The Ahl BASIC corpus: synthesis and KN-86 cart mapping

Related canon: cipher-voice.md, ADR-0015 (CIPHER-LINE + the Null main-grid escape), ADR-0047 (experience tier), roguelike-design.md (the First Law), economy-model.md, verb-taxonomy-map.md (the 17-cart roster), atari-arcade-skeletons.md.

---

0. What this delivers

Batch 10 mined 21 vintage titles for gameplay and closed with a deferral: “The cross-cutting ‘which mechanic goes in which cart’ pass is a follow-on; per-entry applicability is kept deliberately shallow” (inspiration/index.md). This document is that pass, run over a new corpus and over the full 17-cart roster rather than the launch four.

Every mapping below was checked against the current cart spec before being written down. That check killed or narrowed roughly half of the initial proposals, and those rejections are recorded in §6 because they are as useful as the accepted ones. Where a spec already implements something, this document says so instead of re-proposing it.

A caveat on that method, and how to read the rejections

The module specs are evidence, not authority. Most of cartridges/modules/ was drafted by an agent from a one-line premise, and the result frequently runs far ahead of any decision that was actually made. Several specs are over-specified: they elaborate a system in confident vocabulary without ever establishing why the thing would be fun to play, and a few have drifted off the cart’s original premise entirely. A spec asserting its own identity (“this cart is not a puzzle, it is a tempo sport”) is a prior draft talking, not a ruling.

So: where a spec contradicts the cart’s premise, the premise wins. Rejections in §6 that rest only on a spec’s self-description are marked soft and should be reopened rather than cited.

Shellfire is the worked example of the failure. Its spec describes a frequency-band signal-extraction game with discrete menu inputs, and on that basis the first draft of this document rejected the entire artillery family. Shellfire is meant to be a Scorched Earth clone. Gunner is its direct ancestor. That mapping is now §3, and the spec is the thing that needs revisiting.

That caveat has since been confirmed and quantified. A full audit of all 17 specs against the earliest recoverable premise landed as _meta/cartridge-premise-audit-2026-07.md. Its findings that bear directly on how to read this document:

• Shellfire’s artillery premise is attested at the project’s earliest commit (λ records a firing solution (angle + power)), so the recovered premise is documented rather than inferred.
• Five carts drifted severely (Shellfire, ICE Breaker, NeonGrid, The Vault, Relay) and one is unresolved (Takezo).
• ICE Breaker was installed as the template every other spec had to match, including a line-count target, and it is itself the most-drifted spec. That is why so many carts share a four-system shape they did not need.
• The review rubric never asks whether a cart is fun, and agents graded their own inventions. Every T1/T2/T3 approved header traces to that loop, so “approved” does not mean a person accepted the design.
• Three carts never had a game premise at all (The Vault, Relay, Null were System-Module table rows), and ADR-0042 programs have already absorbed their functions.

Read every “the spec already does this, so the corpus adds nothing” conclusion below against that. Where such a claim rests on an elaborate spec rather than on a decision, it means an agent already invented something adjacent, which is a much weaker reason to stand down.

The corpus matters for the same reason the Atari books did: the constraints match. These are keyboard-only, single-screen, monochrome-teletype games with no graphics, built in very little memory, where every mechanic had to earn its place in text alone.

1. The three books in one line

#	Book (year)	What it gives KN-86
1	BASIC Computer Games (1978)	The hidden-target search family across four feedback fidelities; a budget derived from the search space (Depth Charge’s log2); the packed-integer world (Super Star Trek); the economy ancestors (Hammurabi, Stock Market); the generative trio (Amazing, Life, Poetry); a finished optimal AI (Nim)
2	More BASIC Computer Games (1979)	A fifth and higher feedback fidelity (Blackbox’s symbolic three-outcome ray); the moving-target hunt with a defender (Convoy); a living map whose hazards kill both sides (Seabattle); instrument quality as a resource (depth-gated fog)
3	Big Computer Games (1984)	The two adventure-design essays (an IF data model plus a fairness specification); one dial wired to several consequences and deliberately fuzzy instruments (Trucker); spatial arbitrage (Star Merchant); a deterministic resource race (Survival)

2. The corpus’s cross-cutting findings

2a. Feedback fidelity is the difficulty dial, and Blackbox adds a fifth rung

The 1978 capture already names four fidelities in the hidden-target family: binary (Battle), gradient (Depth Charge), continuous-analog (Gunner), range-scalar (Orbit). Blackbox adds a fifth and higher one: a symbolic three-outcome return (absorbed / reflected / deflected-to-port) computed from local three-cell adjacency, shaped by several hidden objects at once. That rewards logical inference rather than gradient-following, which is a different skill and a different cart.

2b. A budget derived from the search space, and the honesty condition attached to it

Depth Charge grants exactly floor(log2 N) + 1 shots, so optimal play is bisection. The condition that makes it fair is easy to miss: it works because feedback is a full compound gradient and nothing more. A log2 budget over a richer readout is generous rather than disciplined. Any adoption has to price the feedback and the budget together (see §5, Depthcharge).

2c. The essays are the third independent source on an open question, and the first with a build order

zork.md already asks whether KN-86 should ship a narrative-cart authoring template (verb-noun dispatch over KEC Lisp, sanctioned meta-commands, a room-graph DSL), and paw-adventure-writer.md extracts the same primitive set from PAW while explicitly stopping at the concept level. Both leave it a scope question.

What the Ahl essays add is sizing and a build order: roughly 40 room nodes, six directed exits per node with a zero sentinel, objects as (object, home-room, value) triples where home-room 0 means “nowhere yet,” one routine per verb, and a dependency chain as the puzzle unit. That turns a standing scope question into something with a shape you can cost.

2d. Fairness is currently scoped to one track

The First Law (roguelike-design.md) is documented and cited, and every citation is inside the roguelike track. Threshold, Black Ledger, and The Vault contain zero references to it. Threshold is the one cart of the three with combat and death and carries no fairness contract of any kind. The play-side essay is a ready checklist for closing that (§5).

2e. The corpus keeps landing on generation primitives the tree already wants and has not built

Three separate cases: Amazing’s carve against NeonGrid’s unbuilt generator, Life against Takezo’s already-specified-and-unbuilt loading-screen automaton, and Poetry’s group cycle against Null’s unspecified passage generator. In each case the spec names the need and stops.

3. The roster review

Superseded on the mapping question (2026-07-27). This section was written before Josh stated his premises and before he made the assignments himself. The authoritative cart mapping is now cart-inspiration-map.md, generated from his calls in the kn86-inspo workbench. What still stands below is the per-game mechanics research; what is superseded is every proposed cart assignment.

All 17 cartridges plus the concept carts. Status is the cart’s own ladder state; verdict is what this corpus contributes after checking the spec.

Carts with a confirmed, specific contribution

Cart	Status	What the corpus contributes
Null	T3, Impl none	Poetry’s group cycle as the missing passage-level generator. Null’s headline CIPHER ANALYSIS contract class (null.md:159, :331) asks for essay-length passages the operator studies, but CIPHER is a one-line generator capped at 32 characters and no passage-level generator exists anywhere in the tree. Today the answer is hand-authoring line by line (null.md:1582). Poetry’s ordered group cycle plus a forced-break counter is exactly the layer above the line, and it composes on top of CIPHER’s existing grammar without touching it.
Cipher Garden	T3, Impl none	Blackbox, as a chosen-plaintext oracle puzzle. Every Cipher Garden puzzle today is static data handed over at load (:1378-1523); the operator never generates new evidence. A costed probe would be the cart’s first interactive information source, in native cryptanalysis vocabulary one step from the existing crib mode (:300-327). Blackbox’s scoring also supplies the par the cart explicitly lacks (:107, “payout is not penalized by failed guesses”). Separately, Battle’s simultaneity (decoding the scheme while probing is still producing evidence) is what the existing phase-sequential Counter-Intelligence campaign (:669-741) does not have.
Pathfinder	T3, Impl none	Three Trucker mechanics, and Star Merchant’s price rule. Fuzzy instruments is the strongest single item in this corpus: every Pathfinder readout is an exact integer, so its own planning rule (“allocate 1.5x calculated fuel as safety margin,” :213) is arithmetic. A coarse gauge makes it the judgment call the rule is pretending to be. Add fatigue as a second clock (Pathfinder tracks one, :663) and a nonlinear fuel cliff. Separately, Star Merchant’s produce/lack trade-class rule gives Pathfinder the thing it lacks: a reason cargo has different value by destination, turning route optimization into arbitrage without touching the fuel or threat model.
NeonGrid	T1-T3 approved, Impl partial	Amazing’s iterative carve, plus a forced decision on braiding. A generator is already named (recursive backtracking, :125-135) in the same spanning-tree family, so Amazing does not add an algorithm. It adds the iterative form, which this runtime requires: the spec says “recurse” over grids up to 300 cells, and the Fe GC root stack is shallow enough that Nodespace’s spec already mandates a worklist for exactly this reason (nodespace.md:396-398). It also forces a decision the spec demands three ways and never states: the braiding rate. Sentry patrol loops (:166) are closed circuits that cannot exist in a spanning tree.
SynthFence	T3, Impl none	Stock Market’s price engine. The current generator is a memoryless ±1 basis-point walk with a volatility multiplier (:225-230): no trend, no regime switching, no correlated news shock. A trend field exists in the cell struct and renders on screen (:955, :1683), but nothing generates or flips it, so the display promises a structure the generator does not produce. The spec concedes the gap (:2135). Stock Market supplies exogenous structure and composes with the already-queued Star Trader mechanic, which supplies endogenous price impact.
Takezo	T3, Impl none	Gomoku’s threat-space evaluator, narrowly. Takezo is a unit-movement tactical wargame, so Gomoku is not its game. But the PATTERN RECOGNITION contract class (:1098-1103) and the pattern-hint tool (:163-166) both promise n-in-a-row detection with no engine behind them. A threat-space evaluator (score a cell by longest own run extended plus longest enemy run blocked) fills a named hole. Also the home for Nim’s detuned-optimal-opponent model (§4).
Depthcharge	T1-T3 approved, Impl partial	Seabattle’s own-depth fog and its living map. Depth-driven visibility is asserted (:87) with no mechanism anywhere; Seabattle supplies one, and adds stochastic hiding plus gating keyed to your own depth rather than to contact distance. The larger contribution is the enemy movement pass that resolves NPC-versus-terrain collisions independently of the player (a ship stepping on a mine dies), which Depthcharge never does. Convoy’s periscope-limited sensing belongs here too, with a caution (§5).
Drift	T3, Impl none	Geowar’s scare-and-relocate, only. Drift already specifies the triangulation math (atan2 at :210, line-intersection with geometry-quality confidence at :1331-1352), so the bearing precompute duplicates it. What Drift genuinely lacks is source agency: its single source has no reaction to being localized, and even the moving-target template follows a predefined cyclic path (:956-960). A near miss that alerts the source and relocates it fills that gap. Sequence it with the confidence-band fold already queued at :1572 or the two will contradict.
Shellfire	T3, Impl none	The whole artillery family, because Shellfire is a Scorched Earth clone. Gunner is the direct ancestor and supplies the honest ballistic model: range = R·sin(2θ), symmetric about 45°, with sensitivity flattening near 45° so the top of the arc reads naturally as a fine-aim zone and two elevations reach most targets. Geowar adds multiple targets in one field, precomputed hit-versus-near-miss windows, and one shot resolving against several collinear targets. Lunar’s Euler integration is the shape for a shell arc under wind. The loop is already in the tree as the LOB skeleton (atari-arcade-skeletons.md §1.5, aim → fire → observe → adjust), and it consumes the shipped render/half-block-line, so the arc is drawable today. Geowar’s scare-and-relocate also reads well here as a target that repositions after a near miss. The current spec describes a different game and needs revisiting (§0).
Sysop Mode	T3, Impl none	Battle’s packer, narrowed to the contract generator. It adds nothing to the sysop’s own placements, which are single-cell and drawn from separate finite budgets (:127-211). It fits the honeypot and real-asset seeding in the contract templates (:379-428), so decoys are not trivially clustered. Bombardment’s residue is its negative result: memoryless-random targeting is a documented AI floor the canned-attacker tapes (:322-336) must beat.
Threshold	T1-T3 approved, Impl none	The essays’ world and object model, and a fairness contract. Threshold is the NARRATIVE cart, and the essays expose three real gaps: an unreconciled dual navigation model, an inverted object model with no reveal verb, and no fairness contract at all despite having combat and death. Detail in §5. The essays’ parser material has no home here (§6).

Carts the corpus confirms rather than changes

Cart	Finding
Nodespace	Presidential Campaign’s electoral structure is already implemented here. “Presence totals never decide a contract; they decide who is able to hold anchors” (:56-57) is winner-take-all summed to a tally, stated as a design principle, and “the thin victory” (:226-228) is explicitly the popular-vote-loses outcome. Read the corpus as confirmation the design is sound. Nodespace is also a candidate for Nim’s detuned-optimal model (§4).
Black Ledger	Already has a stronger lock-and-key dependency graph than the essays describe (:239-247 with :requires / :hold / :reveal-at, plus DOSSIER’s typed fact-links at :176-185). The essays add nothing here, and presenting them as an import would be re-proposing shipped design. One narrow rhyme holds: Tower of Hanoi’s move-count-against-par maps onto the existing move budget plus move-efficiency scoring (:96-100, :263).
The Vault	Already implements the play-side essay’s anti-stuck posture: a hint surfaced after five minutes of unsuccessful search, one per bounty (:82), plus an empty-cross-reference guard (:96). The best fairness behavior in the roster. Its reveal is a passive stat threshold, so the essays’ “revealed by state mutation” primitive lands on DOSSIER instead.
ICE Breaker	Blackbox was rejected here on the spec’s self-description (“not a puzzle… a tempo sport,” :18, :32). That is a soft rejection under §0. A probe-deduction layer inside a netrunner cart is a reasonable thing to want; reopen it against the premise rather than the spec.
Relay, Marty Glitch	Marty’s CIPHER hijack is already fully specified against the existing grammar engine, so Poetry is confirmatory prior art rather than a mechanism it needs. Relay has no contribution from this corpus.

Concept carts and unhomed material

An influence / propaganda cart is the one genuinely new concept this corpus argues for. The shape: the operator composes stories and watches them propagate through a population, so the authoring act is the gameplay and the payoff is a bloc shifting. Four threads converge on it, and none of them has a home today:

• Presidential Campaign’s weighted blocs supply the target model: spend limited reach across constituencies with different weights and different susceptibilities, before a deadline.
• Poetry’s slot grammar is the story generator. The same engine that gives Null its passages produces fabricated copy from a swapped vocabulary bank, and the operator’s job is shaping and selecting rather than typing prose.
• Marty Glitch supplies the broadcast channel and the fiction. Its whole premise is hijacking a signal to put your voice where someone else’s belongs.
• FABRICATION / MANUFACTURE has 0 launch satisfiers (mission-type-catalog.md:58), reserved for the unspecced CLONESHOP. Fabricating a story is manufacture applied to information, which fills that class without inventing a verb.

Open, and deliberately left open: IF play is Threshold, settled. Where IF authoring lives is not. This cart is the authoring side pointed at influence rather than at storytelling, which is a different thing again from a general narrative-authoring template (§2c). It is not Black Ledger, whose forensic-accounting premise it does not touch. Parked as a concept in future-concepts.md.

Presidential Campaign’s other pieces split cleanly: the electoral structure confirms Nodespace, the cash-versus-scandal event matches the Entrepreneur Jungle overlay already queued into SynthFence, and the fund-raise-versus-campaign turn choice is a campaign-economy faucet/sink pattern alongside Hammurabi and Stock Market.

4. The strongest threads

Ranked by contribution against effort. Read them against the §0 caveat: a thread that only makes sense because a spec says so is weaker than one that serves the cart’s premise.

1. Gunner → Shellfire, re-anchoring the cart on its premise. Shellfire is a Scorched Earth clone, so it wants honest artillery: range = R·sin(2θ), an arc drawn with the shipped render/half-block-line, wind, and terrain that deforms where shells land. Geowar supplies multiple targets and near-miss windows. The loop is already catalogued as the LOB skeleton. This is the highest-value item here because it costs a spec revision and returns a game people would actually want to play.
2. Poetry’s group cycle → Null’s passage generator. Fills a named, empty slot in the cart’s headline mechanic. Composes on top of the existing CIPHER grammar without modifying it.
3. Blackbox → Cipher Garden as a chosen-plaintext oracle. Gives the cart its first interactive information source and the scoring par it lacks, in native vocabulary.
4. Trucker’s fuzzy instruments → Pathfinder. One change that converts an arithmetic exercise into the judgment call the spec already claims it is.
5. Amazing’s iterative carve → NeonGrid. The generator is unbuilt, the recursive form will not survive the Fe root stack, and the braiding rate is an open decision the gameplay already forces.
6. Stock Market’s engine → SynthFence. Replaces a memoryless walk with the market structure the cart’s own display already promises.
7. Nim’s detuned-optimal opponent → Takezo §4 and Nodespace §5. The transferable asset is not the game. It is the difficulty model: a provably correct solver whose knob is how often it declines the winning move. Takezo currently claims near-optimal play (:222) from an evaluator that cannot produce it, with search depth as its only knob.
8. Life → Takezo’s loading screen. orchestration.md:672 already specifies a glider-gun automaton that resolves into the Takezo mark. It is specified and unbuilt, and Life’s in-place marker update and living-population bounding box are directly usable.
9. Geowar’s ray-fire → atari-arcade-skeletons.md as a sixth skeleton beside §1.5 LOB (aim → fire → observe → adjust), which already consumes the shipped render/half-block-line. ICBM belongs in the same place or as an experience-tier set-piece per ADR-0047.

5. Conflicts and gaps this pass surfaced

These are findings about existing specs, uncovered while checking the mappings. Several need a decision rather than an edit.

Decisions

• Depthcharge’s ping budget versus its ping fidelity. A log2-derived budget is only honest against gradient feedback. Depthcharge’s active ping already returns range to all contacts omnidirectionally (:133-138), which is strictly more. Adopting the budget means narrowing the ping, which contradicts the current spec. The trade is the actual decision.
• Marty Glitch’s mode bias fails its own merge gate. It declares :silent −0.10 in :active-hack (:1459); the baseline is 0.15 and the style guide’s floor is 0.10, so it lands at 0.05 and check 9 hard-fails. The spec flags the intent (“Marty refuses to shut up,” :1462), so persona and gate are in direct conflict. Resolve by explicit override or a revised bias.
• Threshold’s dual navigation model. The spec declares both a tree (CAR descends a level, CDR moves to a sibling, :164-195) and compass exits on the room cell (:901-906, :732-733), and reconciles neither. The essays’ flat node-with-exits model is one way to collapse it.
• NeonGrid’s braiding rate. The spec wants braided three separate ways and never sets a number.
• Convoy in Depthcharge. Giving the operator a torpedo contradicts the cart’s stated philosophy (:16, “a meditation on tactical patience”). Route it as an inversion of the existing EVASION contract or as an experience-tier set-piece.

Gaps and inconsistencies

• The First Law is scoped to roguelike carts only. Threshold has combat and death (:365) and no death policy, no permadeath statement, no respawn rule.
• No cross-cart clue-markup convention. Threshold, The Vault, and Black Ledger each invented their own glyph vocabulary; screen-design-rules.md is where a shared one would live and has none. The play-side essay names precisely why this matters: players read typographic emphasis as deliberate signal.
• Null’s example passages violate Null’s own merge gate. The illustrative passages are second-person prose (:342, :662-666), which the style guide hard-fails (check 7). They predate the CIPHER-LINE model and were never re-authored.
• OPERATOR_PROFILE / “Template Class” is a dangling term (null.md:355, the only occurrence in the tree). Define a passage-class registry or delete it.
• Marty Glitch §3 is stale. A canned-line utterance model (:196-464) sits unreconciled beside the grammar-contribution model (:1399-1484) that supersedes it.
• Takezo Tier 2 and Tier 3 specify different games. Tier 2 is a unit wargame; Tier 3’s CIPHER vocabulary and nEmacs grammar are pure Go (stone, group, ko, liberty, atari, sente; :1375-1384, :1458-1487). Not among the five conflicts its migration notes already list.
• strategy-passive-modules.md:11-63 carries the pre-rewrite Nodespace, superseded by nodespace.md v2.0 (2026-07-23) and never updated. Anyone reading Nodespace from there designs against a retired model.
• Black Ledger’s red-herring design was lost in the v2.1 rewrite. It survives in docs/reviews/blackledger-round-1-evaluation.md and the campaign doc, and is absent from the current spec.
• “NEGOTIATE has 0 launch satisfiers” is stale (mission-type-catalog.md:120). Threshold declares NEGOTIATE and SOCIAL (:558-564). The claim is true of the launch four and false of the 17-cart library.
• NeonGrid states Wave 1’s grid size two ways (16×10 at :113, 6×8 at :111 and the onboarding screens).
• Batch numbering has diverged. synthesis.md numbers corpus-level batches (Batch 9 = Atari, Batch 10 = Bartle) while inspiration/index.md numbers folder batches (Batch 9 = Entrepreneur Jungle, Batch 10 = gameplay mining). Both next-numbers are 11, so this batch is unambiguous, but the schemes disagree from 9 onward.

6. What does not transfer

Recorded because rejected mappings save the next pass from re-proposing them. Items marked soft rest on a spec’s self-description and should be reopened against the cart’s premise (§0) rather than cited as settled.

• (SOFT) Blackbox in ICE Breaker. Rejected because the cart’s own text refuses “puzzle” twice (:18, :32) and claims an OODA cycle down to 0.5-2 seconds. Cipher Garden remains the better home on loop shape and scoring, but nothing about a netrunner cart forbids a probe-deduction layer, and the spec’s self-assertion is not evidence.
• ICBM in Shellfire — this rejection was wrong and is withdrawn. It was made by reading a spec that describes a frequency-band game. Shellfire is a Scorched Earth clone, so the artillery family is its lineage, not a foreign import. See §3.
• Life is not Nodespace’s spread rule. Nodespace is a real-valued presence count transported up a beacon-generated gradient with proportional attrition and tier-scaled growth over a six-way neighborhood. Life is a binary-state neighbor-count threshold over Moore-8. Life for Two’s integer-packing ownership trick is also numerically unusable here: Nodespace’s own board screen shows presence totals of 4,180 and 6,320, and a 1000×p2 + p1 packing needs p1 < 1000 and leaves Fe’s exact-integer float range shortly after.
• Nim is not a cart. No cart in the roster is modeled as a two-player perfect-information game with no board, and nothing in Takezo survives removing the grid. Fold the difficulty model; park the game concept in future-concepts.md if wanted. ADR-0021 separately rejected reimplementing classic titles as Fe carts.
• Star Merchant is not SynthFence. SynthFence contains no travel, fuel, distance, or geography; the operator trades multiple venues from one terminal at zero travel cost, and its identity is the market’s tempo rather than the operator’s movement. Star Merchant’s constraints are the opposite arrangement. Host it on Pathfinder.
• (SOFT) The adventure essays’ parser in Threshold. Rejected because the spec calls Threshold a choice-based MUD (:26): a verb key plus a numpad menu index, with no vocabulary lookup. “Choice-based” may itself be a drafting decision rather than a settled one, and a MUD with a parser is an ordinary thing to want. Threshold is confirmed as the home for IF play; where the authoring side lives is open (§3, concept carts).
• Tower of Hanoi does not model Black Ledger’s dive/pop. Black Ledger’s hierarchy is a read-only navigation tree where nothing relocates and every traversal along an existing edge is legal. Hanoi is about relocation under a legality rule. Only the move-count-against-par rhyme holds.
• Bombardment’s simultaneous-fire duel is already specced in Sysop Mode (:852) from a better-documented source (Nuclear Destruction), with an additional argument about link-cable timing.
• Combat (1978) is not a targeting cart. It has no spatial component; it is threshold-commitment resource allocation. Mine it for how missions consume credits.
• Grand Prix is not a program-your-driver game. The 1979 listing is manual per-interval throttle and brake against a kinematic model. A KN-86 version where the operator authors a driving policy in the REPL would lean into the deck’s programmable identity, and it would be new design rather than a port.
• Period skins throughout. Star Trek IP, 1980s US-election flavor, literal trucking freight, National Geographic moon rooms, Seabattle’s sea-monster whimsy. Keep the systems, rewrite the fiction.

7. How agents should use this guide

This is the reference layer. The work is per-cart folds and, for the larger items, a spec pass.

• Landing a mechanic in a cart? The assignments are already made: each module spec carries an Inspirations section, and cart-inspiration-map.md is the full map. Do not reopen approved tiers to add a mined mechanic.
• Check §6 before proposing a mapping. Half the obvious ones are already rejected there with reasons.
• Check whether the spec already does it. This pass found several cases where a proposal would have re-proposed shipped design (Black Ledger’s dependency graph, Nodespace’s anchor-count victory, Drift’s triangulation, Pathfinder’s speed dial, Sysop Mode’s simultaneous orders).
• Designing a generative surface? Poetry’s recipe (ordered slots, random fill within a slot, probabilistic punctuation, a context flag, a forced break) is the corpus’s most reusable idea, and it composes with the existing CIPHER grammar rather than replacing it.
• Designing an exploration or narrative surface? Read the two essays in big-computer-games.md first, then §5’s fairness gaps. The play-side essay is a fairness checklist.
• Always: honest budgets priced against the feedback they are paired with; monochrome → dither; keyboard-first replaces the typed-number clunk without touching the math.