Multiple Model Provider Routes
Publisher clue and review code can route through Vertex, Gemini, or OpenAI credentials. Provider models, quotas, behavior, and pricing must be verified at operation time.
Used by configured publisher stages with caching and structured validation. No single model name or cost is guaranteed by this page.
Credential-based routes exist in the model client. Callers still own prompt, schema, retry, budget, and editorial acceptance behavior.
Model clients and callers include error-handling paths, but quota exhaustion, retry behavior, partial artifacts, and spend controls must be verified for the selected route.
Target Owned Quality Gates
Current targets use executable structural, lexical, theme, worst-entry, and publisher policies. The exact gate and receipt for the selected target are authoritative.
| # | Check | Threshold | Why It Matters |
|---|---|---|---|
| 01 | Target Contract | Target-specific | Structure and policy belong to the selected publisher or product target. |
| 02 | Grid Integrity | Executable | Entries, crossings, dimensions, and symmetry are checked by code. |
| 03 | Worst-Entry Quality | No waived bad fill | The primary track rejects crosswordese, partials, and low-quality entries. |
| 04 | Theme Policy | When applicable | Themed targets enforce answer, placement, revealer, and evidence requirements. |
| 05 | Release Evidence | Immutable receipt | Acceptance and release decisions remain auditable. |
From Inputs to Release Evidence
Generation, acceptance, clueing, export, human review, and deployment are separate stages; a target may not use every stage.
rotational symmetry
configured lexicon sources
theme and policy checks
with clue caching
then optional deployment
Supporting Metadata and Publication Tools
Supporting modules can classify, title, resume, and publish artifacts. Availability and authority vary by target, and release remains a reviewed operation.
-
Legacy Difficulty Classifier
Easy / Medium / Hard classification exists for legacy app content. It is not the primary NYT-quality success contract.
-
Optional Puzzle Titling
Some routes can propose titles through model-assisted stages. Editorial review and target policy determine whether a title is usable.
-
Resume-Aware Runners
Selected runners record state or inspect artifacts to support resumption. Recovery guarantees must be verified for the exact command and storage path.
-
GCS Publisher
The publisher writes puzzle objects and an index to configured Google Cloud Storage. Release approval, credentials, bucket state, and client delivery remain separate concerns.
Browse Real Puzzles From the Library
These repository examples demonstrate puzzle structure and presentation. Each artifact needs its own target receipt and provenance before it can be described as currently accepted or publishable.
Interested in the Generator?
The generator is included in the full acquisition. Reach out to discuss terms.
← Return to Overview