Invalid-Input
Vectors

Failure Classification Must Preserve Invalid / Unsupported Separation

The current candidate framework distinguishes at least three operational states:

VALID AND SUPPORTED
VALID BUT UNSUPPORTED
INVALID

Reference Vectors shall test that implementations do not collapse these states into one generic failure condition.

Invalid Rational Structure Must Fail

Reference Vectors shall include malformed CRPC values such as zero denominators, structurally invalid integer fields, prohibited sign forms, or other expressions that cannot represent a valid canonical rational Pang value.

Conforming implementations shall reject malformed CRPC rather than infer, coerce, round, or repair it.

Valid Noncanonical Form Must Not Be Misclassified as Invalid Where Accepted

Where an input interface explicitly permits valid noncanonical rational forms, values such as unreduced fractions shall normalize rather than fail.

Invalid-Input Vectors shall therefore distinguish a normalizable valid representation from an actually malformed rational expression.

Invalid Point Structure Must Fail

The corpus shall include Points with missing coordinate components, malformed CRPC components, prohibited extra structure, or other violations of the canonical Point contract.

A malformed Point shall not acquire canonical meaning through field guessing or positional inference.

Segment Definitions Must Satisfy the Primitive Contract

Reference Vectors shall include malformed Segment structures, invalid endpoint Points, and zero-length Segment candidates where distinct endpoints are required.

Invalid Segment input shall be rejected rather than silently expanded, collapsed, or rewritten into another geometry type.

Polygon Invalidity Must Be Deterministic

The corpus shall include SCPE candidates with too few distinct vertices, consecutive duplicate vertices, zero area, prohibited nonadjacent intersections, malformed Points, invalid closure structure, and other violations of the SCPE contract.

Only explicitly permitted normalization—such as removal of redundant collinear middle vertices or an accepted repeated terminal closure marker—may alter representation before validity is determined.

Geometry Beyond the Survey Domain Must Fail Survey Validity

Reference Vectors shall include exact Points and supported geometry that fall outside D_H = {(x,y) : -H ≤ x ≤ H and -H ≤ y ≤ H}.

Such input shall not be clipped, wrapped, snapped, translated, or reinterpreted as valid Survey geometry.

The earlier ±1,000,000 Pang half-span is a superseded working candidate and shall not be used to construct canonical invalid-input fixtures.

Because the numerical value of H remains open under FMD-04B, symbolic and parameterized outside cases may be used now, while final concrete numerical edge fixtures remain dependent upon a justified capacity decision incorporated into the governing Specification. FMD-04A already supplies the closed-square geometry and boundary semantics.

More Than One Plausible Meaning Is Not Canonical

The corpus shall include supported input forms whose structure is insufficient to determine one canonical meaning and for which no Specification-defined normalization rule resolves the ambiguity.

Such input shall fail rather than receive an implementation-specific interpretation.

Canonical Interchange Structure Must Be Enforced

Reference Vectors shall include malformed FSF-CJSON-1.0 objects, missing required fields, invalid field types, prohibited unknown fields where the profile forbids them, invalid canonical integer strings, and other structural violations.

A conforming parser shall reject invalid canonical structure rather than silently drop, rename, default, or reinterpret prohibited content.

Contradictory Encodings Must Not Be Resolved Opportunistically

Where a representation contains duplicate, contradictory, or mutually incompatible fields, Reference Vectors shall require deterministic rejection according to the governing serialization and validation rules.

An implementation shall not select whichever conflicting value is easiest to consume.

Valid but Unsupported Form Must Remain Distinct From Invalidity

Reference Vectors shall include representations that may be valid in another governed profile, version, or future extension but are unsupported by the implementation under test.

Such cases shall be classified as unsupported where the governing Conformance framework permits that state, rather than being misreported as malformed canonical mathematics.

Unknown or Incompatible Version Must Fail Clearly

The corpus shall include unknown, unsupported, or incompatible Specification versions, FSF-CJSON profile versions, and Conformance profiles.

Implementations shall not reinterpret those inputs under a different version's rules unless the compatibility model explicitly authorizes the conversion.

Normalization Must Not Manufacture Validity

Reference Vectors shall include malformed or semantically invalid inputs that superficially resemble valid noncanonical forms.

The implementation shall reject those inputs rather than “normalize” them by changing their mathematical meaning.

Near Is Not Equal

The corpus shall include values near valid Points, Segment boundaries, SCPE boundaries, and Survey Domain limits to prove that exact canonical truth is not replaced by epsilon, tolerance, snapping, or floating proximity.

Invalid or distinct exact input remains invalid or distinct even when visually close to valid geometry.

Approximation Must Not Masquerade as Exact Canonical Value

Reference Vectors shall include rounded, truncated, decimalized, or generalized representations presented as though they were exact.

The implementation shall preserve the distinction between exact canonical value and lossy derived representation.

Open Mathematics Must Not Be Guessed Into Existence

The corpus shall include requests for operations whose canonical mathematics or result type remain outside the current Specification closure, such as arbitrary exact rotation or general Boolean result geometry where no canonical output type has yet been defined.

A conforming implementation shall report the operation as unsupported or outside the governing profile rather than fabricate an approximate or implementation-specific canonical answer.

Invalid or Unsupported Complexity Must Terminate Deterministically

Reference Vectors shall include pathological redundancy, excessive but bounded structures, invalid recursive forms where applicable, and other cases intended to expose nontermination, hidden recursion, or uncontrolled implementation behavior.

Conforming behavior shall terminate deterministically under the applicable profile and classify the case according to the governing complexity and validity rules.

Every Implementation Must Reach the Same Canonical Judgment

Independent conforming implementations shall reach the same validity judgment and the same required failure or unsupported-state classification for identical input under the same Specification and Conformance profile.

Human-readable diagnostic wording may vary. Canonical validity and governed failure classification may not.

Negative Testing Is Now a Concrete Candidate Proof Surface

The original page correctly deferred actual invalid fixtures while grammar, precision, serialization, and Conformance failure handling were unresolved. Those dependencies are now substantially defined for the candidate core.

Section 09 can therefore support a real negative corpus across CRPC, Point, Segment, SCPE, parameterized Domain validity, normalization, FSF-CJSON, version/profile compatibility, unsupported states, and exactness traps. Remaining open work is concentrated in final error identifiers, FMD-04B numerical Domain capacity, future grammars, and final corpus packaging.

Referenced Requirements

Invalid-Input Vectors will derive especially from Requirements Findings #7–#10, #20–#22, #30, #48, #57–#60, #70–#75, and #78–#83.