not move it. A VDT counts only the Status."). Who follows them: Kai-Zen step Report (hands off to Kai-Zen.EvidenceRules); Agent.Coder.Tier1 item 4, Agent.Tester.Standards, Agent.Concept.Standards (Kai-Zen.EvidenceRules: what counts as proof). P6: Kai-Zen.Cycle.Check (not the agent that built it — with a saved PASS or FAIL), .Measure (by an agent other than the builder); Agent.Tester.Task (A verdict that is only in your reply does not count.); Agent.Coder.AfterIncrement (Fallback "Mark the result self-QA. A self-QA PASS may stand in for an Agent.Tester PASS but is not equal to one"; "prove it in a real browser ... take a screenshot. Count the console errors. A passing test run is not this proof."); Agent.Coder.Solutions (A pass on your own computer is not production). P7: Kai-Zen.LetterRules.Real. P8: GilbApp af0f04ee838a4d55887d2ff1f72edd49a4ff7ca0 (https://gilb.com/app/version.json read 2026-11-10, buildTime 2026-11-09.522Z, dirty false), code read: lib/base/impactActual.ts:51-55 (REAL, MOCK, UNPROVEN, default UNPROVEN), components/records/ImpactActualForm.tsx:402-415 (Evidence tag), lib/base/statusOffer.ts:32-48 (offer reads no tag), lib/base/planImpacts.ts:915-928 (currentStatus: the Status row with the newest date wins; the condition is not read). GilbVeda public read 2026-11-10: t:EvidenceTag (Suggested).
Checked against GilbApp d529620.