Kasparov versus the World — computer analysis

Chessgames.com game link

Updated 2026-05-09 — Stockfish at very deep ply

Fresh analysis with Stockfish in 2026.

The three decisive moments, with diagrams

Analyst scorecard at the decisive moments

The four expert analysts who advised the World Team in 1999 — Irina Krush, Elisabeth Paehtz, Florin Felecan and Etienne Bacrot — each posted a recommendation per move. Krush was the principal expert advisor; the World Team usually but not always voted in her recommendation (early-game exceptions: moves 3 and 6). Cross-referencing those recommendations against the 2026 Stockfish verdict, the picture is striking.

In July 2026 I added seven further contemporary voices, recovered from the Internet Archive and from my own 1999 files (muted columns; sources and highlights below the table): the St Petersburg GM Chess School; correspondence GM Roberto Alvarez’s Ajedrez de Estilo; Barnet Chess Club; the Computer Chess Team (the volunteer engine-analysis group I belonged to); the Finnish IM Antti Pihlajasalo, who posted as “IM2429”; Canadian GM Duncan Suttles, who posted on the MSN board in person; and IM Georgi Orlov’s chessplayer.com column. Rows beyond the original seventeen exist because at least one source recommended something other than the played move. The four analysts’ cells in those rows come from their complete per-move commentary, preserved with each move’s voting results in gameanalysis.txt (originally served at fdl.msn.com/zone/kasparov/gameanalysis.txt); “(absent)” means no pre-vote recommendation survives, and “(none)” that the source explicitly made none.

Green = drawing or winning; red = losing; shaded cell = the move the World Team played. Move 16 is included because it is the only position where any analyst found a move strictly better than the played one. Endgame moves 51–53 are after the game became theoretically lost (Krush's recommendations there are in a position where 51… b5 had already let White win); the colours show whether the recommendation itself was a tablebase draw or loss in isolation. Evaluations in the added rows and muted columns are Stockfish depth 24–67 rather than the 60–77 used elsewhere; the verdicts for 32…b4, 33…Bxg3, 35…Kd5 and 39…Ke4 are tablebase-assisted runs (July 2026), quoted at the depth reached — the losing evaluations were still rising, while 35…Kd5 is flat 0.00 at depth 67. The move-58 and 60 verdicts are 6-man tablebase results — after 54…b4 55. Qxb4 every continuation was lost, so 58…Qf5, which contemporaries believed drew, only resists longest (tablebase distance 65 versus 7 for the played Qe4) — and row 62 records the final play-on-or-resign decision in that lost position.

Tally per analyst (across the seventeen original rows):

• Krush: drawing at 6 (g6, not voted in), 7, 10, 15, 16, 18, 19, 21, 25, 26, 32, 36, 51, 53; losing at 37 and 54. The two times Krush picked a losing move are exactly the two times White had immediate winning options.
• Paehtz: drawing at 7, 10, 15, 18, 21, 26, 51, 52, 54; near-equal at 6 (Ne5, +0.49); uniquely better than the played move at 16 (Ra5, −0.15) and uniquely correct at 37 (e5); losing at 19, 25, 32, 36, 53.
• Felecan: drawing at 6, 7, 10, 15, 18, 25, 26, 32, 52, 53, 54; near-equal at 16 (Nd4, +0.27); losing at 19, 21, 36, 37, 51.
• Bacrot: drawing at 6, 10, 15, 16, 18, 19 (uniquely Qd4), 21, 25, 26, 54; near-equal at 7 (Ne5, +0.26); losing at 36, 37, 52. Absent at 30–33 (winning the French Championship at Besançon) and at 51 / 53 (Beliavsky match in Albert).

Two structural observations:

• At the two losing moments — 37… e6?? and 54… b4?? — Krush picked the losing move. Paehtz uniquely found the only drawing move at move 37 (e5); at move 54 the other three analysts (Paehtz, Felecan, Bacrot) all found drawing moves.
• At the precarious move-36 turning point, the order reversed: only Krush's Kd5 holds among the four analysts; Paehtz's Nb4 and Felecan/Bacrot's b2 all lose by force. Among the archived sources, Barnet's 36…Bc3 — credited in their changelog to GM Duncan Suttles — also holds (+0.00 at depth 55); they recommended it believing Kd5 lost.
• At move 58 the analysts split two against two: Krush (“I recommend the World plays the move 58…Qf5”) and Felecan against Paehtz and Bacrot’s Qe4 — the vote went with Qe4. Every move lost by then, but Qf5 resists far longest. At move 61 Felecan recommended resignation; the World played on one move more, then resigned with 51.01% of the final vote (32.08% wanted to try 62…Qc6+ — the Alvarez/Suttles instinct).

The archived columns. The St Petersburg GM Chess School posted a page per move on gmchess.spb.ru — e.g. move 15, move 26, move 58. Roberto Alvarez's recommendations are from his per-move pages and post-game scoresheet on Ajedrez de Estilo (schach.w3.to) — e.g. move 15, scoresheet. Barnet Chess Club's are from their move pages and the dated changelog of their annotated PGN. Highlights: GM School and Alvarez both backed 26…Bc5, which lost the vote to f4 by 46.61% to 46.17%, and both recommended the drawing 51…Ka1 against the played 51…b5. At the two fatal moments they fared no better than Krush — both picked 37…e6, and GM School 54…b4 — though both then found 58…Qf5, the toughest defence in a lost position. GM School stopped recommending after 58…Qe4 (“White wins in all lines”), and for move 3 declared they recommended “NOTHING!!!”. The Computer Chess Team (eGroups “computergang”, 83 members) posted signed recommendations on the MSN board from move 10 — I was a member, and in July 2026 I recovered the list’s complete traffic (1,005 messages) from my own 1999 mailbox, giving the team’s full official record. The coordinator “-gts” stated the rule: “If the majority of our chess programs like a move then we recommend that move”; ties were broken by Krush and the GM School. The team’s 19…Be5 lost the vote to Qb4 by a nose (35.09% to 33.85%); its 51…Ka1 was a genuine original (“the first move we suggested that Irina recommended”) and lost amid a confessed ballot-stuffing episode while MSN’s server locked the team’s spokesman out for twenty minutes. At 33 the official was b4 — the Bxg3 flirtation was engine churn and my own dissenting vote — and at 58 the vote-day post equivocated toward Qe4 (“our deepest crafty analysis pointed to Qe4, though it was obviously suspect”) after the 7 October mainline had given Qf5; Gawthrop’s reply stands: “Live or die move. Nothing in the database pointed to any other conclusion than Qf5.” IM2429 was Antti Pihlajasalo (IM, Finland, FIDE 2429), who analysed with Crafty on a Pentium 233 for the World side; he was outvoted on all four fatal endgame decisions — 51…Ka1, 52…Kc1, 54…Qd3 (“do NOT vote 54…b4?!”) and 58…Qf5 (“all other moves are PROVEN wins for white”) — and left the board after 58…Qe4 was voted in. GM Duncan Suttles posted under his own name from Vancouver until sustained flaming drove him off around 8 September; his picks were almost all off the consensus path (his own words on 36…Bc3: “not one of the analysts even considered this move!”), and his 29…Qe2 evaluates equal at depth 40, while his too-late 35…Kd5 is confirmed drawn (0.00 at depth 67 with tablebases); his 33…Bxg3 and 39…Ke4, though, lose (+4.4 and +5.8). His move-27 suggestion was relayed after the game by Pihlajasalo as “27…b4!?”, which is not a legal move in that position; the table reads it as 27…b5, the only b-pawn push available, since Suttles’s original post is lost. Orlov’s chessplayer.com ran a short report on essentially every move (36 archived articles) but stated a pick or prediction for only about a third of them; at move 37 he wrote “Black has only one play, 37…e5” — making him, with Paehtz and Suttles, one of only three voices on record for the actual drawing move — and after the game he argued 51…b5?! was the decisive mistake, with “51…d5” deserving attention. Their move-6 page is a curiosity: it formally “recommends 6…Nf6 although 6…Nd4! is better” — a “stunning novelty” they declined to put their name to only because “there is no link from MS Zone to our consultation page so this move wouldn't be chosen anyway”, adding “feel free to vote for 6…Nd4! — it's worth it”. The table takes Nd4 as their effective recommendation; Stockfish is less taken with it (+0.44 at depth 40, behind both Nf6 and g6).

The single non-decisive moment where the World Team had a strictly better choice than what was played — 16… Ra5 at −0.15 — was also Paehtz's recommendation.

The two near-equal positions worth a diagram

Other candidate moves — sorted by move number

Sorted by move number, with the played move marked. The leftmost column is the centipawn evaluation; depth in plies and wall-clock time per group are noted at the end of each line where given.

Move 6 (Black): analyst picks

Krush's g6 and the played Nf6 transpose into the same position after 8. Nxd4 Nf6 9. f3 (Krush's pick is essentially identical at +0.19 vs +0.20). Paehtz's Ne5 trades off Black's c6-knight, doubles White's f-pawns, but gives White a near-half-pawn pull from the broken pawn structure.

Move 7 (Black): analyst picks

The played 7… g6 (recommended by three of the four) equalises; Bacrot's 7… Ne5 leaves White a quarter-pawn pull but still drawn. Note the move-7 PV foreshadows the actual game continuation up to move 12.

Move 10 (Black): three equal alternatives

Move 18 (Black): five equal alternatives

Krush rejected 18… Bd4, e6, and Nd4 in detail in her 1999 commentary. At depth 60 all five alternatives hold; her tactical refutations were depth artefacts.

Move 19 (White): all five tried moves are equal

The 2010 Rybka finding that 19. Bd2 was significantly better (+0.56) does not hold at this depth.

Move 19 (Black): four candidates — Nd4 loses

Move 21 (White): all five tries equal

Move 21 (Black): Felecan's Rh8 loses

Older engine runs preferred 21… f4 over Rxa4; at this depth both are equal.

Move 25 (Black): Paehtz's d5 loses

Move 32 (Black): Paehtz's f3 loses

Bacrot was at the French Championships and didn't post commentary at move 32.

Move 34 (Black): both bishop moves draw

Bacrot did not post commentary at move 34 (he was competing at the French National Championships).

Move 35 (Black): three drawing moves

Move 36 (Black): the precarious turning point

The played 36… Kd5 and the long-recommended 36… Bc3 are only at depth 34 here. Earlier 6-piece tablebase analysis (2017) showed both lead to a draw provided Black follows up with the unique 37… e5.

Moves 51–53 (Black): Q+P vs Q+P endgame — analyst splits resolved by tablebase

From move 49 (Black's d-pawn promotes) the position is in 7-piece tablebase territory and every claim can be checked deterministically. Three of the analyst recommendations in this stretch are confirmed losing.

Move 51 (Black)

Move 52 (Black)

Move 53 (Black)

Across the three moves, each of Felecan, Bacrot, and Paehtz has one losing recommendation, all decisive in tablebase. Bacrot returned to commentary after his French Championship win on 27 August.

Verdict

The game was decided on three moves: Black's 37… e6?? (the only losing move from the analyst suggestions), White's 38. h6?? (missing the only winning move 38. Rd1), and Black's 54… b4?? (with 54… Qd3 or 54… Qd5 drawing). Every other Stockfish-evaluated move on the critical list returns +0.00, with the single exception of Elisabeth Paehtz's 16… Ra5 at −0.15.