Educational Game Design — Outstanding Work

Source of truth is the Markdown: .claude/plans/educational-game-design.md — CodeRaven parses its checkboxes, so tick them there, not here. This rendering is for the review pass and will drift once items get checked off.

Where things stand

35Mechanics5 drafted · 30 seeds

57Themes8 drafted · 49 seeds

0Tested against the appno game category, no Swift

Decided: a game is one category parallel to manual and itinerary — one guide per game, game types as an authored games[] array, entries as content items whose fields arrive per-guide via applies_when. A game type is one mechanic × one theme × one subject. Theme sits at guide level, overridden per game type only where it diverges.

Not built: the game category, any Swift, and the machine export target.

Sequencing

The spine

A (game category + Drive Corps)  →  E (Swift)  →  F (Peregrine)
                                     ↑
B, C, D run alongside, any order  ───┘   (D feeds F, loosely)

The critical path is A. Everything in B, C, and D is genuinely useful and none of it unblocks anything — which is exactly why it will quietly absorb all available time if A doesn't go first.

The avenues

A · Prove the model

critical path10 items

The game category plus Drive Corps. Two game types is enough to exercise applies_when; eleven is the wrong first test.

Create categories/game/ — schema, and a brief carrying the gamification test plus Drive Corps's forbidden vocabulary
Create guides/drive-corps/ with theme: "military" and a games[] array for recon-sweep and intercept
Write per-game-type schema-extensions blocks from each mechanic's content contract, not invented
Migrate 54 signs (preserving concept as the scoring key) and 100 questions (adding the concept tagging the source bank never had)
File Intercept's 12 transmissions as class: flavor; decide where payloads live
Write the corpus → app export script
Verify the document renders and doesn't choke at ~154 entries (previous max: 144)
Verify applies_when shows only the right fields per game type
Round-trip one bank into Drive Corps unmodified

B · Mechanics — open questions

10 items

Cheap now, expensive later — every new entry gets shelved against unresolved axes.

Layers vs. containers — draft Boss Gauntlet and The Container; a layer will fail to say what one unit of content is
Discovering as a sixth motivator — grows to three entries or Rule Discovery returns to Building
Four merge questions carrying live doubt: Stream Go/No-Go (65), Minimum Evidence (70), Match the Tool (75), Call Then Execute (82)
Find the Pivot (62): mechanic or content lens?
Compress It (58): needs a runtime judge — that's the ceiling of authored-content-only
Chase The Half-Map, surfaced by the theme catalog and covered by nothing in the 35

C · Mechanics — drafting

4 items

Name the Trap (92) — cheapest high-value entry; Peregrine already ships the con taxonomy and Trap Atlas it needs
Inspection Budget (90), Build the Chain (90), Fault Sweep (88), Assemble to Order (85)
The entries the drafted themes lean on: Order From Constraints, Place the Mark, Infer the Rule
Close the response-mode gap — nothing built asks the player to locate or assemble. Fault Sweep is the cheapest test

D · Theme catalog

6 items

Draft mechanic fit for the seven fiction-written seeds
Normalize the 15 orphan role strings across both catalogs
Run the subtraction test on the overlap clusters — Detective / Crime Syndicate / Journalist / Courtroom, and three more
Decide the Horror split; decide Natural Disaster's fate
Themes reference sheet generator

E · App / Swift

5 items

Deliberately after A, so the layout contract is written against a real document rather than guessed at.

games: [GameTypeConfig]? on GuideConfig — the array survives in guide.json today, nothing reads it
case "game" in DocumentTemplates with a layout contract
Fix section ordering — both catalogs print alphabetically today
Implement the machine render target, replacing A's script
Decide whether family becomes an enum (means promoting to a real category)

F · Peregrine

5 items

The real test: eleven game types, ~40 mockups that function as the spec, and a pipeline already in place.

Create the guide with theme: "detective-noir" and eleven game types
Migrate the authored banks — vigilance-bank, Target Practice's 16 claims, Go Undercover's 8 rounds, Forensics' job pool
Script-validate index-pointer structure (break_idx) — the schema language can't express it
Decide where Press Raven sits against the existing Python pipeline
Split the guide past ~300 entries

Decisions only you can make

6 items

Theme↔game binding — migrate Drive Corps now, or wait for the theme id scheme? (Likely unblocked now that the catalog exists.)
Discovering as a sixth motivator
Whether layers and containers need a schema class, or stay visible as unplaced entries
The Horror split
Whether best_fit_mechanics stays curated-only — role overlap under-generates badly (Horror derives 1 against 4 curated)
Confirm build order follows asset tier, not enthusiasm — four document-tier themes share one pipeline and three pair with drafted mechanics

Known risks

Not tasks — things kept visible so they aren't rediscovered later.

The machine export gap.

Every downstream payoff of putting content in the corpus depends on a script nobody has written. Until then a game guide is a nicer place to author content that still has to be moved by hand.

Volume is proven only to 144.

Drive Corps at ~154 is a mild stress test. Peregrine is where it breaks.

Cross-guide refs are unvalidated by the app.

best_fit_mechanics → mechanic id and games[].mechanic → mechanic id are plain strings. A script checks them today; nothing in Press Raven does.

Both catalogs print alphabetically

until the template case lands, which makes the motivator and genre orders invisible on the page.

Scope gravity.

B, C, and D are pleasant and unbounded. A is the only avenue that tests whether the model works, and it is the one that will get postponed.