Theme-first construction
Theme entries can be validated for length, placement and grid compatibility before the wider fill attempt proceeds.
Current capability
Lead with what is working now.
Public claims are limited to capabilities demonstrated in the current production system. Additional sizes and output requirements can be packaged through a commercial implementation rather than presented as finished features.
15×15Themed generation
21×21Themeless capability
~2 minutesCurrent themed benchmark
13,000+Weekly cloud-run throughput demonstrated
Quality-banded lexicon
Fill candidates are filtered through quality layers rather than treating every technically valid entry as equally acceptable.
Publisher-mode clues
Clue generation is separated from editorial review so style, fairness and answer sense can be controlled.
Structured delivery
Grid, answer, clue and metadata outputs can be prepared for app delivery, editorial review or downstream publishing systems.
Editorial QA
Quality is treated as a production process, not a single score.
The workflow combines deterministic checks with editorial review and targeted repair. The goal is not to make every generated result pass. It is to prevent weak output from being presented as finished work.
01Structural validation
Checks grid dimensions, symmetry, connectivity, numbering, crossings and answer integrity.
02Fill-quality controls
Flags weak entries, undesirable abbreviations, obscurity, duplication and other lexicon risks.
03Clue exactness
Reviews answer sense, part of speech, plurality, tense, abbreviation signalling and recognisability.
04Editorial style
Checks directness, fairness, clue leakage, awkward grammar and publisher-specific conventions.
05Targeted repair
Repairs identified failures rather than regenerating the entire puzzle without diagnosis.
06Second-pass stability
Confirms that repaired clues remain valid and that previously accepted clues do not regress without evidence.
07Trace and artefact proof
Retains production evidence so a buyer can review what ran, what failed and what was released.
Production path
From specification to deliverable.
The production path can run as an internal tool, a managed workflow or part of a buyer's existing publishing stack.
Define
Set grid size, theme constraints, difficulty, style, vocabulary and delivery requirements.
Construct
Build the grid and fill using validated templates, constraints and quality-ranked entries.
Clue
Create clues with answer context and the target editorial specification available.
Review
Run QA, repair failures, repeat checks and retain the decision trail.
Export
Deliver structured data, puzzle files, PDFs or other agreed artefacts.
Publisher fit
Adapt the system to the buyer, not the other way around.
Commercial implementation should begin with the buyer's actual editorial and technical requirements. That is more valuable than a long list of generic AI features.
| Dimension | What can be specified | Commercial value |
|---|---|---|
| Grid format | Dimensions, symmetry, block density, theme structure and word-count limits. | Supports a defined product rather than one generic crossword type. |
| Editorial style | Clue directness, difficulty, abbreviations, proper-noun anchoring and house conventions. | Reduces downstream editing and protects brand consistency. |
| Vocabulary | Allowed and blocked entries, regional language, cultural scope and obscurity tolerance. | Aligns fill quality to a real audience. |
| Delivery | JSON, app-ready data, printable artefacts and agreed publishing formats. | Reduces manual handoff work. |
| Evidence | Logs, validation results, repair records and release status. | Makes production quality auditable. |
Technical truth
Current limitations should be visible.
Credibility improves when the website separates proven production capability from work that still needs packaging or buyer-specific implementation.
What is not being claimed
The site does not claim that every grid size is already productised, every generated puzzle is publication-ready without review, or operating cost is literally zero. Exact performance depends on template, theme, configuration, infrastructure and editorial specification.
Clear scopeMeasured claimsDue diligence
Technology licensing
Test the system against a real production brief.
A qualified evaluation can cover target formats, sample output, benchmark evidence, documentation and controlled source-code review.