What Is Deliberately Absent

Every item on this list has been proposed at least once, and every one is absent on purpose. Adding one does not extend the system — it replaces the system's answer with a cheaper one.

• No quest log, no objective list, no markers. The map says there is more here without ever saying go here now. A log turns an open world into a checklist, and a checklist turns a player into an auditor.
• No levels, no XP, no character classes as mechanics. Advancement is change — in standing, in scars, in what people believe about you — not a tier. A class is welcome as fiction; it is not permitted as a branch in a schema.
• No endings. No mechanic may hold a terminal state with no successor. Threads close; the world does not.
• No initiative order. Combat is narrative. A turn tracker imports a ruleset this engine deliberately does not have, and once imported everything else has to agree with it.
• No root map in the engine. A System ships contentless; the map belongs to the installed world. The completeness report warns about the missing root map — that warning is accepted, not fixed.
• No model-specific branches. The engine degrades gracefully on a weaker model; it never detects which model it is running on and adapts. A branch you cannot test on every model is a branch that is wrong on most of them.
• No platform memory, combat or campaign-state system. Craft provides none. Behaviour lives in GM instruction files; durable state lives in schemas and files. There is no hidden layer doing any of this for you, which is the reason it can all be edited.

If a world you are building genuinely needs one of these, build it in the world, not in the engine. The engine's job is to keep not having it coherent.

Why the list exists at all

An absence with no reasoning attached gets re-proposed every few months by somebody acting in good faith, and eventually one of them gets added because nobody remembers why it was not there.

That is the actual purpose of this chapter. Not to forbid — to record. If you disagree with an entry, argue with the reason rather than rediscovering the idea.

Techniques considered and not adopted

Four standard approaches were evaluated and rejected on their merits. They are listed so nobody spends a session rediscovering the reasoning.

• Self-consistency — sample N reasoning paths, take the majority. It needs N passes; Craft gives one context and one pass. Its substitute here is the oracle's fixed-likelihood rule: set the odds before the roll, never adjust after.
• Behaviour trees and GOAP planners. The strategic layer is data plus a cadence, not a planner. A planner would make people legible and predictable, which is the opposite of the goal — and a legible NPC is one the player solves rather than meets.
• Markov chains. The engine is deliberately anti-Markov. recentTexture remembers the last several turns precisely so the next one does not repeat them; a chain would do the reverse and do it confidently.
• Pathfinding. Distance comes from the map tool, which measures natively. Reimplementing it in prose produces numbers that disagree with the map, and the map is the thing the player can see.

The test for anything you want to add

Ask what the system's current answer to this problem is, and whether you can state it.

If you cannot state it, the absence is a gap and worth filling. If you can state it and the new thing would simply be easier, you are proposing a replacement, and replacements have to beat the incumbent rather than merely exist.