SpaceWheat: a farming game for building quantum intuition

Venue	Foundations of Digital Games (FDG) 2027, Melbourne, in person
CFP	http://fdg2027.org/authorinformation.html
Deadlines	Secondary, UNVERIFIED after search 2026-11-02. From a committee member's post surfaced by search (the official sub-pages sit behind a self-signed certificate and could not be read): workshop and panel proposals 2026-11-16; full papers and abstract-only submissions 2027-01-11; late-breaking work, games/demos and doctoral consortium 2027-03-08; all 23:59 AoE. The Scout's 2027-01-04 abstract date is contradicted by that post. Check the dates and limits in a browser before relying on them
Recommended track	Demo (realistic). A full paper would need learning-outcome data that doesn't exist
Status	DRAFT. Not submitted. Luke submits by hand, and only if he chooses to. Travel to Melbourne costs money, so that's Luke's call
Revision	2026-11-02 second sweep (related work, venues).
Issue	private, withheld

Deadline check (2026-11-01): still unverified; fdg2027.org presents a self-signed certificate and the FDG and SASDG domains have expired certificates. The 2027-01-04 abstract date is not contradicted. Pattern from FDG 2026 (Copenhagen): abstracts 2025-12-05, full papers 2025-12-15, late-breaking and "Games & Demos" 2026-03-30; the track the draft calls "Demo" was named "Games & Demos" in 2026.

Deadline check (2026-11-02): the site root (fdg2027.org, read once with certificate checking off; still a self-signed certificate) confirms Melbourne/Naarm, 2027-06-21→24, a programme of papers, posters, demos, games, workshops, panels and a doctoral consortium, and EasyChair fdg27; its author-information and submission-guidelines sub-pages could not be read. Deadlines come from a committee member's post surfaced by search (secondary, UNVERIFIED after search 2026-11-02): full papers and abstract-only submissions 2027-01-11 23:59 AoE; workshop and panel proposals 2026-11-16; late-breaking work, games/demos and doctoral consortium 2027-03-08. The 2027-01-04 abstract date in the 2026-11-01 check is now contradicted, and the ~03-30 pattern for games and demos is superseded by 03-08. Confirm on fdg2027.org by hand.

Abstract (draft, demo track)

SpaceWheat is a farming game in which every field is a real quantum system. Each plot is a qubit, each biome is a density matrix evolving under its own Hamiltonian, and every harvest is a Born-rule measurement. The physics is the engine, not a metaphor: a C++ extension on Eigen propagates the state exactly under U = exp(−iH·dt), and 66 of the 164 biomes add authored Lindblad channels for open-system decay. Players learn four verbs that are four pieces of quantum mechanics: strike (projective measurement), gather (surprisal, so rarer outcomes pay more), fast-forward (Hamiltonian evolution) and incorporate (Berry phase). A public companion document grades every concept the game teaches as exact, faithful or suggestive, and says how to check it, so the game never presents a gesture as a calculation. The demo shows the core loop, the raw density-matrix overlay, and building Bell and GHZ states from 14 exact gates. We also describe how the game was built: entirely by AI coding agents directed by one physicist. A rig lets an automated player use only the keys the on-screen guide names, and that player reaches the campaign ending from a fresh save in most runs.

Outline (demo paper; confirm the page limit)

1. What it is (½ p). Qubits as plots, biomes as density matrices, and the four verbs.
2. Honest physics (½ p). The exact / faithful / suggestive grading, with examples: Born-rule collapse and purity are exact, and weak measurement is tested at η = 0 and 1.
3. The demo (½ p). The loop, the matrix overlay and the gate frames. What a visitor does in five minutes.
4. How it was built (¼ p). Agent-directed development, the campaign validator, and the automated player.
5. Next (¼ p). Playtests with pre/post concept questions, which would make the learning claim measurable.

Evidence each claim rests on (public repository AQuantumArchitect/SpaceWheat, MIT)

Claim	Source
164 biomes, 66 with Lindblad channels; 102 factions, 367 icons; 14 exact gates	README.md "By the numbers"
Exact unitary propagation in C++/Eigen	native/, docs/FOR_PHYSICISTS.md (graded exact)
Exact / faithful / suggestive grades	docs/FOR_PHYSICISTS.md
Automated player reaches the ending "in most runs"	README.md status paragraph (the rate itself is not stated: DARK)
Releases v0.2.0 (2026-06-17), v0.1.7-alpha (2026-10-19)	GitHub releases
Built by AI agents, no hand-written code	README.md, docs/HOW_THIS_WAS_BUILT.md

Related work and the negative space

Every entry below was confirmed by a Semantic Scholar record, an arXiv abstract page, or the reference list of one of the two field hubs, as recorded in the research notes dated 2026-11-01. Flags are carried forward.

• Seskir et al. (2022), Optical Engineering 61(8), 081809. The 24-author review of quantum games and interactive tools; each tool's section is written by its own developers.
• Meyer, Goorney, Kushimo & Seskir (2026), American Journal of Physics 94, 593. Resource Letter QIE-1 on quantum information education; its games section lists three resources and finds "general claims about value, with no specific learning-effect evidence".
• Piispanen, Pfaffhauser, Wootton, Togelius & Kultima (2025), EPJ Quantum Technology 12, 7. "Defining quantum games": a survey of collected games and a proposed definition. Read in full on 2026-11-02 (arXiv:2206.00089v5, the published text; verified at Crossref): more than 350 games; zero hits for density matrix, mixed state, open system, Lindblad, geometric or Berry phase, Zeno, topology/SSH/lattice gauge, farming or QuTiP; "decoherence" once, for Quantum Battleship on real hardware.
• Sørensen et al. (2016), Nature 532, 210. Quantum Moves; retracted at the authors' request on 22 July 2020 after an unaffiliated researcher found an error in the code.
• Sels (2018), Physical Review A 97, 040302(R). A simple stochastic optimiser from a random seed outperforms all Quantum Moves players. Grønlund (2019, arXiv:1904.01008) reproduced this and added GRAPE.
• Jensen et al. (2021), Physical Review Research 3, 013057. The replacement paper: Quantum Moves 2, where player seeds help a cluster optimiser; "this is not to say that players are better than algorithms".
• Ahmed, Weidner, Jensen, Sherson & Lewandowski (2021), American Journal of Physics 89, 307. Quantum Composer: a research-grade one-dimensional Schrödinger engine inside an education tool.
• Migdał et al. (2022), Optical Engineering 61(8), 081808. Virtual Lab by Quantum Flytrap: exact pure-state photon optics, up to three entangled photons.
• Evenbly (2024), arXiv:2406.11995 (physics.ed-ph); Steam, free, 29 July 2025. The Qubit Factory: a real-amplitude state-vector simulator, at most six entangled qubits, "decohere" realized as a projective measurement, reduced density matrices only for an entanglement-entropy display; no learner data. Added 2026-11-02 as the nearest neighbour.
• Wootton (2020), FDG '20. Procedural generation using quantum computation; the venue's own precedent for quantum content.
• Passante, Emigh & Shaffer (2015), Physical Review ST Physics Education Research 11, 020135. Student ability to distinguish superposition from mixed states; the one education paper squarely on mixed states.
• Manzano (2020), AIP Advances 10, 025106. The de-facto pedagogical introduction to the Lindblad master equation.
• Johansson, Nation & Nori (2013), Computer Physics Communications 184, 1234. QuTiP 2: the reference implementation any "exact GKSL" claim will be compared against.
• Aharonov & Anandan (1987), Physical Review Letters 58, 1593. Phase change during a cyclic quantum evolution; Berry (1984), Proc. R. Soc. A 392, 45, is the adiabatic original.
• Feinstein & Cannon (2002), Simulation & Gaming 33(4). Fidelity, verification and validation as separate constructs of simulation evaluation.
• Shute, Ventura & Kim (2013), Journal of Educational Research 106(6). Stealth assessment in a physics-engine game, validated against a pre/post test with 167 students.
• Hamed & Rebello (2026), PERC 2026, arXiv:2606.03615. Game versus simulation in a counterbalanced crossover with 83 pre-service teachers; the simulation-first order produced the larger gain.
• Parrondo, Horowitz & Sagawa (2015), Nature Physics 11, 131. The modern synthesis of the thermodynamics of information. Author order confirmed; DOI 10.1038/nphys3230 verified at Crossref on 2026-11-02.
• Ribezzi-Crivellari & Ritort (2019), Nature Physics 15, 660. A continuous Maxwell demon that looks at fixed intervals, acts only on a state change, and is "fuelled by rare events".
• Jacobs (2012), Physical Review E 86, 040106(R). The energy cost of a measurement is the work value of the information it acquires; looking is never free at finite temperature.
• Elouard, Herrera-Martí, Huard & Auffèves (2017), Physical Review Letters 118, 260603. Work drawn from the measurement channel itself; the erasure cost "becomes negligible in the frequent-measurement (Zeno) regime".

In the 48-page Seskir review the strings "density matrix", "mixed state", "decoherence", "Lindblad" and "open quantum" occur zero times, and the 2026 Resource Letter says open systems, decoherence and density matrices are "not discussed as teaching topics". The nearest engines each stop short: Quantum Composer is exact one-dimensional Schrödinger; Virtual Lab simulates pure photon states; the Virtual Quantum Optics Laboratory (La Cour et al. 2022, IEEE QCE) runs a classical Gaussian emulation; Quantum Chess (Cantwell 2019, arXiv:1906.05836) is unitary with a measurement rule and no noise; QuTiP is exact GKSL but a library, not a game. Psi and Delta (Anupam et al., cited through Seskir's reference list) uses measurement counters to express probability, the closest existing mechanic to "harvest is a Born sample". Decodoku (Wootton 2017, arXiv:1712.09649) is the only topology game found. On the economy, no game prices a look or pays surprisal; the nearest objects are Ribezzi-Crivellari and Ritort's demon, which is an apparatus, and the reinforcement-learning curiosity line (Achiam & Sastry 2017, arXiv:1703.01732) where −log p is a reward with no temperature attached. On fidelity, Seskir §3.4 says "there is no common design methodology or an universal evaluation tool for quantum games" and proposes only a rating of whether a game has been evaluated; Feinstein and Cannon's three constructs are the nearest rubric. The check that stood here, Piispanen et al.'s surveyed game list, was read on 2026-11-02 (second sweep, below); "none found" stays the ceiling because two searches were refused by the session budget.

What this changes in the draft. Seven points. First, "the physics is the engine, not a metaphor" stands, and the negative space supports it; but write "we found no published game that evolves a density matrix under open-system dynamics (Seskir et al. 2022; Meyer et al. 2026; Piispanen et al. 2025)", never "first": the survey list is now read, but two searches (a 2024–2026 quantum-game taxonomy / evaluation-framework follow-up; any SSH / topological / lattice-gauge game) were refused by the session budget and are NOT RUN. Second, "players learn four verbs" is a learning claim, and the Resource Letter's complaint about the field is exactly claims "without outcome data"; write "the four verbs are four pieces of quantum mechanics" and keep outcomes DARK, with Hamed & Rebello (crossover design) and Shute (telemetry validated against a pre/post test) cited in §5 as the designs that would end the darkness. Third, the Quantum Moves lineage (Nature 2016, Sels 2018, retraction 2020, Jensen 2021) belongs in §2 as the reason the exact/faithful/suggestive ledger and re-runnable assays exist; cite the three papers together, not the Nature paper alone. The same lesson governs "reaches the campaign ending in most runs": it stays DARK until a rate is counted. Fourth, the "exact" grade will be read against QuTiP; state whether the engine was cross-checked against it, or mark that DARK. Fifth, the abstract says "Berry phase" while the research notes' scope line describes ripeness as the Aharonov–Anandan phase; name the one the code computes and cite Berry (1984) or Aharonov & Anandan (1987) accordingly, with FOR_PHYSICISTS.md as the witness. Sixth, "gather (surprisal, so rarer outcomes pay more)" has no published game neighbour; the demo paper may say so and cite Jacobs for "looking is never free", but the companion document's temperature read from live entropy needs an explicit defence (Jaynes 1957, Physical Review 106, 620, is the licence), since the physics literature keeps kT as the bath's. Seventh, cite Wootton (2020) in §1 so reviewers see quantum content has appeared at this venue before.

Second sweep (2026-11-02). Piispanen et al. (2025) was read in full (arXiv:2206.00089v5, the text of EPJ Quantum Technology 12, 7; verified at Crossref and arXiv). It catalogues more than 350 games (over 250 examined between 2019 and 2022), and across the whole text "density matrix", "mixed state", "open quantum system", "Lindblad", "geometric phase", "Berry phase", "Zeno", "topology", "SSH", "lattice gauge", "farming" and "QuTiP" occur zero times; "decoherence" occurs once, for Quantum Battleship (Wootton 2017–18), where it is noise on real hardware and not a simulated channel; its eight "simulation" entries are all wavefunction- or circuit-level. The sentence the demo paper may write is "we found no published game that evolves a density matrix under open-system dynamics (Seskir et al. 2022; Meyer et al. 2026; Piispanen et al. 2025)", still not "first", because two searches were refused by the session's budget and are NOT RUN (a 2024–2026 quantum-game taxonomy / evaluation-framework follow-up; any SSH / topological / lattice-gauge game). The nearest neighbour is The Qubit Factory (Evenbly 2024, arXiv:2406.11995; Steam, free, 29 July 2025): a real-amplitude state-vector simulator with at most six entangled qubits, in which "decohere" is a forced projective measurement and reduced density matrices are computed only for an entanglement-entropy display; its paper sets noise and decoherence aside as engineering challenges and reports no learner data. FDG's own dates moved to a committee member's post (see the 2026-11-02 Deadline check above; secondary, UNVERIFIED). Learning outcomes stay DARK.

DARK (do not claim a number)

• Learning outcomes (pre/post), playtest counts, retention.
• The automated player's completion rate ("most runs" is the only witnessed phrase).
• Downloads and players.

Luke's steps, if he wants to submit

1. Open the FDG pages in a browser (authorinformation.html, submissionguidelines.html), confirm the 2027-01-11 full-paper and 2027-03-08 games/demos deadlines and the page limits, and fix the dates above; they rest on a secondary post.
2. Decide whether travel to Melbourne is worth it. That's money, and only Luke can decide it.
3. A lean web build (planned for GitHub Pages) would let reviewers play it, which is the strongest evidence a demo can have.
4. Have a seat expand this into the template, then submit.