Representation is a tactical material.

The player can reduce a signal’s precision to carry it more cheaply, route it through an exposed relay, or distribute it to several recipients. None of those actions establishes that its contents are true. This distinction lets a packet remain perfectly encoded while the claim it carries is wrong.

A relay encounter can reward the player who recognizes the valid packet and preserves its precision, while another build spends bandwidth broadcasting a coarse threshold. Both manipulate representation; they should produce visibly different transport and precision tradeoffs.

The same symbol can mean different things.

Semantics binds symbols to explicit contexts and registered meanings. Translation, reframing and disambiguation must state what mapping they use. Analogy is useful, but it is not factual support. Literal physical responses are authored allowlisted interactions.

An inscription puzzle should make its context legible through the room, not demand arbitrary word guessing. A clever build can preserve or scope a binding while a drift-oriented enemy changes another one.

The language itself becomes a reusable object.

Grammar handles inert syntax, named bindings, typed fragments and schema transformations. Quoting a graph does not execute it. Simplification cannot remove effects, lineage or charges. Every reconstructed graph still passes through the compiler.

Advanced players can abstract a reliable spell into a template, substitute a compatible fragment and preserve a versioned schema. The interface should make structural changes explainable without presenting arbitrary source code execution as magic.

Knowing begins with an available measurement.

Observation exposes visible fields, uncertainty and measurement age. It does not look through walls or authenticate a record’s origin. Missing samples become gaps; calibration cannot invent resolution. A remembered marker is not continuous tracking.

The player can illuminate a clue, instrument a construct or compare readings. Enemies can move within an error interval, break a sensor line, or change an unmeasured variable. Make those limits perceptible.

Loss should have a readable object.

Entropy attacks current values, maximum capacity, maintenance and recovery links through separate operations. Those distinctions matter because the healthy response changes: replenishing current shield helps against a drain but cannot remove a lowered capacity.

The player should recognize whether armor is rusted, a ward is unraveling or restoration is severed. Debt limits sustained manipulation. It should not become an excuse for unexplained random punishment.

A record is not a reset button.

Memory separates capturing, reading, replaying and restoring. A snapshot carries identity, generation and expiry; a recall reads without changing reality. Restoration must honor current capacity, navigation and permanent progression boundaries.

The strongest memory play rewards preparation and timing. An opponent can block a bookmarked position or sever a recovery link. Echoes replay eligible committed events once and never create an unlimited chain of echoes.

A relationship can carry a consequence.

Causality adds bounded event-response behavior to explicit links. It distinguishes querying an event’s instigator from tracing its longer path, and a modeled intervention from a live world edit. Every hop costs budget and recursion stays guarded.

Players can defend a mediator, isolate a target or sever an early link instead of absorbing the final cascade. The architecture should reveal cause and response in sequence, so the encounter teaches its own causal rule.

Timing changes without erasing history.

Temporality owns ages, intervals, local schedules and duration. A stretched status lasts longer; it does not heal its target. Hasten changes future windups, not every game clock. Stasis leaves damage and cooldown clocks running.

Readable time rings and half-open windows make timing a skill rather than a hidden exception. Enemies can commit an attack before slow lands, interrupt a synchronized group, or pressure a stationary construct.

A clever condition can be the spell.

Logic produces explicit Boolean results, bindings and premise-linked claims. It does not hide execution inside a truth operator. Imply computes material implication; Gate decides whether another expression runs. Unknown facts must remain visible as missing input rather than convenient false values.

A logic-focused build trades broad effect coverage for the right response at the right time. Give predicates readable thresholds and let enemies invalidate one premise. The player needs to understand why the graph chose not to cast.

Classification changes what can interact.

Ontology owns encounter types, admissible relationships and carefully bounded transformation. A membership query does not retype an enemy. A local subtype does not rewrite the world’s permanent taxonomy. These boundaries keep type manipulation powerful without becoming unrestricted reality editing.

A precise classification can turn a modest effect into a good answer for one enemy family. Broader classifications should lose specificity. Make temporary tags and relation proxies visible so opponents can escape, cleanse or attack them.

Uncertainty is something the player can shape.

Weights, distributions and explicit seeds let the player choose how much variability to tolerate. Sampling a distribution and changing its outcomes are different operations. A shared roll can create correlation, but never becomes knowledge that the underlying claim is true.

The interface should expose possible outcomes and bounds before a commitment. A low-variance build and a reroll-focused build can feel different without requiring opaque luck. Replays preserve exact keyed draws so observed behavior can be explained.

Alternatives help decide; they do not rewrite the past.

Counterfactual branches are copied scenarios with explicit assumptions, missing fields and finite horizons. A forecast is hypothetical. Contrast, regret and sensitivity mean something only when the compared baselines and metrics match.

The player can rehearse a plan, identify a robust choice and precommit conditionally. Live checks still happen at execution. An opponent can change an assumption and invalidate the forecast without the game pretending the simulation had perfect knowledge.

Identity has appearance, behavior and limits.

Persona works with masks, temporary archetypes, traits and owned aspects. A perceived faction never changes actual damage permissions. Copying an observed attack does not grant every behavior of its origin. Archetypes keep weaknesses as well as bonuses.

A familiar can wear a role or project a convincing decoy, while a witness-focused opponent reveals the mismatch. The player keeps control of their own actions; these mechanics affect NPC interpretation and bounded owned actors.

A claim needs a route back to its support.

Witnessing, citing, tracing, checking and attesting are separate operations. Ten copies of one record are not ten independent witnesses. An attestation authenticates origin, not truth. A supported claim may still become disputed when relevant counterevidence appears.

Restoration becomes a meaningful culmination of combat and observation. Canonize requires a supported claim, independent sources, no unresolved dispute and explicit Authority. Rejected commits spend no Authority, and required evidence remains recoverable.

What an actor notices is not what it must do.

Attention governs salience, focus, retained observations and allocation. A stimulus may influence an NPC without forcing its next action. An old last-seen position stays visibly stale. Observation owns field visibility; Agency owns an owned actor’s goals.

Decoys, divided attention and narrow perception create spatial tactics. Counterplay should be concrete: inspect the decoy, flank the cone, break sight or create stronger legitimate evidence of danger.

A plan is conditional, not a promise.

Agency separates goals, plans, evaluation and commitment. An owned construct has a bounded action allowance, an authored ability schema and safe fallback behaviors. Planning can fail honestly when no valid route exists inside its search budget.

A familiar can defend, pursue or hold without gaining unlimited autonomy. The interesting choice is which goal to reserve resources for, and when new evidence should make the player abandon that plan.