Quality Principles Engine

Deterministic validation, independent of AI generation

An LLM that generates test cases and an LLM that grades them is one system marking its own homework. These five rules are code: each has a yes/no answer and a pointer to what failed.

This engine does not call the AI Model Gateway. If a rule needed a model to decide, it would not be a quality principle — it would be more generation.

rules v1.0 · 0 model calls

Quality gate

REVIEW REQUIRED

The specification does not publish until a person decides what to do about the findings below.

Pass

Fail

Findings

Human review required

QP-01

Requirement traceability

rule v1.0

Every test case references at least one requirement, and every in-scope requirement has at least one test case. Orphans in either direction fail.

PASS

Test cases linked to a requirement147 / 147

Requirements covered by a test case37 / 37

184 of 184 links satisfy this rule

No findings. Nothing in this specification violates QP-01.

QP-02

Expected results

rule v1.0

Every test case states an outcome that could be observed and disagreed with — not 'verify it works', and not a restatement of the last step.

PASS

Test cases with a stated, observable outcome147 / 147

Vacuous phrasing detected0

147 of 147 test cases satisfy this rule

No findings. Nothing in this specification violates QP-02.

QP-03

Boundary coverage

rule v1.0

Where a requirement states a limit, range or threshold, tests must cover the value at, below and above it. The limit is read from the structured requirement, never inferred from prose.

FAIL

Requirements stating a limit10

Limits covered at, below and above9 / 10

Boundary test cases generated29

9 of 10 bounded requirements satisfy this rule

1 finding

REQ-1042Rolling 24-hour transaction total — no test at below the limit (₹49,999).

Each finding points at a specific artifact. There is nothing to interpret — the rule either held or it did not.

QP-04

Negative coverage

rule v1.0

Every requirement carrying a validation or permission rule has at least one negative test case. A rule nobody tried to break has not been tested.

FAIL

Requirements requiring negative coverage29

Requirements with negative coverage27 / 29

Negative test cases generated41

27 of 29 requirements satisfy this rule

2 findings

REQ-1028Requirement states 1 validation rule but has no negative test case.

REQ-1047Requirement restricts access by role (Adjuster, Supervisor) but has no negative test case.

Each finding points at a specific artifact. There is nothing to interpret — the rule either held or it did not.

QP-05

Test-data completeness

rule v1.0

Every test case that needs data either specifies it or references a data definition. 'As appropriate' is not test data.

PASS

Test cases requiring data147

Data specified or referenced147 / 147

147 of 147 test cases satisfy this rule

No findings. Nothing in this specification violates QP-05.

AI generated the tests. SALC checks whether they satisfy measurable quality rules.

The separation is the product. A deterministic engine that rejects generated output for stated, measurable reasons is a different thing from a model reviewing its own work — and it is what makes the approval below meaningful. Stage 2 extends these five to eleven, including detection of assertions the source never supported.

147 cases · 5 rules · 0 models