

20 Years of Engine-Assisted Chess and the Future of Competition on LSS
Lechenicher SchachServer · 2006–2026 · By Erik L. van Dijk & Ortwin Paetzold
A longitudinal analysis of 452,542 games played over two decades — the most comprehensive study of engine-assisted correspondence chess ever published. What the data reveals challenges every fashionable narrative about the "death of chess."
The headline finding is unambiguous: at the Master level (Category 4), standard correspondence chess has moved from a draw rate of 55.7% (2006) to 97.3% (2026) — an asymptotic march toward the theoretically drawn game. Yet we reject the conclusion that this constitutes a "draw death." Five deeper findings tell the real story.
The draw explosion is concentrated in Categories 3 & 4. Novices rose only to ~18.4%. The engine amplifies existing competence — it does not play for you.
If rising draws reflected cheating, the rating gap between strong and weak players would be widening. It is not. Everyone stands on the shoulders of algorithmic giants.
By multiplying starting positions 960-fold, it strips opening preparation. Masters draw only 54.2% in Chess960 vs. 97.3% in standard play.
No-Engine Masters sit at 41.5%; Thematic Masters at 82.4%. Each forces human error back into the equation.
Today the gap is small (95.5% vs. 97.3%), but this is an artefact of low FAST volume (8.5% of games). As volume migrates to FAST, the gap is expected to widen materially.
When Kasparov proposed "Advanced Chess" after his 1997 defeat by Deep Blue, he envisioned human intuition guiding machine calculation. Over twenty years on LSS, that vision reached maturity — and pushed past it.
The revolution was not engine strength alone, but engine democratization. A novice today, running free Stockfish on a laptop, commands more raw analytical force than the entire Deep Blue team possessed in 1997.
This collapsed the cost of chess education to near zero. Players now study, evaluate, and improve without expensive coaches or vast opening libraries. Each generation begins at a higher plateau — precisely Newton's "standing on the shoulders of giants," except the giants are now largely silicon.
The "draw death" debate appears on LSS in an unusually clean, measurable form, because correspondence chess removes the noise of clock-driven blunders. The rising draw rate is empirical proof of progression — our players are simply getting too good.
The most dramatic single finding of our 20-year study: Master-level standard correspondence chess has moved from a draw rate of 55.7% in 2006 to 97.3% in 2026 — a near-complete convergence toward the theoretically drawn outcome.

The asymptote is category-dependent — concentrated at the top. Category 4 (Masters) rockets from 55.7% to 97.3% while Category 1 (Novices) barely moves from 5.4% to 18.4%. The engine is a force multiplier on existing human competence, not a substitute. A Novice + the same engine still misnavigates complex positions — so decisive games persist. This directly refutes the "the computer plays it for them" narrative.

55.7 → 97.3% — practical solution reached.
18.2 → 46.3% — large rise from a low base.
5.4 → 18.4% — under one-in-five. Engine cannot compensate for foundational blindness.
At first glance, FAST barely differs from STANDARD at the top (95.5% vs. 97.3% in 2026). But this conceals a critical fact: FAST historically was the game players took on "on the side," never under genuine time pressure. The current near-parity is a sampling artefact of low FAST volume (8.5% of all games). As players migrate volume to FAST, the compounding Fatigue Factor will force errors and the FAST draw rate should fall away from STANDARD.
The most practical design tool for organizers confronting the draw problem is format selection. Each format attacks draw inflation at a different layer:

Practical solution. Opening prep + engine validation = near-zero error.
Partial cure. Unfamiliar structures force genuine decisions.
Opening preparation removed. Near coin-flip at the top.
Most decisive. Raw human calculation restored.

The moment engines are removed, human flaw and error re-enter immediately and decisive games flourish. The No-Engine Cup has held draw rates below 30% throughout its entire history — even as engine-assisted Master events crossed 90% and approached 97%. The gap has grown from +23.2 pts in 2006 to +55.8 pts in 2026.
Centaur chess is Formula 1: the engine compresses the error band and the result tends toward the drawn "racing line." No-Engine chess is the marathon: raw, unassisted, wider spreads, more decisive results. Neither is "real" chess and the other "fake" — they are different disciplines testing different things.
Within Category 3 alone, removing engines collapses the draw rate from 55.2% to 20.2% — a reduction of 35 percentage points within the same player pool. This is the cleanest within-category evidence of engine causation in the dataset.

The three-part inflation mechanism: Chess960 ablates opening preparation; Thematic reduces the comfort zone; No-Engine removes engine validation entirely.
As decisive games become rare at the top, the few that occur are increasingly won by White. In 2006, decisive Master games split roughly 54/46 in White's favour. By 2026 the ratio sits near 69%. The first-move advantage has not grown in absolute strength — the draw-saturated environment has concentrated decisiveness into White's column. This is itself an argument for formats (Chess960, Pentathlon) that distribute winning chances more symmetrically.

If the rising draw rate were caused by a minority secretly leaning on engines, we would observe a widening rating dispersion — the "cheaters" pulling away. We do not see that. The distribution of players across strength tiers is sticky; the spread between strongest and weakest is not growing.
The improvement is broad-based, not concentrated. Better engines, better free training tools, and better books lift everyone — most visibly at Category 4, but demonstrably at Category 1 too. This is the scientific model of progress: new generations standing on the shoulders of algorithmic giants.
Selected evidence from individual tournament series confirms the aggregate patterns. The LSS portfolio spans from flagship World Championships down to specialist No-Engine and Chess960 events — each a distinct laboratory.
Draw rate climbed 56.3% → 91.7% — the asymptote visible in our crown event. A near-perfect reflection of the category-level trend.
20.5% → 79.7% — a near-fourfold rise, the steepest proportional rise in any named series, illustrating how a broader-entry format responds to engine democratization.
Notably flatter — 47.0% → 39.9%, actually declining. Arena formats appear to resist the climb: a possible design lesson for organizers seeking decisiveness.
10.1% → 20.2%, never crossing 30%. With engines removed, draw inflation is modest and the game retains decisiveness across the entire study period.
Remarkably stable low draws — 7.8% → 14.7% across 2009–2026. Even with 17 years of accumulated Chess960 experience, the format continues to generate decisive games.
Quad Masters (>2200), 2026. Very small sample — illustrative only.
Novice 2-game (<1000), 2021. The truly unaided game.
The No-Engine Cup has never crossed 30% in its entire history.
If standard play at the top is approaching a solved, drawn state, the real design problem is how to adjudicate skill so championships aren't decided by a single freak blunder amid a sea of draws. Two forward-looking frameworks emerge directly from our findings.
Borrowing from gymnastics and figure skating, LSS can deploy high-depth engine analysis to score average centipawn loss (accuracy). A drawn game where Player A maintained 99.5% accuracy and Player B 98.0% could be scored in A's favour — rewarding superior quality without coin-flip tiebreaks.
To crown the best all-around chess athlete, the Pentathlon tests the same opponents across five disciplines, each probing a different faculty of chess mastery. A multi-format aggregate guarantees a high proportion of decisive games — Chess960, No-Engine, and Thematic drag the composite draw rate down sharply.
Deep research and engine guidance. Tests navigation of complex, engine-validated structures.
Time management and pressure calculation. The sleeping variable made active.
Pure creativity and opening adaptability. Masters draw only 54.2% — ensures meaningful results.
Structural understanding and versatility. Forces both colors into specific, unbalanced structures.
Raw, unassisted human calculation. At 41.5% draws among Masters — the purest test of chess mastery.

The rising draw margins on LSS are a testament to the progression of human-machine collaboration — broad-based, not corrupt; concentrated at the top, but lifting every tier. After 20 years and 452,542 games, the data tells a coherent story: our players are not cheating — they are improving. The engines have not killed chess — they have raised its floor.
The complete 20-year LSS corpus — the foundation of every finding in this report.
2006–2026: the most comprehensive longitudinal study of engine-assisted chess ever published.
The asymptote reached: practical solution of standard correspondence chess at the elite level.
Indicative paths forward for the next five years — all member-contingent, all community-driven.
"The evolution of perfection is not an ending. It is the most interesting beginning we have ever had."
— Erik L. van Dijk & Ortwin Paetzold
Lechenicher SchachServer · 20th Anniversary Edition · 2026

The Evolution of Perfection