Noding and precision guarantees

Noding makes every topological intersection an endpoint on every incident segment and normalizes collinear overlap before graph construction. Precision controls which XY coordinates participate in that topology. They are separate choices: a fixed grid can round coordinates without noding, and floating noding can insert intersections without rounding to a grid.

Choosing a guarantee

Configuration	Work performed	Contract
node_input = false, Unchecked	No intersection insertion. Fixed precision still rounds endpoints.	The caller asserts that the resulting linework is already noded.
node_input = false, Validate	No intersection insertion, followed by the independent validator.	Returns a typed error if the supplied or rounded linework is not fully noded.
node_input = true, Unchecked	Iterative floating or grid candidate noding.	Produces the noder's result without certifying its postcondition.
node_input = true, Validate	Ordinary noding, followed by the independent validator.	Succeeds only when the produced segments pass the full-noding checks.
CertifiedFixedPrecision	Hot-pixel snap rounding, followed by the independent validator.	Succeeds only for a validated fixed-grid result.

CertifiedFixedPrecision requires all of the following:

• node_input = true;
• PrecisionModel::FixedGrid with a finite positive grid_size;
• the Snap backend; and
• SnapStrategy::Grid.

Other combinations return UnsupportedOptionCombination before geometry work. Certification is a postcondition for the selected grid, not a claim that the grid preserves the application's intended narrow features or distinct points.

Validator postcondition

The independent validator rejects the first deterministic segment pair that contains:

• a zero-length segment;
• an intersection that is not an endpoint of both segments; or
• an unnormalized collinear overlap.

Failures use the structured noding-validation error family and retain segment indices plus the failure kind. Validation checks the segments produced by noding, precision rounding, coordinate restoration, and Z reconciliation; it does not merely re-check the original input.

Floating and fixed precision

PrecisionModel::Floating preserves source endpoints and computes inserted intersection coordinates in f64. It avoids intentional quantization, but it does not make floating predicates or constructed coordinates exact real-number arithmetic.

PrecisionModel::FixedGrid { grid_size } maps topology coordinates to integer grid indices separated by grid_size in the input coordinate units. A smaller grid retains more coordinate detail but may require larger exact grid indices; a larger grid is more aggressive and can collapse short segments, narrow gaps, or nearby vertices. Certified inputs must fit the supported integer grid range.

Use SnapStrategy::Grid when grid coordinates are the desired output contract. GeosCompat uses the grid for topology and then restores one deterministic nearest source XY per snapped node. That improves source-coordinate fidelity for compatibility work, but it is not set_precision emulation and is intentionally ineligible for CertifiedFixedPrecision.

Pre-snap and execution controls

pre_snap_tolerance optionally moves endpoints to nearby reference vertices before noding and therefore requires node_input = true. It changes semantic geometry and should use a tolerance expressed in the same units as the input.

Execution budgets and cancellation are non-semantic controls. They can stop noding or validation with a structured resource/cancellation error, but raising a budget does not change the selected precision or guarantee. Diagnostics and bounded traces report physical candidate, predicate, split, and validation work without selecting a different algorithm through public semantic options.