What this axis covers
Pure-logic specialist class for functionality bugs distinct from security. Off-by-one in array indexing, wrong-branch in conditional logic, business-rule violations, missing edge cases the test suite forgot to cover.
Weakness axis #2
Correctness
Logic bugs, off-by-one, wrong-branch, missing-edge-case — caught at PR time, not at runtime.
Why now
Business-critical code can compile and pass stale tests while still violating real product rules. Logic bugs are frequent enough that human review cannot keep up with every edge case, migration, and integration path.
Where it hides in your codebase
- Arithmetic on indices, lengths, and date arithmetic (off-by-one)
- Conditional branches where the AI picked the wrong arm
- Business-rule violations: discount-on-discount, currency-precision, leap-year
- Missing edge-cases: empty inputs, single-element arrays, identity transforms
- Type-confusion across language boundaries (TS↔Python via JSON, etc.)
How Securie handles it
Intent-graph + property-based test generation
Securie extracts the declared intent (docstrings, ADRs, type signatures) and synthesizes property-based tests that fuzz the surface. A mismatch between intent and behavior fires a finding.
Cross-function taint + dataflow
The taint-analyst specialist tracks data through function boundaries — a wrong-branch in module A only matters if it reaches the user-facing handler in module D.
Sandbox replay of the failing case
When a logic bug is detected, Securie runs the failing input in a sandbox and confirms divergence from declared intent — never flags without proof.
What this axis is NOT
Not a lint tool
Linters flag style + simple bugs. Correctness specialists track semantic correctness against the codebase's own intent — same code, different intent → different verdict.
Not a unit-test generator
Test generation (axis #5) generates tests; correctness detects + fixes logic bugs and ships the test that reproduces them. The two axes interlock.