ChessVision Engine Recognition · Reconstruction · Analysis

Available for acquisition

Universal browser chessboard recognition. Local engine analysis.

ChessVision Engine is a Chromium extension architecture designed to discover, reconstruct and analyse chess positions across diverse browser interfaces — using layered DOM, semantic, geometry and visual-recognition methods, with analysis executed locally by the bundled Stockfish 18 engine.

Manifest V3 Local analysis Site-agnostic fallback Stockfish powered v1.6.0
Explore the technology Acquisition details

§ 1 — The technology

More than a move overlay.

The difficult component of browser chess software is reliable board understanding: reconstructing a valid position from a page you do not control, without knowing in advance how it was rendered. ChessVision Engine treats this as a layered engineering problem rather than a single selector.

Authoritative sources — first priority

1.1

Direct FEN / state attributes

Authoritative board-state attributes are read directly where a page exposes them, avoiding inference entirely.

1.2

Known DOM structures

Proven readers for well-known site families, including Chess.com-style square classes and Chessground/Lichess-style geometry and orientation.

1.3

Move-list reconstruction

When exact metadata is required, positions are rebuilt from a bounded move-list replay rather than guessed.

Site-agnostic fallback layer — added in v1.6.0

1.4

Generic 8×8 board discovery

Coordinate metadata (data-square, data-coordinate, file/rank attributes), semantic grids and chess-labelled custom containers.

1.5

Semantic square & piece metadata

ARIA labels, titles, alt text, data-piece attributes and piece-type/colour tokens such as white-knight or bP.

1.6

Unicode & image-name recognition

Unicode chess glyphs and common image/SVG piece filenames are interpreted into structured piece placement.

1.7

Geometry-based square mapping

Where board orientation can be established, piece positions are mapped by geometry as a bounded fallback.

1.8

Visual / pixel fallback

A lazy template-matching path exists for boards where no DOM representation can be extracted.

1.9

Explicit side-to-move inference

Turn metadata is recovered from the board or its local wrapper — and when it cannot be trusted, no turn is invented.

Termination

1.10

Fail-safe refusal

If the board state cannot be established reliably, ChessVision Engine does not fabricate a position. It stops and reports why.

§ 2 — Architecture

Recognition, validation, analysis — deliberately separated.

Detection, policy and analysis are separate concerns in the codebase. Board understanding never leaks directly into the engine interface, and the engine is never handed a position that has not passed state validation.

i

Browser interface

Any Chromium-rendered page containing a chessboard, on demand. No <all_urls> content script; the reader is injected into the explicitly activated tab only.

ii

Board discovery

Known-structure readers run first; the generic fallback layer takes over only when they cannot reconstruct a board.

iii

Piece & square recognition

Pieces are read from classes, metadata, Unicode, image names or geometry — always bounded, never document-wide.

iv

Position reconstruction

Square-wise readings are assembled into a full FEN board description.

v

State / turn validation

Position legality is verified; the side to move is recovered explicitly or treated as unknown.

vi

Compliance gate

Recognised active competitive-game contexts suppress engine execution before any search begins.

vii

Local analysis engine

Stockfish 18 Lite (WASM) in an offscreen document, single thread, bounded memory and lifecycle.

viii

Ranked analysis output

Best move, a representative mid-tier move and the lowest-ranked legal move — plus mate-threat warnings, rendered as an overlay.

Recognition Policy gate Analysis

§ 3 — Analysis engine

Local Stockfish, engineered for bounded memory.

The bundled Stockfish 18 Lite engine runs as local WebAssembly inside an offscreen document. Core board recognition and analysis require no account, no API key, no remote LLM and no hosted analysis service — nothing leaves the machine.

Best move Original full-strength single-PV Stockfish pass — the source of the primary recommendation.
Ranked moves A separate bounded MultiPV pass ranks all legal moves to expose a representative mid-tier move and the engine-ranked lowest legal move.
Mate logic Mate-distance detection plus an opposition forced-mate warning derived from Stockfish mate distance and detected board orientation.
Sanitation Between every recommendation the engine crosses an explicit barrier: stop → ucinewgame → MultiPV reset → Clear Hash → position startpos → isready. No search state bleeds between positions.
Resource bounds One thread, a 4 MB transposition hash, PV retention capped at eight plies, background queue capped at the newest position.
Lifecycle Explicit start/stop session control. The engine host is destroyed on stop, tab detach, game end, session switch or after idle.
Privacy No board image, FEN, move, PV or game history is persisted. Only user settings are stored.
Cloud services None required for core recognition or analysis. No account, API key, remote model or hosted analysis service.

§ 4 — Built to fail safely

Unreliable board understanding is worse than no analysis.

A reconstructed position that silently substitutes an error for a fact produces confident nonsense. Uncertainty terminates the pipeline instead of flowing into the engine.

Unknown means unknown

  • No fabricated positions. If a board cannot be established reliably, analysis does not start.
  • No silent assumptions. When the side to move cannot be determined reliably, White is not assumed by default.
  • Transient state is discarded. Detection candidates and analysis state do not survive a scan or a session.

Detection never drives the engine

  • Detection and analysis are separated. The engine is only ever handed a validated state.
  • Competitive contexts gate execution. Recognised active competitive-game contexts can prevent the engine from running at all.
  • Fail-closed posture. The architecture is designed to fail closed in recognised competitive contexts and when critical board-state information cannot be established reliably.

§ 5 — Fair-play architecture

Intended for legitimate analysis — and architected accordingly.

ChessVision Engine is built for chess analysis, training, puzzles, completed-game review, accessibility, broadcasting and development use. It is not positioned as a competitive-game assistant, and the sale build contains safeguards designed to suppress engine recommendations when recognised active competitive-game contexts are detected.

These safeguards are a design commitment, not a universal guarantee: the architecture is designed to fail closed in recognised competitive contexts and when critical board-state information cannot be established reliably. No claim is made of absolute regulatory or platform compliance, and third-party sites evolve independently of this software.

§ 6 — Potential applications

Where the recognition architecture can go next.

No. 1

Chess training

Recognition and analysis embedded into educational workflows — from lesson material to interactive study.

No. 2

Post-game analysis

Reconstruct positions displayed in browser interfaces for completed-game review and study.

No. 3

Accessibility

Convert visually rendered board state into structured, machine-readable information for assistive tooling.

No. 4

Broadcasting

Extract board state from browser-based displays for analysis and presentation workflows.

No. 5

Developer tooling

Use the recognition architecture as the foundation for APIs, extensions or a reusable SDK.

No. 6

Chess platforms

License or integrate browser position-recognition technology rather than rebuilding it internally.

§ 7 — Acquisition

What is being acquired.

The package is structured so a buyer receives a complete, transferable software asset — subject to the eventual transaction agreement.

01
Proprietary source codeChessVision Engine extension codebase
02
Browser-extension architectureManifest V3 service-worker, offscreen and popup design
03
Recognition / detection modulesKnown-structure readers and the universal fallback layer
04
UI & analysis orchestrationPopup, overlay, ranking and lifecycle logic
05
Project documentationREADME, changelog and validation record
06
Validation / test materialsFixture harnesses and regression methodology
07
ChessVision Engine brandingName, mark and visual identity
08
Acquisition websiteThis site, as a standalone asset
09
chessvisionengine.comTransferable primary domain
10
Associated project assetsAdditional materials supplied through due diligence

Source code is not published here. The complete package is supplied exclusively through the acquisition and due-diligence process, on request.

Acquisition Listing Coming Shortly

§ 8 — Licensing

Proprietary project code, clearly bounded by third-party licences.

ChessVision Engine contains proprietary project code as well as third-party and open-source components. The bundled Stockfish chess engine is GPLv3 software and is not represented as exclusive proprietary intellectual property — it is included under its own licence, and that licence is shipped with the package.

The buyer receives the transferable ChessVision Engine proprietary assets and must continue to comply with applicable third-party licences, including the Stockfish GPLv3 terms when redistributing the bundled engine. Full licence texts are included in the release package.

§ 9 — Validation

What was actually tested.

The v1.6.0 release was validated with automated syntax, manifest, fixture and live-engine checks. Only tests that were performed are listed — no numbers are invented and no guarantees are implied beyond them.

CheckAreaResult
PASS Package integrity JavaScript syntax validation for all extension JS files; Manifest V3 JSON validation; ZIP integrity and re-extraction checks.
PASS Baseline preservation Behavioural regression harness compared v1.5.0 with v1.6.0 across Chess.com-style DOM reading, Chessground geometry, board discovery order and direct FEN attributes — identical outputs. SHA-256 checks confirmed unchanged baseline files.
PASS Generic browser fixtures 64 named squares, Unicode pieces, ARIA-labelled pieces, image/SVG piece naming, wK/bP-style codes, data-square/coordinate/file-rank attributes, white/black turn metadata, unknown-turn handling, generic board discovery, geometry placement, open Shadow DOM, 8×8 table boards.
PASS Integrated lifecycle harness A completely generic board with no site-specific selectors was discovered, reconstructed as the exact starting FEN, updated correctly after e2-e4, automatically re-analysed, stopped without further work and resumed correctly.
PASS Fail-safe & gating Side-to-move recovery, unknown-turn fail-safe, competitive-context gating and engine-request gating fixtures.
PASS Resource stress On a synthetic page of 20,000 unrelated DOM nodes, the universal reader completed three rounds of 300 find/read cycles with the JS heap approximately flat after forced garbage collection.
PASS Live engine smoke test The actual bundled Stockfish 18 JS/WASM executed a starting-position search, a 20-MultiPV pass exposing ranks 1–20, and a forced mate-in-one fixture returning score mate 1.

Automated harness and package validation is distinct from guaranteeing compatibility with every third-party webpage or future browser version. The v1.5.0 baseline is the user-tested live-browser build; v1.6.0 preserved those files byte-for-byte and exercised the new universal module in real Chromium DOM fixtures.

§ 10 — Engineering principles

How the codebase is built.

Layered

Independent recognition techniques complement rather than replace one another.

Local

Core engine analysis executes locally — no cloud inference in the analysis path.

Conservative

Unknown state is treated as unknown rather than guessed.

Maintainable

Detection, policy and analysis concerns are separated.

Transferable

Architecture, documentation and project assets are structured for acquisition.

§ 11 — Enquiries

Questions an acquirer should ask.

Is ChessVision Engine a chess platform?
No. It is primarily a browser chessboard-recognition and local-analysis architecture. The recognition pipeline is the asset; platforms are one of many things it could be embedded into.
Does it require a cloud AI API?
Core board recognition and Stockfish analysis do not require a paid cloud LLM service. The extension needs no account, API key, remote model or hosted chess-analysis service.
Does it work with arbitrary chess websites?
It is designed to recognise chessboards across diverse browser interfaces through layered generic detection, but no claim is made that every possible website implementation can be recognised. Encrypted canvases, inaccessible cross-origin frames and proprietary WebGL scenes are examples of what generic DOM reading cannot reach.
Can it be developed into an SDK?
Yes. The recognition pipeline — discovery, piece reading, state reconstruction and validation — is a self-contained module boundary that lends itself naturally to extraction into a reusable library or SDK.
Is Stockfish included in the proprietary IP sale?
No. The bundled Stockfish engine is third-party GPLv3 software and is not represented as exclusive proprietary intellectual property. The sale transfers the ChessVision Engine proprietary assets; the buyer must comply with applicable third-party licences.
Is the software intended for competitive-game assistance?
No. ChessVision Engine is intended for legitimate analysis, training, puzzles, completed-game review, accessibility, broadcasting and development use. The sale build contains safeguards designed to suppress engine recommendations when recognised active competitive-game contexts are detected, and the architecture is designed to fail closed in recognised competitive contexts.
What assets are transferable?
The proprietary source code, the extension architecture, recognition modules, UI and analysis orchestration, documentation, validation materials, branding, this acquisition website and the chessvisionengine.com domain — subject to the eventual transaction agreement. The complete package is supplied through the acquisition and due-diligence process.

Acquisition

Acquire the technology, not just the interface.

The buyer receives an established browser chessboard-recognition architecture that can become the foundation for analysis, training, accessibility, broadcasting or developer products — with the domains, branding and documentation already in place.

Acquisition Listing Coming Shortly Explore the technology