# Additional-generation seam continuity fix

Date: 2026-08-07

## Scope

This revision addresses the remaining defect from the previous geography-integrity package: visible terrain discontinuities along the established-map / newly-generated-map frontier.

The previously reported disconnected prefecture component is **not** treated as an error merely because it is small. A sea-separated component can be a legitimate island. Existing component cleanup already preserves sea-isolated administrative components when there is no adjacent land region to merge into, and this revision does not add an island-erasing cleanup pass.

## Changes

1. **Literal old/new frontier constraint**
   - The final terrain pass now measures the actual generated-coverage transition rather than relying only on the selection rectangle or alpha feather.
   - Newly generated land cells directly touching established land are constrained to the established elevation neighborhood.
   - The established side remains unchanged.

2. **Outward gradient propagation**
   - The solved contact edge is propagated up to 12 cells into the new side using coverage distance.
   - This avoids simply moving the visible seam one or two cells inward.

3. **Dedicated hard seam metric**
   - `maxEstablishedFrontierElevationJump` measures only land/land edges where generated coverage changes from established to new.
   - Hard limit: `0.075`.
   - A candidate above this limit fails with `established-frontier-elevation-jump`.

4. **Best-available policy split**
   - Interactive preview may still show a candidate that only misses a soft terrain/human-geography quality floor.
   - A hard seam failure is never accepted as `best available`; it is rolled back.

5. **Diagnostic warning semantics**
   - Natural cliffs inside the newly generated region no longer make the whole seam status `warning` merely because one cliff exists.
   - Cliff density and literal frontier continuity are evaluated separately.
   - Advanced diagnostics now show `Established frontier elevation` and the hard limit.

## Validation

### Reported-bug regression

`node VALIDATE_REPORTED_PATCH_BUGS.mjs`

- river path points on sea: `0`
- duplicate prefectural-capital prefectures: `0`
- tiny generated prefectures: `0`
- municipal breaks on established frontier: `0`
- prefecture breaks on established frontier: `0`
- maximum established-frontier elevation jump: `0.0268`
- seam status: `clean`
- seam gate reasons: none

### Direction / seed stress validation

`VALIDATE_SEAM_STRESS.mjs` was run in four separate two-case processes (to avoid cumulative process runtime limits): 2 base seeds × right/left/down/up.

- cases: `8`
- hard limit: `0.075`
- maximum independently measured frontier jump: `0.0364`
- all cases: seam hard gate PASS

See `SEAM_STRESS_RESULT.json` for per-case values.

### Existing regression suites

`STEP15_VALIDATION.mjs`: PASS
- expansion seam status: `clean`
- generated-footprint escaped cells: `0`
- unresolved road portals: `0`
- unresolved rail portals: `0`

`STEP17_VALIDATION.mjs`: PASS
- Alternative variants differ
- explicit render revision retained
- single-patch Worker lifetime retained

`FIX_ADDITIONAL_GENERATION_VALIDATION.mjs`: PASS
- strict quality caller still rolls back on quality failure
- interactive preview still accepts a soft-quality best-available candidate when seam continuity passes