Keep the question.
Test the alternatives.

Start with a question.

1. Open Game library, search a player or opening, and select a game. Or use Input for a FEN/PGN and Load PGN for a local file.
2. Step through the game with ← / →. Open DCC analysis to compare the available candidate moves and their measured continuations.
3. Open Study to keep alternate lines, write a question and pin an A/B comparison.
4. Use Deep analysis to challenge the position with local Stockfish, and Evidence to preserve the source data behind a DCC result.

Game library starts with seven curated Top Picks; All collections opens the full library. Marked games open at a verified interesting position, and the player names remain above the board. Use the history arrows to review the game, or move pieces deliberately to explore a variation. Open Review game for a guided overview of the recorded game.

More opens Workspace tools: FEN/PGN input and export, board orientation, Settings, Study, Evidence and Benchmark. The three resource buttons below them link to PGN picks, Why DCC? and this Help.

Board analysis source offers CDB → SF and SF. It selects the board badges and left evaluation bar. The three clickable cards below the board stay visible in both modes: CDB refreshes the database result, SF starts deeper analysis or stops its current search, and DCC switches Moves/DCC. The separate DCC checkbox controls board details. Choosing SF does not disable remote CDB requests.

Ordinary SF analysis defaults to depth 11; a previously saved depth is retained. Analysis adds two plies for this position, or stops a running search. The bar retains the previous position’s score while waiting and identifies it in its description; it is replaced when the new score arrives. Cards and the bar use White’s perspective; DCC candidate rows identify their mover perspective.

Review the recorded game.

After loading a game, choose Review game beside DCC analysis. Use the slider to visit any recorded position, or select a listed moment to stop before its next move. Show on board opens the position; Study / A-B takes that same position to Study, where you can investigate alternatives. The Review drawer closes when you choose a position and opens again with the Review game button.

Highlights come from the game’s PGN comments, annotation glyphs and verified curated positions. Plain navigation checkpoints help with PGNs that have no comments. A checkpoint is not an engine finding, and a source annotation describes what its author wrote, not a newly verified ChessBest verdict. When you branch by moving a piece, Review follows your current line and hides markers that belonged to the original line.

The current-position CDB, SF and DCC cards in Review use only snapshots for the exact board position. CDB and SF scores are shown from White’s perspective. Recorded move vs source choice shows the next played move; its centipawn gap is calculated from one source only if that source scored both moves with comparable exact centipawn values. If either is missing, the gap stays unknown. A DCC number is a heuristic measure, never a centipawn score. Review does not score every move or draw a whole-game evaluation graph; the separate DCC replay operation analyzes recorded moves and may take time and source requests.

For TCEC games, Review shows the engine names, opening, round and termination only when the PGN supplies them. Adjudication is identified by its recorded termination details; it is not presented as a played checkmate. Historical engine scores, PVs and a paired opening game require source records and are not inferred from the game result.

More room for the position.

Sim Focus automatically folds the lower tools during a local automated Sim. The central moves/DCC area receives the saved space on desktop and phone. Pause and Tools stay at the bottom. Opening Tools is an overlay; results arriving in the background do not move its buttons.

In Settings → Workspace & simulation, choose Always expanded to retain the full toolbar. A wider workspace is available on wide screens, and Compact analysis spacing reduces reading gaps while keeping usable touch targets. Clocks remain at the top and navigation immediately above the reading area.

The workspace does not automatically scroll to a new move. Your reading position is retained as the list updates. Settings and Game library open over the reading area and close with × or Esc.

The bar beside the board always uses White’s perspective: +1.00 favors White, −1.00 favors Black. It follows the board orientation when flipped. Its height is a compressed score scale, not a win probability. … means waiting, — unavailable, # an actual checkmate, and W/B a decisive database score without an invented mate distance. Hide Eval hides its score too.

Choose how to play.

Click a history move to pause Simulation on that exact position. Pause stops at the current position. The next Simulation opens the engine choices again, so you can change either side before continuing. New game interrupts local Simulation, a game against the engine, or DCC replay and resets immediately; late responses cannot play on the reset board.

Optional event pauses are in Settings: a different CDB/DCC choice, a selected evaluation change in centipawns, or missing required analysis. These are observations for study, not time limits. Connected Lichess games retain their own server state and clocks.

Keep the branches, not just the final line.

Study opens three views: Variations, A / B comparison and Preview. A study keeps alternate moves and comments as a tree. Starting another study retains the earlier ones.

Click a move written in normal chess notation, such as Nf3, to preview it on a separate board. Your main workspace stays where it was. Choose Open position in workspace when you want to continue from the preview.

Import one PGN game at a time with its nested variations, comments and annotation glyphs. A saved import does not replace the current workspace position. PGN + variations exports the active study; use this study export when preserving branches matters.

Pin a fair A/B comparison

Write your question, then pin two candidates from the same starting FEN as A and B. Candidate previews offer Pin this candidate as A/B; Pin workspace analysis as A/B stores the displayed analysis. The comparison keeps its captured CDB, SF and DCC measurements, their source labels and position context as you explore other lines.

Up to 20 studies are saved in this browser. Export all studies + A/B JSON preserves the collection and comparisons for backup or transfer to another computer. The website is available across devices; browser study storage is local to each browser, with no account synchronization.

Manage Studies is available from Study, including before the 20-study limit. When Studies are full, the manager keeps the pending save and offers export, a searchable selection of Studies, confirmed selected/all removal, and Continue pending action once enough slots are free. Export alone does not free space. A verified local recovery copy allows Undo last removal; if that copy cannot be written, removal needs a separate explicit decision. A/B and the main board are retained.

A collection import is validated and committed as one operation. A quota or storage failure is reported as unsaved, not as a successful import. Concurrent tabs coordinate Study writes; a conflicting edit requires reviewing the current version. Pending work can be exported. A pending draft survives reload only when its local write was verified; the restored draft must be resumed explicitly.

On first use of this LAB release, existing browser data is copied into LAB-only storage and verified. Existing LAB data takes priority; the original CURRENT records are neither moved nor deleted. Separate names do not provide a separate browser quota. If copying fails, the page explains the problem before starting the workspace. Keep exported backups for portability.

Test the choices where they are close.

1. Choose a position, open Simulation, and set White to CDB/SF and Black to CDB/SF + DCC, or the reverse.
2. Read the comparison card: raw first choice, DCC choice, equally scored candidates, and the gap in raw centipawns. A 0 cp gap means the provider gives the two choices the same score.
3. Pause on a relevant move, pin the alternatives in Study, or return to the experiment’s start and run another branch.
4. Use Position experiments → Return to start, PGN and CSV exports to compare the same starting position with recorded engines and settings.

CDB top 1 preserves the provider’s order within tied raw scores. DCC is also calculated for comparison when CDB is the selected player. The observed 10 cp choice band is fixed; a wider inspection window does not widen it.

Exports record the starting position, engines, decisions, coverage and timing. An unfinished or paused experiment keeps result *. A different selection alone does not show an improvement; investigate the resulting lines and preserve the source snapshot.

What DCC contributes.

DCC means Dynamic Complexity Controller. ChessDB supplies engine evaluations. DCC uses the observed shape of candidate continuations to rank close choices and decide where additional inspection may help.

The default Balanced policy includes every available legal candidate within 10 cp of the best observed raw score. The Extra inspection pool provides room for additional candidates inside the wider window; it does not truncate the near-tied group. The Legacy policy retains the earlier capped pool for comparison.

Balanced scheduling gives candidates an initial measured pass before spending more requests on longer continuations. Alternative defensive replies can be inspected too, so a favorable main line is challenged by more than one sampled response. The reply count and continuation horizon are configurable. This is bounded branch inspection, not exhaustive search of every defence.

Individual ranking sensors can be switched off for comparisons. Descriptive observations do not all contribute separate weights. In particular, structure affects ranking only when its experimental mode and sensor are both enabled. Use Benchmark to inspect the effect of a change rather than assuming an extra sensor helps.

Attack and defence relations can also suggest questions tied to a measured legal line, such as an undefended target or a defender with several duties. These describe geometric relations; pinned pieces and tactical exchanges still need concrete checking. The questions do not add an unverified rank bonus.

From Sudoku to chess

The transferable idea is controlled allocation of work: begin with observations, deepen unresolved comparisons, keep missing data explicit, and retain a raw-engine fallback. Sudoku results do not establish chess strength. When required evidence is incomplete, DCC keeps the raw choice rather than inventing a confident override.

Different numbers answer different questions.

A ★ marks the shared DCC choice and the blue badge marks raw best. A favorable trend describes the sampled line; it does not prove that an opponent has no stronger response.

Read the gaps as carefully as the scores.

A supplied principal variation is a possible continuation. It does not mean that every position has a separately measured score. Read the observed half-moves, funded horizon, requested horizon and defensive-branch coverage together.

Unknown is not zero, a draw, or stability. Invalid replies, unavailable positions and missing samples remain visible gaps. A complete sampled line does not mean every legal candidate or defence has been covered.

DCC half-moves and engine search depth are different. Five DCC half-moves means sampling five moves by alternating players, including the candidate. A source PV may report its own search depth. Other source replies do not report depth or nodes, so those fields remain unknown. Scores obtained at different times or depths can vary for reasons unrelated to the position’s practical stability.

Challenge the position with local Stockfish.

Deep analysis runs Stockfish 18 Lite on your device with an explicit search budget. The same local engine also supplies the SF comparison card during ordinary analysis, so its resources may already be loaded. No local installation or engine API key is needed.

1. Open Deep analysis to pin the current position.
2. Choose Search budget: depth 14 or 18, 250,000 or 1,000,000 nodes, or Until I stop. Choose 1, 3, 5 or 10 lines.
3. Optionally enter Only these first moves, for example e4, d4, Nf3. Leave it empty to search all legal moves.
4. Click a SAN move to preview its variation. Use Save line to study to retain a line, or Export analysis to save the measurements.

The engine stays with its pinned position even if the main board changes. Use current board position explicitly updates it. Stop retains partial results; an unfinished search is not shown as a finished one.

Lite uses one CPU thread and a compact network suitable for browsers. Actual speed depends on the device; a phone can use the same tools with smaller budgets. Its evaluation is separate from the ChessDB score and DCC rank. Build, corresponding source and license.

Keep what the decision was based on.

Evidence → Save current analysis freezes the full FEN, settings, DCC decision and captured source replies. The summary shows capture time, available source timestamps, version, reply count, dropped collector records and a SHA-256 identifier. A timestamp records collection; it does not establish when ChessDB originally computed the evaluation.

Restore offline loads a snapshot and reuses only captured replies. Uncaptured answers stay unknown; offline replay must not silently fill them from live ChessDB. Choose Return to live data when you want fresh queries again.

Export evidence and Import evidence JSON transfer a snapshot between devices. Storage holds up to 30 snapshots within a 16 MB limit in this browser. If durable browser storage is unavailable, the panel says that the snapshot is temporary; export it to keep a copy.

Reusing the exact evidence helps distinguish a policy change from a changing database. The identifier checks the recorded content; it does not certify that the source’s chess assessment is correct.

Compare methods with the budget visible.

Compare policies

Replay a saved snapshot through raw choice, Legacy, Balanced and sensor variants. Each receives the same logical query allowance; actual requests and unobserved queries are reported. This controls requests, not ChessDB’s unknown underlying search work. A policy asking a new question cannot obtain an answer absent from the snapshot.

Run diagnostic fixture uses synthetic frozen data to check reproducibility and coverage handling. Its result is a software diagnostic, not a chess-strength result.

Engine experiment

Compare raw browser Stockfish with an experimental DCC allocation using the same total requested node budget. The experiment divides its budget between candidate discovery and continued inspection. Actual nodes are reported because engine limits can overshoot slightly.

A separate larger Stockfish search then compares the selected first moves. Its reported loss is against the best of those compared moves, not necessarily the globally best legal move. Deeper analysis by the same engine is useful evidence, but is not an independent engine verdict or proof of improved Elo.

Use Stop comparison to cancel and Export comparison JSON to keep the configuration, budget and results. Several repetitions of one position remain one position’s evidence; a playing-strength study needs a broader controlled sample.

Tune the view and the workload.

Reset Settings restores LAB preferences. More candidates and deeper defensive checks need more source requests. A simulation’s limits come from its selected mode: an untimed observer run and a match configured with a game clock have different timing rules.

Time, timestamps & two humans.

Player timers are on by default at the top; timestamps are off. In Settings, toggle Show player timers and Show move timestamps independently without pausing Sim.

Untimed observer Sim counts elapsed time and waits for planned analysis until it completes or you pause/reset. This describes the untimed run, not every configuration reachable from Sim.

Simulation → Matches & tournaments supports depth, root-node and per-move time budgets, or a player clock with increment. In clock mode the recorded result can include a flag fall. Engine initialization and the selected display pause are excluded from the player clock. The archive keeps analysis time, pause, remaining clocks and the actual stop reason; a source failure is reported explicitly.

For a duel or round robin, choose the Current board position or a Game library collection as the opening source. A collection can provide several opening positions. Auto uses opening metadata when available, otherwise up to nine full moves; Choose move count lets you set the opening length. Paired formats swap colors for each opening. Review the planned game count before starting.

History & standings holds saved events with Pause/Resume, a table of completed games, head-to-head results and PGN/CSV/JSON exports. Open game reviews a recorded game. Saved games also appear in their own Game library collection, where you can search and load them. Import JSON restores a compatible exported event in this browser; it is separate from importing a Study.

Settings → Two players · same board lets two humans play on one device, with optional time and increment. CDB+DCC reviews the positions in the background. Looking back or choosing Pause game pauses the local clock; zero offers Continue untimed without inventing a tournament result. This mode has no remote rooms or matchmaking.

Timestamps use the browser’s local timezone in the display and UTC in exports. Historical PGNs do not receive invented move times. Games against the local engine show elapsed time; connected Lichess games retain server clocks.

Ask about the current position.

Click ✦ beside YOUR WORKSPACE and ask in Slovenian or English. A question includes the current position, legal moves, recent history and available CDB, SF and DCC measurements with their actual sources. Its answer is marked with the position it refers to if the board advances.

Gemini receives instructions explaining DCC, score perspective and missing data. It can explain and suggest questions, but does not move the pieces or replace engine verification. Chat remains in this tab; Clear removes the displayed history. A service quota error is shown explicitly rather than retried in a loop.

The temporary GitHub Pages release cannot run the Gemini assistant service. The assistant needs a deployment with its server endpoint to return answers; the board, Study and local analysis remain available here.

Stay on the keyboard.

Navigation keys do not move the main board while you are typing in a field or using an open dialog. Focus returns to a visible control after closing Tools.

Install this chess workspace.

Open CURRENT for the stable release or LAB for the development release, then choose Install LAB on this page. On iPhone or iPad, open the page in Safari, tap Share → Add to Home Screen and confirm. Each page installs under its own name and receives its own offline cache. The old separate PWA link now opens LAB.

The board, bundled games, saved studies and local Stockfish are available offline after the first complete install while online. ChessDB, Lichess and Gemini still need a connection. Hover over or focus the LAB link for its offline information; on this page it says Offline ready when LAB's own worker controls it. If the download cannot finish, reconnect and reopen the page. Offline files occupy browser storage alongside your studies; export studies regularly.

When Update ready · reload appears, finish your current analysis and choose the button when you are ready. The new release waits; it does not reload an active game by itself. CURRENT and LAB keep distinct data keys, though they share browser storage quota on the same site. Installed apps on different devices or in separate browser storage do not synchronize; export and import to transfer data.

When an answer is missing.

No database evaluation? The position may be absent from ChessDB or a request may have failed. Continue studying the board, or try Deep analysis. A longer DCC horizon cannot manufacture missing database data.

Offline replay has gaps? The snapshot may not contain replies needed for the new configuration or position. Keep the gap for an exact replay, or explicitly return to live data.

The engine cannot start? Read its loading status. Local Stockfish needs a browser that supports WebAssembly and workers; other study and ChessDB features remain separate.

A study is missing on another device? Studies and evidence are saved locally to each browser. Import the exported study or evidence JSON there. LAB and CURRENT use separate workspaces. There is no automatic cross-device synchronization.

The library is empty? Clear the search and select All collections, or load your own PGN. A connected game is stale? Read its server-connection status before relying on an estimated clock.