Reconstructing the Greek game, from the sources up

No complete rulebook survives. Every reconstruction is a chain of inferences, and the honest work is marking where each link is iron and where it is wire. So every rule below carries a tag for how the sources stand behind it:

§0Method, and why one game

Two standing commitments govern everything that follows. Primary sources outrank secondary reconstructions: where Bell or Schädler assert what the texts do not license, we drop it (§5). And we import nothing from a cousin game — Seega, tafl, Kharebga — except as a tiebreaker where the sources are silent.

Why we reconstruct one game, not two

An earlier draft built Latrunculi and Petteia as parallel reconstructions. Carried out honestly, they converged: once the unwarranted accretions are stripped, two equal-army custodial-capture games on a grid have nothing left to keep them apart. The historical record cannot sustain a stable difference, and the reception history proves the danger — every attempt to differentiate the Roman game (Bell's "Dux," the leaping king of the popular rule-sheets, a checkmate-style win) pulls it toward chess, while every attempt to keep the two distinct without such accretions collapses them back together. Two failure modes, one cause: there is only one game here.

So we anchor in Petteia / Polis and treat Latrunculi as its Roman reception (§4). The reasons are structural, not sentimental. The Greek game is the drift-resistant one: its conceptual core is more completely attested (board, pieces, capture, and victory are all Greek-fixed), and its defining feature — victory by immobilisation — is the one idea that cannot slide into chess, because blockade is the opposite of checkmate. The Roman game can be built clean, but only by continuously overriding its own reception; the Greek game is clean by nature.

On "reception." We do not claim a documented line of transmission — the evidence for direct Greek-to-Roman descent is sparse. We claim only the parsimonious reading: once anachronisms are bracketed out, the Roman material is most economically understood as the same game received into Latin culture and re-dressed, rather than as an independent invention that happens to share every structural feature. Latrunculi is conventionally "believed to be a variant of the earlier Greek game"; Schädler grants that polis and latrunculi are "largely the same." Reception is held here as the most economical hypothesis, flagged as such, not as fact.

§1The evidence dossier

1A · The Greek game — the canonical witnesses

1B · The Roman reception — corroborating and illustrative witnesses

The Latin sources are read here not as a rival game's evidence but as the receiving tradition's testimony to the same game — and, crucially, as the place where two questions the Greek record leaves open are answered in practice.

Archaeology. Boards survive at 7×7, 7×8, 8×8, 9×9, 9×10; bone / glass / pottery counters, uniform and distinguished only by colour (the Perugia hoard, the British Museum sets); the Vindolanda and Corbridge boards. The Stanway grave (Colchester, c. 50 CE) — rectangular (~2:3), possibly un-latticed — is the sole basis for the "Dux" theory and is plausibly a different (native British) game; we set it aside (§5).

Computational check (Crist, Piette, Soemers, Stephenson & Browne, 2024, via Ludii). Independent support for two of our calls: slide movement is playable and source-consistent; the large rectangular boards (8×12 Colchester-type) stall out under every custodial ruleset — evidence they were a different game, and that the real game lived on the squarer 7×8–9×9 boards.

1C · The witnesses in their own words

The tables above say what each source fixes; here is the source itself. A reconstruction that grades its rules by attestation owes the reader the attested words, not just a citation to chase.

The Greek game

The Roman reception

Further witnesses — corroborating the keystone

Pollux IX.98 is the keystone, but it is not alone. Three later or lexicographic witnesses independently confirm the two things most of the reconstruction hangs on — that the board is a polis and the pieces are "dogs," and that our game is one of placement and position, not of dice. They are paraphrased here with their references.

§2What the sources fix vs. leave open

1. Board is a grid. A Pollux; Varro. Not in dispute.
2. Capture = custodial enclosure. A The best-attested rule of all — Pollux, confirmed by Ovid and Martial. Bedrock.
3. Two players, two colours, otherwise-identical pieces. A Pollux ("dogs… two colours"). A single named piece-type, no hierarchy.
4. No special / leader piece. I strong. No Greek or Roman text mentions a second piece type. The Dux rests entirely on the Stanway arrangement; we treat all pieces as equal (§5).
5. Victory by immobilisation / blockade. A Plato, Aristotle, Polybius make cornering and being unable to move the defining image. Foregrounded as the primary win. Material reduction is a secondary path I, and a material tie-break decides a stalled game I/D — the plain sense of ad incitas redigere.
6. Board size. D Open — no Greek dimension survives, and the Roman archaeology spreads across 7×7, 7×8, 8×8, 9×9, 9×10 with none privileged. 7×7 default — equally attested, the squarest of the small boards (inside the computationally-supported 7×8–9×9 range, §1B), and deliberately neither the 8×8 chessboard nor a Go-sized grid, so the game reads as a thing of its own. Size remains tunable; a 9×9 array is first-player-skewed (§6), and board size does not change the blockade outcome (§6C).
7. Opening — placement. I Nowhere stated in Greek. But the receiving tradition's richest source puts play on the open board (Laus Pisonis), the opposed-rows array is "a chess analogy for which there is no authority" (Wayte, CR 1892), and self-play makes placement the fairest, most varied opening. The Roman reception thus fills a Greek gap: open-board placement, captures suspended during it.
8. Movement — slide. I Nowhere stated. The reception again supplies it: the Laus Pisonis' "breaking the line" reads as ranged movement, and single-step movement is computationally degenerate (games never resolve; §6) while slide gives a living game. Slide (rook-like).
9. No suicide. I Capture is mover-initiated; moving into a pre-existing bracket is safe. Universal in the family and required to function.
10. Multiple simultaneous capture. I One move closing two brackets takes both. (Echoed by "itself checks two.")
11. Corner capture. D A corner piece taken by bracketing its two approaches. Near-inert in play (§6); on by default, toggleable.
12. No freeing in the canonical game. I No Greek source supports a rescue mechanic. Freeing belongs, if anywhere, to the Roman reception (§4) — and even there is weakly attested and demonstrably pacifying (§6).

§3The reconstruction — the rules

This is the game published as Pessoi. Below, the historical name petteia / polis marks the tradition; the rules are the playable core.

Players 2.  Board open square grid, 7×7 default, size tunable D (§2.6).  Pieces the "dogs" (kynes) — identical, two colours; player-agreed count, 12 per side a clean default on 7×7, roughly half the city occupied D.

§4Ludus Latrunculorum as Roman reception

We read the Roman game as the same game in Latin dress. The hypothesis is economical rather than documented (§0), but it earns its keep three ways.

It shares the entire canonical skeleton. Grid (Varro), equal two-colour pieces (Martial), custodial capture (Ovid) — point for point Pollux's game.

It supplies the two rules the Greek record leaves open. This is the strongest reason to keep the Roman material close rather than discard it. The Greek sources fix the concept (board, pieces, capture, victory) but are silent on opening and movement — exactly the two open choices. The receiving tradition's richest source, the Laus Pisonis, answers both: play on the open board → placement; "breaking the line" → slide. The reception is not decoration; it is evidence for the canonical game's two underdetermined rules.

It corroborates the victory condition. The Latin idiom ad incitas redigere — "reduced to immobility," i.e. beaten (Plautus, Lucilius; the terminology via Isidore) — independently confirms, from the Roman side, that immobilisation is defeat, the very thing Plato and Polybius make central.

Roman-flavour variants (offered, not canonical): the military vocabulary (latrones, ordinarii, the vagus → ordinarius progression); and the freeing / incitus mechanic — a bracketed piece frozen for a turn and removed next turn unless a flanker is itself trapped. We treat freeing as a possible Roman accretion only: it rests on Schädler's reading of a Senecan metaphor (not an explicit rule), it has no Greek warrant, and under skilled play it nearly abolishes combat (§6). It is the most likely single point of genuine Roman divergence — and precisely the kind of differentiating accretion that, pushed further, becomes chess. We log it and leave it off by default.

Honest caveat. Direct transmission is sparsely attested. A reader who weights embodied texture over parsimony could instead anchor in Ludus (for the Laus Pisonis and the archaeology) and treat petteia as ancestral background — equally coherent, but it trades drift-resistance for richness and demands permanent vigilance against the chess-pull. We have chosen parsimony and the drift-resistant core.

§5Where we break from the standard reconstructions

1. No Dux / no leader piece (vs. Bell). The leader-piece has no text behind it; it enters through readings of the Stanway arrangement (Parlett floated it, Bell built it). Pollux's undifferentiated "dogs," the silence of every text, and the uniform physical pieces weigh against it; Schädler reads Stanway as plausibly a native British game. All pieces equal.
2. Blockade restored as the primary victory (vs. Schädler / Murray, who centre reduction-to-one). Plato, Aristotle, and Polybius make immobilisation the image of defeat; we promote it, with material as secondary and a material tie-break. (But "primary" is a claim about attestation and theme, not frequency: the engine shows blockade rarely decides a game even under strong play, and board geometry does not fix this — §6C. We foreground it as the source-attested victory and the meaning of defeat, while acknowledging the material tie-break does the practical work.)
3. One game, not two — Latrunculi as reception (vs. the whole tradition of parallel Greek / Roman reconstructions). Stripped of accretions the two cannot be told apart; we name the Greek game canonical and the Roman game its reception (§4).
4. Slide movement (vs. Schädler's single-step), on the Laus Pisonis' "breaking the line," the 2024 computational results, and our own finding that single-step is degenerate.
5. Placement, not opposed rows (vs. Murray's chess-style array), which has "no authority" (Wayte 1892) and plays worse than placement (§6).
6. Board honesty. The big rectangular boards (Colchester 8×12) are treated as probably a different game, following the computational stall-out, rather than stretched to fit.

§6The tested program

The design choices were not argued to a conclusion in prose but played to one: a custodial-capture engine with every rule as a flag, a unit-tested capture core (14 tests), an alpha-beta agent that defends and angles for blockade, confidence intervals over self-play, and a playable client. The canonical preset is the §3 game.

What the engine found

• Single-step movement is degenerate — step games never resolve through play (they run to the cap, at every search depth and however the cap is scored), while slide resolves cleanly. The firmest result; it settles movement in favour of slide.
• Placement is fairest; the array is worst. A 9×9 array hands the first player ~80% of decided games and collapses opening variety; placement is the fairest and most varied opening.
• A material tie-break rescues the austere game — deciding a capped game by dogs-standing takes the blockade game from ~⅔ drawn to fully decisive while staying fair. Adopted.
• Freeing suppresses combat — under skilled play the incitus rule drops captures from ~18/game to ~0.7/game; a careful player won't risk a capture a counter-capture can undo. This is why freeing is a Roman-reception variant only, off by default.
• Corner-capture is near-inert — it rarely binds; settled on coherence, not consequence.

The structured-decision rationale and per-rule confidence grades are kept in a companion essay, Design Decisions & Evidential Standing.

6C · The blockade verdict — an open editorial fork

The program was rebuilt for speed (a bitboard engine, parity-verified against the reference on 2,304 games and exact on the evaluation) and extended to three independent agent architectures — a static heuristic, alpha-beta search at depths 2–3 under three evaluations (one written specifically to seek blockade), and an evaluation-free MCTS. This let us put the one question the rest of §6 left implicit: does blockade, which §3 foregrounds as the primary victory, actually decide games under competent play?

The result is stark, and it is robust. Across all three architectures — and again once the placement opening is played competently rather than blindly — blockade decides essentially no games: zero, over several hundred decided games, with upper 95% confidence bounds of only a few percent. What decides them is capture — annihilation under sharp play, the material tie-break otherwise. The finding meets the agent-invariance bar this project set for itself (trust a result only when ≥2 architectures agree); here three do, including the eval-free architecture best suited to a siege.

Why — and it is structural, not a weakness of the players. Blockade and capture pull against each other. To immobilise an enemy stone you must crowd its neighbours, but crowding it custodially tends to capture it (or spring a bracket) first; a true blockade requires keeping enemy stones alive and trapped, a delicate global plan no shallow search assembles and no rollout stumbles into. And the movement rule compounds it: slide movement gives a hemmed-in stone a long-range escape down any open file, so total immobilisation on an open board is extraordinarily hard. The irony is exact — slide was adopted because single-step is degenerate, yet slide is also what puts the canonical victory out of reach.

This forces a choice the engine cannot make. It is set down here as an open editorial fork, not a settled revision:

The responses are not exclusive — one could adopt A in the prose now and flag board size as a live experiment. What the engine establishes is only the fact that forces the choice: as the canonical rules stand, the soul of the game and the way the game is won are not the same thing.

§7Sources

The key passages are quoted in the original and in fresh translation in §1C, "The witnesses in their own words."

Greek (the canonical game). Homer, Odyssey 1.107 · Plato, Republic 487b–c; Phaedrus 274c–d · Aristotle, Politics 1253a · Polybius, Histories 1.84.7–8 (on Hamilcar Barca) · Pollux, Onomasticon IX.97–98 (keystone).

Roman (the reception). Varro, De Lingua Latina X.22 · Ovid, Ars Amatoria III.357–360; Tristia II.477 · Martial, Epigrams XIV.17 · Laus Pisonis 190–208 · Seneca (the alligatus metaphor) · Isidore, Etymologiae XVIII · Plautus, Poenulus 905 & Lucilius (ad incitas redigere).

Secondary (interpreters). Wayte, Classical Review VI (1892) · Austin, "Roman Board Games" (1934–35) · Murray, A History of Board-Games Other Than Chess (1952) · Bell, Board and Table Games · Schädler, Latrunculi (1994) & "The Doctor's Game" (2007) · Parlett, Oxford History of Board Games (1999) · Crist, Piette, Soemers, Stephenson & Browne (2024).

↑ Back to top  ·  Play Pessoi →