After the standard eval badges load, the top 3 moves get a small arrow:

• +30 ↑ Green arrow (↑ rising) — the eval improves as you look deeper. This move leads into a position that gets better with best play from both sides. Safe choice.
• +28 → Gray arrow (→ stable) — the eval holds steady through depth. Solid, no surprises.
• +32 ↓ Red arrow (↓ falling) — the eval drops when the opponent responds optimally. The score looks good now, but there's a trap or tactical risk ahead.

The arrow tells you: "The score looks good NOW, but what happens NEXT?"

A colored border glow appears around badges that have lookahead data:

• +30 ↑ Green glow — the eval path through depth is compressible (measured by LZ76 complexity). This means the position is robust — few surprises, structurally simple.
• +28 ↓ Red glow — the eval path oscillates (high LZ76 complexity). Tactical minefield — one wrong move and the position collapses.
• +25 No glow — no lookahead data yet (still loading, or position not in ChessDB).

Each badge with DCC data shows a small shape symbol after the trend arrow. This is the ADSR signature — the same sensor that achieved ρ = −0.50 on Sudoku difficulty prediction, confirmed across TSP and Flip4M.

ADSR reads the eval path like an audio waveform: Attack (how fast the advantage grows), Decay (how much it drops after peaking), Sustain (the steady-state level), Release (how the line ends). From these four measurements, each move gets a shape classification:

• ▬ Sustained — flat, solid advantage through depth. The safest choice.
• ▲ Building — advantage grows steadily with depth. The position improves.
• ⚡ Spike — sharp gain followed by a fade. Looks great short-term but doesn't hold.
• ▼ Collapse — falls below starting level with best play from both sides. Dangerous.
• 〜 Volatile — wild swings up and down. Tactical chaos — outcome depends on who calculates deeper.
• ◆ Mixed — no clean pattern. Needs more depth or the position is genuinely complex.

ADSR tells you something LZ stability can't: the shape of how the evaluation evolves, not just how compressible it is. A sustained +30 is fundamentally different from a spiking +30, even if both have the same stability score.

When two or more moves have evaluations within 10 centipawns of each other (practically equal), one gets a gold ★ star. That's the MDL pick: the move leading to the most compressible resulting position.

Structurally simpler positions have fewer things that can go wrong. When the engine says two moves are equal, the one that produces a cleaner board state is the more robust choice. This is the founding hypothesis of the 8Z research program applied directly to move selection.

Click any badge that has DCC data (arrow visible) to see the full evaluation path in the info panel below the move list:

+30 → e2e4 +28 → e7e5 +32 → g1f3 +30
Trend: ↑ rising  |  Stability: 87% · d52  |  ▬ Solid — holds advantage through depth

This shows the principal variation (best play from both sides), intermediate scores, the overall trend, stability percentage, and the search depth from ChessDB.

When you load a PGN game and step through the moves, a small accuracy tracker appears:

ChessDB accuracy: 72%  |  DCC accuracy: 78% (34 moves)

This compares how often the player's actual move matched ChessDB's raw top pick versus the DCC stability-adjusted top pick. If DCC accuracy is consistently higher, it means looking deeper through existing data picks moves that align better with expert play.

Open Settings and scroll to the DCC Lookahead section:

• Enable DCC lookahead — toggle on/off (default: on)
• Lookahead depth — how many half-moves deep to analyze (1–10, default: 5)
• Top moves to analyze — how many of the top moves get the lookahead treatment (1–5, default: 3)

DCC runs silently in the background after each position loads. Higher depth means more API calls to ChessDB — if you're on a slow connection, try depth 2–3. DCC governance automatically stops probing stable lines early, so depth 10 doesn't always mean 10 calls.

For each top move, the system makes a single querypv call to ChessDB, which returns the full principal variation (best play line) in one response. Then selective lightweight queryscore calls probe key intermediate positions based on DCC governance rules — stable positions get fewer probes, unstable ones get more.

Total API load: ~3–12 calls per position (for 3 candidate moves), all read-only (learn=0). Results are cached in your browser's localStorage for instant replay.

Two buttons next to the DCC View toggle let you watch DCC compete against raw ChessDB live:

• SimW — DCC plays White, raw ChessDB plays Black
• SimB — DCC plays Black, raw ChessDB plays White

Both sides read from the same 48 billion positions. Same data, same API. The only variable is the selection algorithm: DCC uses stability, ADSR shape, and compression tiebreaking. Raw just picks the highest number. If DCC wins from both sides, it's the reading — not the color advantage.

At normal speed, you watch the game unfold on the board move by move. At max speed, the board dims and the simulation runs as fast as the API allows, then shows a stats dashboard: wins, draws, win rate, average game length.

Each game starts with 6 random moves from ChessDB's top-3 pool (to create variety between games), then switches to DCC vs Raw for the remainder. Configure the number of games and speed in Settings under DCC Lookahead.