combat-and-spellcraft.md

music/combat-and-spellcraft.md

Novel Combat + Player Spell-Creation — full combat-canon enhancement

CANON SUBORDINATION — this document is PROPOSAL-TIER: it serves canon and never outranks it.
The canon: the CVD · the T1 foundation docs · the T0 registries (registries/) · the spine
(docs/spine/CH_*.md) · the region pages (_source/02_Tier_2_Region_Pages/). Authority order: docs/DOC_MAP.md § 0.
Canon served (scanned from this document's own citations — widen it by hand where it is thin):
CVD §12 · T1_Ability_Tree · T1_Combat_System_Spec · T1_Trades · (+1 more tier) · T0_Ability_Tree_Registry · T0_Weapon_Registry
READ THAT CANON FIRST — open it and derive from it before you build anything from this document.
If this document disagrees with canon, CANON WINS and this document is the defect — fix the
document, never the canon. Nothing here is applied until it is ratified into canon.

Status: PROPOSAL (Opus 4.8, 2026-07-13). Grounds in T1_Vril_and_Magic_System_Master v2.2, T1_Combat_System_Spec v1.0, T0_Ability_Tree_Registry (185 rows), docs/COMBAT_ENCOUNTER_SYSTEM.md, docs/RUNTIME_GENERATIVE_LAYER.md, docs/proposals/DIFFICULTY_SYSTEM.md, AAA_REARCHITECTURE_BRIEF.md (Layer-model v2), CVD §12/§13/§17. Nothing here edits canon; it proposes enhancements, splits them into lock-now / Josh-ruling / Fable-5-handoff, and flags every count/§17 touch.

This is the deepest system dive. It does three things: (1) resolves the local-model-tier question with a concrete recommendation; (2) reviews and enhances the existing combat + weapons + abilities + vril corpus; (3) designs the marquee new system — player Spellcraft: a proficiency + rune-component + generative-merge authoring layer that produces genuinely personal spells without adding a single new ability, weapon, or mode to the locked counts.

---

1. Vision

Humanity already has the best difficulty idea in the genre — *difficulty is a density knob on the world, and that knob is the cosmology* (COMBAT_ENCOUNTER_SYSTEM.md §0). It already has Composure you break with any of seven modes, a Withdrawal clock instead of an enrage bar, and a boss that reads your habits. What it does not yet have is the thing that makes a magic action-RPG a decade-long obsession: a language the player speaks back. Right now the player is handed 185 finished verbs. The enhancement is to turn those 185 from a *fixed spellbook* into a *closed vocabulary* over which the player writes their own sentences — inscribing rune-components, directing a generative merge, and binding personal spells that carry their own vril-signature into battle.

The pitch in one line: the 185 abilities are the alphabet; runes are the grammar; the player is the author; and vril physics — real mass, heat, conductivity, structural integrity — is what keeps every authored spell honest. You do not pick "Fireball" off a bar. You extract iron from an arena's ore-vein, superheat it molten, magnetise it, and fire it as a rail-slug you designed — because that boss's Composure node is weak to exactly that, and you *worked out* that it was. Then you name the spell, the world remembers your signature, and you never see another player's identical build. That is Layer-1 novel-combat, Layer-2 completionist obsession, and Layer-3 "it was all real inscription lore" — in one system that respects every hard line.

---

2. The design & the moment-to-moment loop

2.0 The load-bearing distinction: AUTHOR vs MODULATE vs CAST

Josh's idea-seeds mix two different activities that must be split cleanly, because one is safe to be slow and generative and the other must be deterministic at 60 Hz:

Getting this split right is what makes the whole thing shippable: authoring can be as clever as we like because a wrong output is a menu re-roll, not a dropped frame; modulation and casting never touch the model.

2.1 The atomic layer — the 185 abilities as a closed vocabulary (enhancement, not addition)

T0_Ability_Tree_Registry already carries the exact columns a composition grammar needs, and they are ~90% populated:

The enhancement is to treat these columns as the spell grammar's type system, and to finish populating them uniformly for all 185 rows. No new ability is authored; instead each existing ability declares (a) what elemental *material* it produces or consumes, (b) its default *shape/delivery*, and (c) its *physical properties*. Once every row is typed, the 185 stop being a static list and become a combinable vocabulary — which is exactly the closed-action-vocabulary discipline the strategy-token architecture already uses (RUNTIME_GENERATIVE_LAYER §2.1: "the emergence lives in ordering/emphasis; the vocabulary is closed").

Firewall: the count stays 185. A player spell is a *parameterised composition* of vocabulary entries, not a 186th ability.

2.2 The physical-property engine — where the novelty actually lives

The physical_properties_array turns element magic into a small material simulation, and that is the source of the genuinely novel loop. Properties are not flavour text; they are inputs to deterministic interactions the player can chain:

Josh's worked seed, expressed as a real chain: *extract metal from earth → melt → forge → fire bullets* becomes:

Earth (ore extraction, produces metal-material construct)Fire (thermal_state → hot past melt-point, structural_integrity → molten)Earth/shape modulation (structural_integrity → reformed as slugs)Earth+Lightning (conductivity + magnetic — Magnetic Pull/Lock are authored T3/T5)rail-slug volley. Every step is an existing ability whose physical_properties make the next step legal. The player is not scripting; they are doing applied vril-physics, and the sim adjudicates whether the chain is coherent.

This is the addictive core: the boss's Composure node (§5) has a physical weakness, and the player invents the material-chain that exploits it. You craft the key to the lock. No shipped AAA game exposes a physics-grounded elemental-property simulation as its spell-authoring substrate.

2.3 The ability-modifier grammar (Josh's seeds, generalised past them)

The grammar is four operator families applied to a base ability. Each operator is unlocked by proficiency and gated by an owned rune-component (§2.4). This is a superset of Josh's list:

SHAPE operators (what the vril takes as form):

DELIVERY operators (how it leaves you and travels): maps onto cast_pattern + range_class.

MOTION / TARGET-LOCK operators (Josh's real-time modulation set — the marquee feel):

TRANSFORM-CHAIN operators (the physical-property pipeline of §2.2): extract, melt, freeze, forge/shape, magnetise, ignite, conduct — each is a legal transition on the physical_properties state, checked by the sim.

MOBILITY (pre-activated) operators — Josh's ice-slide / wind-hover / ground-wave / ride-boulder. These are the existing mobility and mobility_combat_hybrid ability classes (8 + 2 rows) extended into traversal-that-is-also-a-cast: pre-activate an Earth wave under your feet → surf it (ground-wave); a Water base + freeze → ice-slide; an Air base + field under you → wind-hover; move a heavy Earth construct and stand on it → ride the boulder. Mobility spells are their own bindable class so they compose with the platforming/agility-course system (Combat_Spec §10) rather than sitting only in combat.

ELEMENT-COMBINE operator — the Layer-B matrix (§2.1) exposed as a live fusion: hold two vril-seat elements active and the composer offers the authored pair-ability as a craftable base (Water+Fire → Steam/Plasma). At high proficiency, a real-time combine (soak then shock) is a modulation, not just an authored pair.

2.4 Runes — the component system (Josh's "rare/epic runes that make it possible")

Corrected 2026-07-13 (Josh). Inscription/runes are WORN, not consumed — a permanent equipped enhancement layer, *"except when turned into something else or learned like the spell creation."* Josh's hard mandate: no ordinary-rune grind. "I don't want this game where you have to constantly farm and craft ordinary runes to enhance — mindless grinding to make combat unfun is not the goal." So the acquisition-heavy Common/Rare/Epic/Legendary *farm-and-slot* framing an earlier draft implied is REPLACED by the worn-layer + single-catalyst model below. And consumables (bombs, potions, poisons) are NOT inscription — they belong to Engineering (Trade 8) and Alchemy (Trade 5). See the dated Correction section at the foot of this doc.

Runes are already triple-canonical, so this invents nothing: (a) RUNES is the Sorcery resource (Orb generates on basic attacks, Tomb spends — Combat_Spec §Resource economy); (b) "runic weapon alterations" are a canon worn enhancement unlocked by Trade 3 Blacksmithing + Trade 9 Inscription at Adeptus Minor (T1_Trades §9.1Weapons_Enhancements_Alterations); (c) the game's XP curve is literally called the "runesque progression curve" (T1_Ability_Tree). The enhancement resolves these into three cleanly-separated senses — and only the third is ever consumed:

Rune-inscriptions and catalysts are acquired canon-true ways that do not reward mindless farming: Inscription-trade crafting (Trade 9 masters, home bases Ch 15 Egypt + Ch 29 Sámi — the worn alteration surface), milestone/boss/vril-site rewards (Sigma-bounded, RNG_Sigma, as one-time or bounded drops, never a grind faucet), and cultural-tradition unlocks as the player learns each region's real script (cuneiform, Maya glyph, Elder Futhark) — where the era-rhyme and assimilation systems compose (a rune-aesthetic learned in an ancient chapter returns in the modern arc). If a design pass ever needs a repeatable-farm sink to make this work, that is a [DECISION NEEDED - JOSH], because the standing mandate is the opposite.

Consumables belong to other trades — NOT inscription (Josh, 2026-07-13): bombs, potions, and poisons are crafted through Engineering (Trade 8 — form-and-mechanism consciousness) and Alchemy (Trade 5 — substrate-transmutation), with Herbalism (Trade 4) feeding the botanical inputs to Alchemy's potion/preparation line (canon: T1_Trades §Apothecary "plant materials feed potion and elemental preparation"; Alchemy+Herbalism "elemental infusion preparations" at Adeptus Minor). Inscription owns the worn enhancement layer and the spell-creation catalyst only — it does not craft consumable combat items. This keeps spellcraft (worn + learned) and consumables (crafted-and-used-up via Engineering/Alchemy) as cleanly separate economies.

Care firewall (hard, unchanged): protected-culture inscriptions are reference-only and never craftable or wearable rune-inscriptions — no joik-rune, no songline-rune, no Vodou-rune (Ch 29 Sámi, Ch 57 Australia, Ch 42 Haiti, HL_0002/§17.1). The authoring validator hard-fails any spell that would instrumentalise a protected tradition. Vril reads mathematical/physical/substrate throughout — the physical_properties engine *reinforces* this (mass/heat/conductivity, never race-mystical "polar energy"; the Vril-Society firewall holds).

2.5 The generative merge (the "AI birthing," disciplined)

At authoring time the player gives a direction — either by wiring a node-graph (base ability → shape rune → transform → delivery → lock; fully deterministic, the shippable floor and the accessibility path) or in natural language ("a slow homing sphere of molten iron that splits into three on release"). The composer merges direction + owned runes + active vril-seat elements into a candidate spell-token:

spell_token = {
  base_ability_id  ∈ the 185 (closed),
  element          ∈ legal-for-base ∪ active-vril-seat,
  shape            ∈ owned shape-runes,
  delivery         ∈ owned delivery-runes (cast_pattern),
  motion_lock      ∈ owned motion-runes (split N / direction_lock / hold_to_home / orbit),
  transform_chain  ∈ ordered, physically-legal transitions over physical_properties_array,
  knobs            ∈ ranges (charge_max, split_N ≤ vril-seat tier, home_window),
  worn_runes       ∈ equipped rune-inscriptions (worn, NOT consumed at cast — §2.4),
  vril_cost        = DERIVED deterministically, drawn from the depletable VRIL pool per cast
                     [+ RUNES-resource cost only for Sorcery-off-hand-routed casts]
                     (never author-set — balance integrity)
}

This is the exact strategy-token discipline from RUNTIME_GENERATIVE_LAYER §2, applied to authoring instead of speech:

1. Context pack = the player's owned runes + unlocked proficiency + active vril-seat elements + the physical_properties type-graph + the §17 rule vector (perfect-precision recall — the composer can only reference what the player actually has).

2. Grammar-constrained decode = the token schema; an out-of-vocabulary ability, an unowned rune, or an illegal physical transition is unsampleable, not merely rejected.

3. Deterministic validator = element legality, rune ownership, physical-property coherence (you cannot "melt" something with flammability:zero, thermal_state:cold past its melt-point without a heat source in the chain), proficiency gate, split_N ≤ tier, and the care/reveal denylist. PASS → the spell binds and the player names it. FAIL → the composer explains *why* ("you don't own a hold-to-home rune"; "molten iron needs a Fire step first") — the failure is *teaching*, mirroring the death-knowledge economy.

4. Authored/deterministic fallback = the node-graph editor. If the LLM is absent, low-end, or the NL parse fails, the player composes the identical token by hand. The system is 100% complete with the model turned off — the model is the natural-language front-end and the "surprise-me / merge-my-two-favourites" creative suggester, never a dependency (the core safety guarantee of the whole runtime layer, restated for authoring).

Because vril_cost is derived by the balance formula, not author-set (and drawn from the depletable vril pool, not a free per-cast rune-currency), the player cannot craft an overpowered spell — power comes from cleverness of composition against a specific lock, and the cost scales with what you asked for. Balance integrity is deterministic.

2.6 Bringing spells into battle without a mana-hotbar (the hard-line resolution)

Corrected 2026-07-13 (Josh). An earlier draft of this section read T1_Vril §1 as "no mana pool." That was WRONG — see the dated Correction section at the foot of this doc. T1_Vril §1 says the opposite about the resource: *"It is chargeable through recharge sources and depletes through vril-intensive actions"* and *"Abilities cost vril as resource, not mana as arbitrary pool."* A depletable, rechargeable vril resource pool EXISTS and is spent per cast. What §1 forbids is (a) the spell-casting hotbar UI and (b) an arbitrary abstract mana meter — not the existence of a depletable resource. The paragraphs below are corrected accordingly.

T1_Vril §1 is explicit about the two forbidden things: no spell-casting hotbar, and vril is not an arbitrary mana pool ("Humanity does not have a spell-casting hotbar magic system … Abilities cost vril as resource, not mana as arbitrary pool"). It is equally explicit that the resource itself depletes and recharges. Josh wants spells "to bring into battle." These reconcile cleanly because a bound spell is not a hotbar entry — it is cast through the existing consciousness-operation + weapon-channel input grammar, paid for out of the depletable vril pool:

Two distinct casting registers, and the correction matters for both:

So "bringing a spell into battle" is not selecting off a mana wheel mid-swing and is not inscribing base abilities onto gear — it is having *climbed the gates* (for the 185) or *authored-and-learned* the merge (for custom spells), then channeling from within/around against the depletable vril pool. The weapon is the spell's delivery organ (§2.6 weapon-channel bullets below), not a required slotting surface. Modulation (§2.0) then shapes the live cast.

2.7 Proficiency progression (rides the 12-tier, adds no new track)

Spellcraft does not become an 11th trade and does not add a mastery track — both would break locks. It rides existing rails:

2.8 The moment-to-moment loop (how it plays)

1. Read the world's tell and the boss's Composure nodes (§5) — one node is physically-typed (e.g. a structural_integrity:high carapace weak to melt→shatter).

2. In the pre-fight/rest window (or on a prior death, having *learned* the node — death-knowledge economy), author or retune a spell whose transform-chain targets that node. Bind it to your Wand/Chakram.

3. In the fight, cast the primary verb; when the node exposes, cast your authored counter, modulating live — charge-hold to raise the melt, split to hit three carapace plates, or hold_to_home to thread it past adds.

4. The node breaks on the *mode-correct physical answer*; the phase evolves (evolve-not-swap); you re-read and adapt.

5. Win or die, you bank knowledge — a death reveals the node's physical type as a Grimoire fragment, so the *next* authoring pass is sharper. The realm remembers your signature; an NPC with substrate-perception later *recognises the spell as yours*.

This is Souls-grade demand answered not by twitch alone but by authored problem-solving under time pressure — a combination the genre has never shipped.

---

3. Composition with existing canon (what it enhances, node by node)

---

4. Model-tier resolution (Josh's explicit ask)

Josh's instinct is right and can be stated sharply: the marquee "reasoning boss" needs no strong LLM — because a boss's tactics should not come from an LLM at all. The recommendation splits the question by job:

JobRecommendationWhy
Combat tactics / boss "what is it trying to do" / attack selection / phase logic / spell resolutionDeterministic code — NO LLM, any tier.The ~30-clock sim + Sigma-bounded habit-adaptation (COMBAT_ENCOUNTER_SYSTEM §4/§8) already produces adaptive, learnable, speedrun-stable tactics *better* than an LLM and at 60 Hz. An LLM here would be slower, non-deterministic, and add nothing to *what* the boss does. This is where Josh's "we likely don't need a strong LLM" lands hardest: we don't need one at all.
Voice / barks / faction intent / NPC dialogue / fauna flavour (the strategy-token layer)One shared 3–4B (Qwen3-3B / Phi-4-mini class), Q4_K_M, + 0.5–1B speculative draft.Exactly the RUNTIME_GENERATIVE_LAYER §B conclusion. Short, persona-conditioned, heavily context-packed outputs where sub-4B ≈ 8B once the pack carries the facts and the validator catches misses.
Spell-authoring merge (natural-language front-end + creative suggester)Same shared 3–4B on the baseline; opt-up to 7–8B on ≥12 GB-VRAM PC.Authoring is off the hot path (seconds of budget), so a bigger model *can* earn its keep on richer NL parsing — this is the one place Josh's "7B" instinct is correct, and it costs nothing at runtime because it is not frame-critical. Never a dependency: the node-graph editor is the deterministic floor.

Net resolution: ship deterministic-only for every combat-critical decision and all spell execution; ship the 3–4B shared model for voice/flavour/faction and the authoring merge; allow a 7–8B opt-up for the authoring front-end on capable PCs; and degrade the entire LLM layer to authored/node-graph fallback with zero content or correctness loss (§B.9 graceful-degradation ladder). The "reasoning-boss risk" dissolves because bosses reason via the clock-sim; the model only ever talks and assists authoring. A deterministic strategy-token for the boss (a small GOAP/utility selector) is not just an acceptable fallback — it is the primary and recommended path; the LLM is never in the boss's tactical loop.

---

5. Alternatives considered (steelmanned, then chosen)

Fork A — Where do player spells live relative to the 185 lock?

Fork B — The no-mana-hotbar hard line vs "bring spells into battle."

Fork C — Model tier (see §4).

Fork D — Where does authoring happen? *Workshop/rest (off hot path)* for AUTHORING vs *live* for MODULATION. [RECOMMEND the split] — authoring is slow and generative-safe; modulation (split / direction-lock / hold-to-home / charge) is live and deterministic. Conflating them is the trap that would either bloat the frame budget or dumb-down the authoring.

---

6. Canon composition & firewalls (adversarial pass)

---

7. Opus-can-spec-now vs Fable-5 handoff

OPUS-CAN-SPEC-NOW (lockable + wireable this session):

NEEDS-JOSH-RULING (genuine canon/numerical/§17 forks — options + rec in §5/§6):

FABLE-5 HANDOFF (deep implementation / population / tuning, well-scoped):

---

8. The AAA hook (one line)

You don't pick spells off a bar — you channel from the vril within and around against a depletable pool, wear inscribed runes that shape what you channel, and (rarely) spend a spell-creation catalyst to author a spell that is physically the key to a specific boss's lock; the world learns your signature, and no two players ever fight the same way. Souls-grade demand answered by authored problem-solving, on a real elemental-material simulation, with the local model as pure creative upside over a deterministic spine that cannot lie — every count, every hard line, and the no-mana-hotbar constitution intact.

---

Correction (2026-07-13, Josh): the resource + inscription model

This section supersedes any earlier statement in this doc that conflicts with it. Four things were corrected against canon (T1_Vril §1/§5/§10, T1_Combat_System_Spec Resource economy, T1_Trades §3.4/§3.5/§3.8/§3.9/§9.1).

C1. A depletable VRIL resource pool EXISTS — the "no mana pool" reading was WRONG

Josh: *"vril and mana and chakra and prana and all of that is still LIFE ENERGY. we can DEPLETE it and need to RECHARGE and/or REST."*

What T1_Vril §1 actually says (quoted, so we stop mis-citing it):

Reconciliation: §1 forbids exactly two things — (1) the spell-casting hotbar UI (shop/slot a spell onto a wheel), and (2) treating vril as an arbitrary abstract mana meter divorced from the physics. It does not forbid a depletable resource — it explicitly affirms one ("depletes … cost vril as resource"). The depletable, rechargeable vril pool is canon. Every cast draws it down. It refills via the canonical recharge network — waterfalls (passive), obelisks, pyramids, tombs, orbs (§5), passive atmospheric absorption, out-of-combat regen, and rest. The earlier draft's "no mana pool spent per cast" contradicted §1's own words and is retracted. The distinction that survives: vril is a *physically-grounded, density-gated, recharge-sourced* pool, never a free-floating arbitrary mana bar, and never selected from a hotbar.

C2. The resource bars — what is canon vs what is NEW (flagged, not asserted)

Josh's model: ~4 bars — HEALTH / VRIL (mana) / STAMINA / RAGE — each with a MAX that grows via (a) level/character progression and (b) Zelda-style exploration rewards (heart-container / stamina-vessel analog from puzzles + easter eggs).

Grounding against T1_Combat_System_Spec (Resource economy = five systems) and T1_Vril:

Exploration-grows-MAX (Zelda heart-container / vessel analog):

C3. Inscription is WORN, not consumed — no ordinary-rune grind

Josh: *"I don't want this game where you have to constantly farm and craft ordinary runes to enhance — mindless grinding to make combat unfun is not the goal,"* and inscription is *"worn, not consumed, except when turned into something else or learned like the spell creation."*

Model (canon-true; see §2.4 as corrected):

C4. Consumables belong to Engineering + Alchemy — NOT Inscription

Josh: *"We have Engineering as a trade and Alchemy as a trade for those consumable things — we'll actually be able to craft bombs and potions and poisons."*

Canon-status summary (for the flag log)

C5. Enchantment runes — a Borderlands-style affix lane (Josh, 2026-07-13 follow-up)

Josh, refining C3/C4: *"maybe there are some consumable runes — we could further enhance consumable bombs and things like that to create different effects or features (think Borderlands 4 enchantments on weapons and grenades) — maybe that can be innocent RNG where we can re-roll and improve those enchantment slots."*

This SUPERSEDES the C4 line "no consumable-enhancer-rune exists" and adds a third, distinct rune lane — without reopening the grind C3 barred. There are now three non-overlapping enhancement lanes:

1. WORN inscription (C3) — a permanent vril-body/gear enhancement, never consumed, never farmed.

2. Spell-creation catalyst (C3) — a rare/epic rune consumed once at authoring into a *learned* custom spell (a merge of a few of the 185); thereafter runeless.

3. ENCHANTMENT runes (NEW) — a consumable affix layer that sockets enchantment slots on gear + consumables (weapons, grenades/bombs, potions, poisons) to add effects/features: a fire-affix grenade, a homing weapon-enchant, a lifesteal poison, a chain-lightning bomb.

The anti-grind guarantee — "innocent RNG." The enchantment rolls are re-rollable and improvable: you *tune toward the build you want* (Borderlands-4 / Diablo theorycrafting-and-reroll), and are never luck-gated into farming a thousand drops for one roll. The fun is OPTIMIZATION, not attrition — which is exactly Josh's mandate ("mindless grinding to make combat unfun is not the goal").

Composition: Engineering (Trade 8) + Alchemy (Trade 5, Herbalism-fed) craft the base bombs/potions/poisons (C4); enchantment runes are the affix layer that makes each one *yours*; weapons/gear carry enchant slots too. The worn-inscription lane (C3) stays worn-not-consumed and untouched. Clean separation: three lanes, three feels (permanent identity / one-time spell birth / re-rollable affix optimization).

[DECISION NEEDED - JOSH] (numerical/system canon): enchantment-rune SOURCE (crafted vs dropped vs both), which items carry slots (consumables only vs +weapons/gear), slot counts per item tier, the affix/effect pool, and the re-roll cost/curve. All numerical — flagged, not asserted.

C6. The HUD / resource layout + the exploration max-growth mechanic (Josh, 2026-07-13)

Josh specified the on-screen resource layout and how the maxes grow — which RESOLVES the C2 "four-bars-vs-canon's-five-archetype-keyed-systems" flag. There are not four always-on bars: there are Health + Vril always, plus ONE weapon-swapped Stamina/Rage slot, with Runes/Ammunition as archetype overlays.

Layout (the left cluster):

Max-growth mechanic (heart-container / MK-style): exploration / puzzle / achievement pickups come as fifths — collect 5 pieces = one full additional unit that overlaps/stacks (a full health bar, a full stamina/rage bar, OR a full mana-bubble layer). Max also grows with level/character progression; the two sources compose. This makes Zelda-style exploration + easter eggs tangibly reward the core power fantasy.

Runes (Orb/Tomb) + Ammunition stay the sorcery/ranged build-overlays (per C2's rec) — surfaced only with the relevant weapon/archetype, never a 4th/5th always-on bar.

[DECISION NEEDED - JOSH] (numerical): the fifths-per-full-unit (5?), the cap on stacked health bars / bubble-layers, the exploration-pickup counts across the 77 chapters, and whether Rage keeps build/decay vs deplete-from-full. Flagged, not asserted.

Generated by harness/site/structure_site.py — the URL path is the repo path. review root