Close evaluations.
Better questions.
Several moves can receive the same engine score while leading to very different positions. ChessBest helps you inspect those alternatives, challenge their continuations and keep the evidence behind your conclusions.
The engine and the controller.
ChessDB and local Stockfish 18 Lite provide source evaluations. DCC — Dynamic Complexity Controller — inspects how candidate lines behave. It compares measured stability, floors, volatility and trends, and allocates additional inspection among close choices. Alternative defensive replies help test whether a favorable main line survives another response. Raw source scores remain visible so the controller can be compared with the engine evidence rather than replacing it.
The default Balanced policy includes every available candidate within 10 cp of observed raw best. Its separate inspection pool can add other moves. A 10 cp guard constrains the choice relative to those source scores; it does not guarantee the objective loss under deeper analysis. Incomplete required evidence retains the raw-engine fallback.
Structure is descriptive by default. Its experimental ranking contribution and other sensors can be tested independently, so an added signal must earn its place through comparison.
Built around a serious analysis session.
Preserve the idea
Study keeps a variation tree, comments and a pinned A/B question. Preview a line before explicitly opening it on the main board. PGN and JSON carry the work to another device.
Challenge the line
Local Stockfish 18 Lite supplies a separate deeper search. Request several variations or restrict the first moves, then save a useful continuation back to Study.
Keep the evidence
Freeze the position, settings, source replies and DCC decision. Offline restore replays captured answers; a missing answer stays unknown until you explicitly return to live data.
Compare the method
Benchmark exposes policies, sensor variants, actual queries and node budgets. Synthetic diagnostics, frozen evidence and deeper engine experiments serve different purposes.
A workspace that stays out of the way.
Simulation supports CDB/SF, CDB/SF + DCC, SF and SF + DCC, chosen independently for each side. Run a single game, paired duel or paired round robin, or play against the engine. Supported local runs preserve their schedule, settings and committed moves for pause/resume and reload recovery; the browser must remain active for computation.
Sim Focus gives the lower toolbar’s space to the analysis while keeping Pause and Tools visible. Clocks stay at the top, reading position stays steady and the workspace adapts to desktop, tablet and phone. Width, compact spacing and optional event pauses are available in Settings.
Gemini can explain the current evidence and help frame a question. The engine and saved measurements remain the basis for checking a proposed continuation.
What a result can establish.
A different DCC choice is a finding to investigate. It does not by itself establish a stronger move, a higher win rate or an Elo gain. Measured coverage describes the sampled data; it is not confidence in a universal claim.
A frozen comparison makes policy differences reproducible. An equal-request comparison cannot equalize unknown ChessDB server compute. A browser-engine experiment reports actual nodes and uses a larger follow-up search, but its same-engine adjudication is still an estimate. These distinctions are visible so that conclusions can match the evidence.
Study collections and evidence are saved in the current browser, with export/import as the portability path. There is no account or automatic cross-device synchronization, so exported backups remain the reliable way to move or preserve work.
Start exploring.
Open ChessBest, or follow the Study & A/B guide. For a controlled comparison, see Benchmark. The browser engine’s build details, corresponding source and license are included.
Created by Bojan D. with AI collaborators, as part of BD × AI Lab’s MDL×DCC research.