ChessDB computed 48 billion positions. ChessDCC reads them intelligently.
ChessDB.cn is one of the most important resources in chess: 48+ billion positions evaluated at depth 50+ by Stockfish, backed by over 100,000 CPU-years of computation, 7-piece Syzygy tablebases, and continuous TCEC tournament feeds.
But ChessDB shows you a number. +30. Is that safe? Is it a trap? Will it hold when the opponent responds optimally? The number doesn't tell you.
ChessDCC adds the intelligence layer that answers these questions. It reads up to 10 moves deep through ChessDB's existing evaluations — not to compute new positions, but to understand the shape of how evaluations evolve with depth. Zero new Stockfish computation. Same data, smarter reading.
| Feature | ChessDB alone | + ChessDCC |
|---|---|---|
| Evaluation | +30 (one number) | +30 ↑ sustained ▬ 92% stable |
| Depth of insight | Current position only | 10 moves of best play traced |
| Move selection | Highest number wins | Stability + ADSR shape + compression |
| Tiebreaking | None (tied = tied) | MDL picks most compressible position ★ |
| Risk detection | Not available | ⚡ Spike / ▼ Collapse / 〜 Volatile warnings |
| Opening novelties | Shows what exists | Finds what's overlooked but robust |
| Self-testing | Not available | SimW/SimB/DCCT live simulation |
The core principle: the shortest description of the data IS the best model. In chess: a position that requires fewer bits to describe its evaluation path is more predictable, more robust, less likely to contain hidden traps. ChessDCC uses LZ76 compression (the same algorithm used in ZIP files) to measure the complexity of evaluation sequences through depth.
Not every position deserves the same depth of analysis. DCC governance decides where to invest: falling evaluations get deeper probing, stable lines stop early. The system governs its own resource usage — the same algorithm that managed 10 parallel workers optimizing a 3,496-city route for 60+ hours.
Borrowed from signal processing, validated across four independent domains (TSP optimization, Sudoku difficulty prediction, board game AI, chess). ADSR reads the evaluation path like a sound waveform: how fast does the advantage grow (attack), how much does it drop (decay), where does it settle (sustain), how does it end (release). Two moves at +30 can have completely different shapes — one sustained and solid, the other a spike that collapses. ADSR tells them apart.
The MDL+DCC architecture powering ChessDCC is the same mathematical kernel that solves traveling salesman problems (1.2% gap on 3,496 cities), detects structure in DNA (Z-scores 28–74), predicts Sudoku difficulty (ρ = −0.50), governs board game AI (60% win rate), and compresses audio beyond FLAC. Chess is domain #12. Software development is domain #12. The kernel doesn't change — only the alphabet does.
Everything ChessDCC does is built on top of ChessDB.cn, managed by Bojun Guo. Here's what makes the foundation extraordinary:
A single position at depth 50 takes many minutes on a strong PC. ChessDB has precomputed billions of them. That's centuries of computation delivered in milliseconds. Even rare openings and unconventional lines are covered. No human bias — pure computational analysis forms the opening book.
ChessDCC doesn't replace any of this. It reads it smarter. 48 billion positions were already computed. We just built the reader they deserve.