Restructure balancing docs into docs/balancing/

Split the cluttered progression_design.md and content_design.md into
separated documents by role:

- docs/balancing/rules.md    - design rules and principles (moved from
                               progression_design.md, which is removed)
- docs/balancing/targets.md  - the chosen base numbers: run shape,
                               factory curve, threat ladder (achieved
                               values adopted), combat anchors, pacing
                               anchors
- docs/balancing/derived.md  - current tuned state of all derived
                               numbers, mirroring the configs
- docs/balancing/process.md  - pass order, tuning discipline learned in
                               arena rounds 1-5, tools, next-round
                               checklist
- docs/balancing/history.md  - chronological record: decisions, numbers
                               pass, calculator bugs, arena rounds,
                               pacing pass
- docs/balancing/README.md   - index, status, open action items (moved
                               from progression_design.md)

content_design.md slims back down to actual content: footprint gating,
hull grids, gating matrix (module names updated to railguns), and the
production tree structure with its fiction - numbers, anchors, arena
logs, and pacing all moved to docs/balancing/. Stale references in the
config comments updated to the new locations.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DyCu8vwChKMbLJQ3xosYEN

docs/balancing/README.md

@@ -0,0 +1,56 @@
+# Balancing Documentation
+
+Everything about balancing Dota Factory, separated by role:
+
+- **[rules.md](rules.md)** — the design rules and principles. Timeless;
+  changes only when the design changes.
+- **[targets.md](targets.md)** — the base numbers (roots/anchors) chosen
+  by design. Change these first; everything else re-derives.
+- **[derived.md](derived.md)** — the current tuned state of all derived
+  numbers, mirroring the configs. Updated whenever configs change.
+- **[process.md](process.md)** — how balancing is done: the pass order,
+  tuning discipline, tools, and the checklist for the next round.
+- **[history.md](history.md)** — chronological record of decisions,
+  findings, bugs, and arena rounds.
+
+Related: game content (hull grids, footprint gating, tree design and
+fiction) in [../content_design.md](../content_design.md); rules with
+REQ-* ids in [../requirements.md](../requirements.md).
+
+## Status
+
+First full balancing round complete (2026-07-06): targets → tree →
+numbers → threat-calculator parity → combat stats (arena-converged) →
+pacing. Next step: full-game playtests against the run-shape targets in
+`targets.md`.
+
+## Open action items
+
+Agreed changes that require edits to `requirements.md`, the code, or the
+configs. Completed items are removed (their outcomes live in
+`requirements.md`, `history.md`, and the git history).
+
+1. **Fill unfillable schematic slots with artifacts.** With duplicates
+   removed, the schematic drop pool can run dry — previously unreachable.
+   Decision: every slot in the choice dialog that cannot be filled with a
+   schematic because the eligible pool is exhausted is filled with an
+   artifact option instead (in addition to any artifact option granted by
+   the regular artifact roll). A push therefore always awards a full
+   dialog. Update REQ-DEF-SCHEMATIC-DROP.
+2. **Confirm wave scaling in playtests.** `threat_rate_formula` is the
+   only time-scaling axis; verify the tuned curve (see `derived.md`)
+   produces the intended difficulty race in real runs.
+3. **Gate shortcut-recipe drops on their inputs.** Extend the assembler
+   recipe schematic pool eligibility in REQ-DEF-SCHEMATIC-DROP: in
+   addition to the existing station-level and output-item checks, all of
+   the recipe's input item types must be implicitly unlocked as well.
+4. **Resource deposits.** Add a terrain deposit layer per the Resource
+   deposits rules (`rules.md`): deposit patches generated in expansion
+   columns (deterministic content per expansion, randomized placement
+   within the new columns), deposit rendering, and a miner condition (a
+   resource recipe is selectable only if the miner's footprint overlaps
+   at least one matching deposit tile). Touches REQ-BLD-MINER ("every
+   asteroid tile is equivalent" no longer holds),
+   REQ-GW-ASTEROID-EXPAND / REQ-EXP-*, `world.toml`, and `visuals.toml`.
+   Until this lands, quartz mines anywhere and the mid-game is
+   knowledge-gated only.

docs/balancing/derived.md

@@ -0,0 +1,164 @@
+# Derived Values (current tuned state)
+
+Everything here is derived from `targets.md` under the rules in
+`rules.md`, and mirrors the config files. Item threats, ship threats,
+ratios, and belt checks are verified by `tools/threat_report.py` — re-run
+it after any recipe or material change and update this file when values
+move. Combat stats were tuned empirically against the arena suite in
+`bin/balancing/data/balancing.toml` (round-by-round record in
+`history.md`).
+
+## Economy constants
+
+- `scrap_per_threat = 0.25` — 1 scrap per 4 threat destroyed (a cruiser
+  kill drops ~59 scrap); threat(scrap) = 4.
+- Scrap smelting: 1 scrap → 1 iron_ingot, 1 s — deliberately
+  value-losing; reprocessing is the value-preserving path.
+- Reprocessing: 4 scrap per cycle, 4 s; full-pool weights iron_ingot 30 /
+  copper_ingot 30 / silicon 20 / voidsteel 20 → threat(voidsteel)
+  = (4·4 + 4)/0.2 = 100.
+- `scrap_despawn_seconds = 120` (a capital kill drops hundreds of scrap,
+  collected one per salvage cycle).
+
+## Recipes and item threats
+
+(dur in seconds; threat is per output unit)
+
+| item | recipe | dur | out | threat |
+|---|---|---|---|---|
+| iron_ore / copper_ore | miner | 1 | 1 | 1 |
+| quartz | miner (deposit) | 2 | 1 | 2 |
+| iron_ingot | 1 iron_ore | 1 | 1 | 2 |
+| copper_ingot | 1 copper_ore | 1 | 1 | 2 |
+| silicon | 1 quartz | 2 | 1 | 4 |
+| steel_plate | 2 iron_ingot | 3 | 1 | 7 |
+| copper_wire | 1 copper_ingot | 1 | 2 | 1.5 |
+| copper_coil | 2 copper_wire | 1.5 | 1 | 4.5 |
+| building_block | 2 steel_plate | 2 | 4 | 4 |
+| control_chip | 1 silicon + 2 copper_wire | 5 | 1 | 12 |
+| capacitor_bank | 2 copper_coil + 1 silicon | 5 | 1 | 18 |
+| hardened_steel | 3 steel_plate | 12 | 1 | 33 |
+| ceramic_plate | 2 quartz | 4 | 1 | 8 |
+| drive_unit | 2 steel_plate + 2 copper_coil + 1 control_chip | 8 | 1 | 43 |
+| voidsteel_plate | 1 voidsteel + 1 hardened_steel | 8 | 1 | 141 |
+| capital_core | 2 voidsteel + 1 capacitor_bank + 1 control_chip | 10 | 1 | 240 |
+
+Shortcut recipes (drop-only; item threat stays defined by the base path
+via the max rule): `shortcut_steel_plate` 3 iron_ore → 1 plate (2 s,
+level 1), `shortcut_control_chip` 2 quartz → 1 chip (4 s, level 2),
+`shortcut_hardened_steel` 4 iron_ingot → 1 hardened (8 s, level 2).
+
+Ratio curve realized: t1 all 1:1 (miner:smelter); t2 clean 2:3
+(ingot→plate, wire→coil); t3 strange — 2:5 (silicon→chip), 3:5
+(coil→capacitor), 3:4 (plate→hardened, plate→drive); t4 inverted 3:2
+(hardened→voidsteel_plate). Belt check: worst input demand 1.33 items/s,
+under the ~2/s single-belt cap everywhere.
+
+## Module prefabs
+
+(contribution = item threat + module production time)
+
+| module | recipe | dur | mod. time | contribution |
+|---|---|---|---|---|
+| railgun_s | 1 copper_coil | 1 | 1 | 6.5 |
+| salvager | 1 steel_plate + 2 copper_wire | 2 | 1 | 13 |
+| repair_tool | 1 steel_plate + 2 copper_wire | 2 | 1 | 13 |
+| armor_plates | 4 steel_plate | 3 | 1 | 32 |
+| maneuvering_thrusters | 1 steel_plate + 1 copper_coil | 2 | 1 | 14.5 |
+| sensor_booster | 2 copper_wire + 1 copper_coil | 2 | 1 | 10.5 |
+| afterburner | 2 copper_coil + 1 steel_plate | 3 | 1 | 20 |
+| weapon_stabilizer | 1 steel_plate + 1 copper_coil | 2 | 1 | 14.5 |
+| weapon_primer | 1 capacitor_bank + 1 copper_coil | 4 | 2 | 28.5 |
+| weapon_upgrade | 1 control_chip + 1 copper_coil | 4 | 2 | 22.5 |
+| railgun_m | 1 capacitor_bank + 2 steel_plate + 1 copper_coil | 4 | 3 | 43.5 |
+| drone_bay | 1 control_chip + 2 steel_plate + 1 copper_coil | 4 | 3 | 37.5 |
+| railgun_l | 1 capacitor_bank + 2 hardened_steel + 1 ceramic_plate | 6 | 4 | 102 |
+| drone_hangar | 1 voidsteel_plate + 2 control_chip + 1 drive_unit | 10 | 6 | 224 |
+
+## Ships
+
+(fitted = hull item + ship base time + default loadout; the default
+loadouts are the `default_modules` used by enemy waves and are
+geometry-validated against the hull grids)
+
+| ship | hull recipe | dur | base | default loadout | fitted |
+|---|---|---|---|---|---|
+| drone | 1 iron_ingot | 1 | 1 | railgun_s | 10.5 |
+| frigate | 2 steel_plate + 1 copper_wire | 2 | 2 | 2× railgun_s, maneuvering_thrusters | 47 |
+| destroyer | 3 steel_plate + 2 copper_coil | 4 | 3 | 3× railgun_s, armor_plates, sensor_booster | 99 |
+| cruiser | 2 hardened_steel + 2 control_chip | 6 | 4 | 2× railgun_m, armor_plates, maneuvering_thrusters | 233.5 |
+| battlecruiser | 3 hardened_steel + 2 control_chip + 1 drive_unit | 8 | 5 | 3× railgun_m, armor_plates, 2× railgun_s | 354.5 |
+| battleship | 3 voidsteel_plate + 1 drive_unit + 2 control_chip | 10 | 6 | railgun_l, 2× railgun_m, weapon_stabilizer, 2× railgun_s | 722.5 |
+| dreadnought | 5 voidsteel_plate + 1 capital_core + 2 drive_unit | 12 | 8 | 3× railgun_l, 4× armor_plates, railgun_s | 1491.5 |
+| carrier | 5 voidsteel_plate + 1 capital_core + 2 drive_unit | 12 | 8 | drone_hangar, 2× railgun_m, 2× armor_plates, sensor_booster | 1436.5 |
+
+## Combat stats
+
+(arena-converged, 2026-07; see `history.md` rounds 1–5)
+
+**Weapons:** railgun_s 2 dmg × 2.0 Hz (4.0 DPS), range 50 m;
+railgun_m 14 × 1.5 (21), range 70; railgun_l 52 × 0.8 (41.6), range 100.
+
+**Hull HP** (15/threat prior + empirical trims): drone 60, frigate 300,
+destroyer 550, cruiser 1500, battlecruiser 2400, battleship 6300,
+dreadnought/carrier 24000.
+
+**Mobility ladder** (speed m/s | main accel | maneuvering | angular |
+max rot): drone 45|60|30|12|6, frigate 35|45|22|8|4,
+destroyer 30|35|18|6|3, cruiser 24|25|12|4|2, battlecruiser 20|20|10|3|1.5,
+battleship 15|14|7|2|1, dreadnought/carrier 10|8|4|1|0.5.
+Sensors: 150/200/220/250/260/280/300/350 m.
+
+**Other modules:** armor_plates +1200 HP; repair_tool 9 HP × 1 Hz,
+range 80; salvager range 60, cargo 20, 0.5 collections/s; afterburner
+×1.6 speed +60 accel; maneuvering_thrusters ×1.2 speed +10 maneuvering;
+sensor_booster +50 m; weapon_upgrade ×1.2 damage; weapon_primer ×1.2
+rate; weapon_stabilizer ×1.3 range ×0.8 rate.
+
+**Stations:** HQ 5000 HP. Player station 3000 HP, 25 dmg × 1 Hz,
+range 120, scrap 40. Enemy station: 3000+1500x HP, 25+12x dmg,
+1.0+0.1x Hz, range 120, scrap 40+30x (x = push level).
+
+## Pacing
+
+**Unlock ladder** (level → unlocks; ← marks `unlock_requires`; starting
+set at −1: drone, frigate, railgun_s, salvager, building_block recipe):
+
+| level | ships | modules | recipes |
+|---|---|---|---|
+| 0 | destroyer | repair_tool, armor_plates | |
+| 1 | | maneuvering_thrusters, sensor_booster | shortcut_steel_plate |
+| 2 | cruiser | railgun_m, afterburner | shortcut_control_chip, shortcut_hardened_steel |
+| 3 | | weapon_stabilizer | |
+| 4 | battlecruiser ← cruiser | weapon_primer, weapon_upgrade | |
+| 5 | | drone_bay | |
+| 6 | battleship ← battlecruiser | railgun_l ← railgun_m | |
+| 8 | dreadnought ← battleship | | |
+| 9 | carrier ← battleship | drone_hangar | |
+
+Level 0's pool has exactly three entries (a full first dialog). Level 2
+is the quartz gate: cruiser and railgun_m are the first schematics whose
+chains reach quartz; the shortcut outputs only become implicitly
+unlocked alongside them, so shortcuts cannot drop early.
+
+**Threat rate** `2*x + 0.15*x*x` (x = boss cycle counter), against the
+factory-size curve with ~half the player's output assumed military:
+
+| cycle x | rate (threat/s) | player military (≈ curve/2) |
+|---|---|---|
+| 2 | 4.6 | ~12 |
+| 6 | 17.4 | ~30 |
+| 15 | 63.8 | ~60 |
+| 20 | 100 | ~75 |
+| 24 | 134 | — |
+
+**Economy:** `starting_building_blocks = 200`; expansion cost formula
+`300 + 50*x + 10*x*x` (x = expansions already purchased: ~1 affordable
+per cycle mid-game at ~1/3 of block income, stretching to 2–3 cycles
+late — quadratic so costs outrun the roughly linear block income
+gradually, never with a hard wall); `artifact_win_count = 5` with
+`artifact_chance_formula = 0.05*x`. Building costs: belt 2, splitter 3,
+tunnels 5, miner 15, smelter 20, assembler 35, reprocessing plant 40,
+salvage bay 25, shipyard 60 — averaging ≈18 blocks per placed building
+(belts included), which meets the 4-minute doubling target at block
+threat 4.

docs/balancing/history.md

@@ -0,0 +1,96 @@
+# Balancing History
+
+Chronological record of the balancing work: what was decided, what was
+found, what changed. Current values live in `derived.md`; this file
+explains how they got there.
+
+## 2026-07-02/03 — rules and structural decisions
+
+- Rules document written (now `rules.md`): ratio curve, shortcut
+  recipes, refactorability, cost archetypes, threat model, growth curve.
+- Scrap derived from threat (`scrap_per_threat`), replacing authored
+  per-ship scrap drops; scrap threat became the constant
+  `1/scrap_per_threat`, removing the old min-scrap_drop derivation and
+  its circularity.
+- Duplicate schematic drops removed (no level-ups); ship/module levels
+  removed entirely — all time scaling lives in the threat rate, push
+  scaling stays on stations. Mk2 upgrade recipes noted as the future
+  per-item progression.
+- Growth-curve rules added: escalating expansion costs, designed
+  doubling time, growth limited by economy not waiting; resource
+  deposits designed (deposit-gated mid resource in expansion territory).
+- Production tree v2 decided: iron/copper everywhere (M-type asteroid),
+  quartz in geodes (mid), voidsteel battle-forged from scrap (late);
+  titanium dropped; lasers renamed to railguns, lasers reserved as a
+  future weapon type.
+
+## 2026-07-03 — targets, tree, numbers
+
+- Balancing targets fixed: ≤2 h run, phases 1–5/6–14/15+, factory curve
+  25/60/120/150, threat ladder, 25-ship swarm, block roots.
+- Tree structure drafted and numbers computed (recursive threat
+  calculator); ratio curve realized; fitted ships within 96–124% of the
+  strawman ladder (small end hot from fixed chain overhead — ladder
+  later adopted the achieved values).
+- **Rule bugs found by the numbers work:** the scrap→ingot smelter
+  recipe would inflate basic materials via the max rule (fixed:
+  scrap-consuming recipes are threat fallback only); recipe output
+  amounts were ignored (fixed: per-unit division); items downstream of
+  reprocessing-only items never resolved (fixed: fixpoint resolution);
+  a shortcut recipe resolving earlier than the base path silently
+  underpriced items (fixed: commit only when all eligible recipes are
+  computable). All four fixed in `ThreatCostCalculator` with tests, and
+  implemented in `tools/threat_report.py`.
+- v2 tree written into the configs; `default_modules` loadouts
+  geometry-validated (the numbers-pass loadouts for battlecruiser and
+  dreadnought were geometrically impossible — L-modifiers don't fit
+  beside full gun complements; corrected loadouts landed closer to the
+  ladder).
+
+## 2026-07-04 — combat stats, arena rounds 1–5
+
+Initial stats derived from the anchors (weapon DPS ≈0.6/threat flat,
+hull 15 HP/threat, armor 20/threat, repair 2 HP/s/threat, station
+range 200).
+
+- **Round 1:** concentrated fleets won all equal-threat cross-tier
+  matchups flawlessly; glass beat armored; repair escort flawless; two
+  stations shrugged off a 3× swarm. Changes: concentration tax on m/l
+  gun damage (railgun_m 17→14, railgun_l 70→52), armor 640→1000,
+  repair 25→12, station range 200→120. (Team-1 "bias" in mirrors later
+  shown to be noise.)
+- **Round 2 (EHP-margin logging added):** battleship +33% while
+  dreadnought −37% (stabilizer range + opposing armor); glass still
+  +11%. Changes: stabilizer range ×1.5→×1.3; per-hull trims introduced
+  (BC 2700→2500, BS 7500→7000, DN/CV 15500→19000).
+- **Round 3 (narrow lanes — geometry fixed into the fixture):**
+  DN closed to −11%, BS +22%, swarm flipped to +14% over cruisers,
+  glass +12% third time. Changes: armor 1000→1200, BC 2500→2000,
+  BS 7000→6300, DN/CV 19000→22500.
+- **Round 4:** glass-vs-armored resolved (+3% armored); noise floor
+  established (~±10%/run: BS ignored a −10% EHP cut; repair drifted
+  14→24% untouched). Convergence policy adopted: two-round signals only,
+  ±20% converged. Changes: BC 2000→2200, DN/CV 22500→24000; BS +23%
+  accepted as doctrine texture (mechanical range edge vs. pure small
+  fleets).
+- **Round 5 (durations logged; end-condition bug fixed upstream):**
+  TTK anchor validated (mirrors 23/71/95 s; DN-vs-swarm 214 s outlier
+  accepted); dreadnought +3%, everything else inside band. Final
+  changes: BC 2200→2400, repair 12→9 (persistent +24% escort margin).
+  **Combat pass declared converged.**
+
+## 2026-07-05/06 — pacing pass
+
+- Unlock ladder set (starting set drone/frigate/railgun_s/salvager;
+  quartz gate at level 2; capitals at 8–9 with `unlock_requires`
+  chains); threat rate `2*x + 0.15*x*x`; starting blocks 1000→200;
+  expansion 400 flat pending the cost formula; artifacts 3→5.
+- **Bug found:** the building_block recipe was silently locked at game
+  start (building blocks appear in no schematic's materials, so implicit
+  unlocking could never reach the recipe) — fixed with an explicit
+  `unlock_at_station_level = -1`.
+- Expansion cost formula implemented and set (`300 + 50*x + 10*x*x`):
+  quadratic, so costs outrun the roughly linear block income gradually
+  — ~1 expansion per cycle mid-game, 2–3 cycles apart late.
+- **First full balancing round complete.** Next: full-game playtests
+  against the run-shape targets.

docs/balancing/process.md

@@ -0,0 +1,98 @@
+# Balancing Process
+
+How balancing is done in this project: the pass order, the tuning
+discipline, and the tools. Refer to this when starting the next
+balancing round.
+
+## The pass order
+
+Each pass depends on the ones before it; a change in an earlier pass
+invalidates the later ones (but not vice versa). Redo from the earliest
+pass whose inputs changed.
+
+1. **Targets** (`targets.md`) — choose the root numbers: run shape,
+   factory curve, threat-cost ladder, fleet size, block roots, combat
+   anchors, pacing anchors. These are design decisions, not
+   measurements. Everything else is derived from them.
+2. **Tree structure** (`../content_design.md`) — items, chains,
+   what-consumes-what, per the production tree rules (one input per
+   phase transition, generic parts, archetypes, refactorability).
+   Structure only, no quantities.
+3. **Numbers** (`derived.md`, recipes/materials in the configs) —
+   quantities and durations so every fitted ship sums to its ladder
+   value, the ratio curve is realized, and the belt/buffer guardrails
+   hold. Verified computationally by `tools/threat_report.py`.
+4. **Calculator/tooling parity** — the game's `ThreatCostCalculator`
+   and `tools/threat_report.py` must produce identical values; the
+   Python tool is the design reference. Any semantic change to
+   REQ-THREAT-* needs both updated plus tests.
+5. **Combat stats** (arena-driven) — derive stats from the combat
+   anchors, then iterate against the arena suite
+   (`bin/balancing/data/balancing.toml`) until equal-threat matchups are
+   near-draws. Threat costs are stat-independent, so arena ship counts
+   stay valid across stat changes.
+6. **Pacing** — unlock ladder, `unlock_requires` edges, threat rate,
+   block/artifact/expansion values, per the pacing anchors.
+
+Then: **full-game playtests**, which are the only check for the pacing
+pass and feed back into targets.
+
+## Tuning discipline (learned in arena rounds 1–5)
+
+- **Change anchors, not symptoms.** When a class of results is off,
+  adjust the anchor that explains all of them (e.g. the concentration
+  tax) rather than individual stats.
+- **Fewest knobs per round.** Attribution dies when many knobs move at
+  once. Prefer one anchor change plus its mechanical compensations.
+- **Shared vs. local knobs.** Guns and module stats are shared across
+  many hulls — changing them moves many matchups. Per-hull HP moves
+  exactly one matchup; it is the designated per-ship trim knob on top of
+  the HP-per-threat prior.
+- **Mind the ride-alongs.** A module buff lands on every default loadout
+  containing it (e.g. an armor buff strengthens the destroyer swarm that
+  opposes the dreadnought). Compute the net effect per matchup before
+  choosing step sizes.
+- **Two-round signal policy.** Single arena runs re-roll by ~±10% EHP
+  margin; a margin inside ±20% counts as converged for v1. Only act on
+  signals that persist across two rounds.
+- **Arena geometry is part of the fixture.** Lane width/height changes
+  the results (full engagement vs. fleets slipping past); margins are
+  only comparable within the same geometry.
+- **Accept mechanical texture.** Not every deviation is a bug: a margin
+  that survives a stat change is mechanical (usually range/kiting under
+  the orbit AI) and may be desirable doctrine texture. Document the
+  acceptance in `targets.md` instead of chasing it.
+- **Range is the strongest stat** under the orbit AI — free approach
+  fire. Price range modifiers conservatively; station dominance is
+  controlled via range, not HP.
+
+## Tools
+
+- `tools/threat_report.py` — item threats, module contributions,
+  hull/fitted ship threats, producer:consumer ratios, belt feasibility;
+  reads the real configs. The design reference for threat semantics.
+- `tools/verify_recipes.py` — recipe tree closure, visuals coverage,
+  orphans, reprocessing-only items.
+- `tools/verify_layouts.py` — module footprint gating matrix per hull.
+- **Balancing tool** (`balancing` target) — parallel arena simulation of
+  `bin/balancing/data/balancing.toml`; logs winner, surviving counts,
+  team EHP %, and fight duration per arena. The suite covers: class
+  mirrors (expect near-mutual annihilation, symmetric winners),
+  equal-threat cross-tier matchups (expect near-draws — power-per-threat
+  made empirical), a 2:1 decisiveness check, doctrine matchups
+  (armored-vs-glass, repair-escort), and station assault.
+
+## Checklist for the next balancing round
+
+1. Pull; run `verify_recipes.py`, `verify_layouts.py`,
+   `threat_report.py`; compare against the tables in `derived.md`.
+2. If recipes/materials changed: re-check fitted threats vs. the ladder
+   in `targets.md`; update arena suite ship counts if fitted values
+   moved.
+3. Run the arena suite; read EHP margins and durations against the
+   expectations noted in `balancing.toml` and the anchors.
+4. Apply changes per the tuning discipline (two-round signals only);
+   record the round and its knob changes in `history.md`.
+5. Update `derived.md` where values moved; if an anchor moved, update
+   `targets.md` and state why.
+6. Commit and push (the review workflow reads the remote).

docs/balancing/rules.md

@@ -0,0 +1,333 @@
+# Balancing & Progression Rules
+
+Rules and principles that govern the production tree, progression pacing,
+and balancing. This document contains **rules only** — the chosen base
+numbers live in `targets.md`, everything derived from them in
+`derived.md`, and the concrete content in the config files and
+`../content_design.md`. All of those must follow the rules stated here.
+
+## Player-experience goals
+
+What each phase of a run should feel like:
+
+- **Early:** learning belts and ratios with forgiving chains. The building
+  block economy is the main constraint; the player bootstraps a
+  self-sustaining factory from the starting stock.
+- **Mid:** deeper chains, the first real ratio puzzles, and the first
+  meaningful drop decisions (which schematic, when to push).
+- **Late:** combat feeds the factory — capital production requires salvage.
+  Progress means extending and refactoring the existing factory, not
+  rebuilding it. Strange ratios are deliberate optimization puzzles.
+
+Overarching: an experienced player gains efficiency through **knowledge** —
+layout foresight, understanding chains, exploiting shortcut recipes — never
+through hidden mechanics. An inexperienced setup should not cost much more
+than an experienced one; experience pays off in how easily the factory
+adapts later (see Refactorability).
+
+## Resource phases
+
+- A run has exactly **four base inputs**:
+  1. Two mined resources available from the start, minable on **every**
+     asteroid tile.
+  2. A third mined resource unlocked mid-game, minable **only on deposit
+     patches** found in expansion territory (see Resource deposits).
+  3. A fourth input unlocked late-game, obtainable **only** from
+     reprocessing salvaged scrap.
+- The fourth input is the core loop hook: capital ship production requires
+  fighting (salvaging and reprocessing), not just mining.
+- Every gating has a fictional reason (concrete fiction in
+  `../content_design.md`): the asteroid is a metal-rich body, so its bulk
+  rock is minable anywhere; the mid resource sits in rare pockets; the
+  late input is battle-forged — created only in the violence of ship
+  destruction, which is why any wreck (including the player's own)
+  yields it and no foundry can make it.
+- The mid resource is **dual-gated**: schematics (knowledge, via drops)
+  and territory (deposits, via expansions). Tuning must guarantee the
+  deposit-bearing expansion is comfortably affordable by the time the
+  first mid-tier schematics drop, or those drops are dead picks.
+- There is no direct "resource unlock" mechanism. Miner recipes unlock
+  **implicitly** (REQ-LOCK-IMPLICIT) when some unlocked schematic's material
+  chain reaches that resource. Resource pacing is therefore controlled
+  through the `unlock_at_station_level` values of ships, modules, and
+  assembler recipe schematics — and the content must guarantee that the
+  chains actually connect (a mid-game schematic must require an item whose
+  chain reaches the mid resource, or it never unlocks).
+
+### Resource deposits
+
+- **Rule: freedom first, geography later.** The starting resources are
+  minable everywhere, so the player has full layout freedom while
+  learning. Later mined resources are bound to deposit patches — fixed
+  geography as a layout puzzle, introduced once the player is competent.
+- **Rule: deposits exist only in expansion territory.** Expansions buy
+  space *and* access to resource tiers — the second leg of the growth
+  curve (see Building block economy).
+- **Rule: patch area is the throughput cap.** Deposits never deplete but
+  are finite in area; the number of deposit tiles caps how many miners
+  the chain supports. Buying deeper expansions raises the throughput
+  ceiling of high-tier chains.
+- **Rule: no empty expansions.** Deposit content per expansion is
+  deterministic and config-defined; only the placement within the new
+  columns is randomized. Buying an expansion never rolls "nothing".
+- **Rule: mining is binary.** A miner whose footprint overlaps at least
+  one deposit tile of a resource can select that resource's recipe; no
+  partial-coverage rate scaling.
+- Deposits arrive at the periphery (expansions add columns on the left),
+  so each new chain starts in fresh space — supporting the
+  refactorability property — and high-tier chains have the longest belt
+  runs to the shipyards, escalating the logistics puzzle with tier.
+
+## Production tree rules
+
+### Structure
+
+- **Each phase transition adds exactly one new base input chain.** A base
+  input is a bottom-level resource entering the factory from outside — a
+  mined resource or the scrap-only input. The early game starts with two
+  ores as the baseline; the transition to mid adds one (the deposit-bound
+  mid resource), the transition to late adds one (the scrap-only input).
+  No transition ever introduces more than one unfamiliar bottom-level
+  chain, so the factory grows in one direction at a time.
+- **Intermediates are generic shared parts.** Keep the item count low —
+  modules and hulls of a tier draw from a shared pool of that tier's and
+  lower tiers' intermediates rather than each having bespoke inputs.
+- **Thematic naming over thematic items.** Inputs should be plausible for
+  what the recipe produces (crystals for lasers, heat sinks for bigger
+  lasers). Achieve this through naming and chain membership, not by adding
+  item types: rename a generic part, don't add a parallel one.
+
+### Ratios
+
+- **Ratio "niceness" degrades with tier.** The producer:consumer ratios
+  needed for 100% throughput follow a curve:
+  - Tier 1 (ore → basic material): trivially nice (e.g. 1:1 or 1:2
+    miner:smelter).
+  - Tier 2: slightly complex but still clean (e.g. 2:3).
+  - Higher tiers: increasingly strange ratios, as deliberate optimization
+    puzzles.
+- Exceptions in both directions are allowed when there is a reason — a
+  clean late chain as a breather, an odd early chain as a teaser — but the
+  curve is the default.
+
+### Shortcut recipes
+
+- Some strange chains get a **shortcut recipe**: an explicitly unlockable
+  assembler recipe schematic (`unlock_at_station_level ≥ 0`, drop-only per
+  REQ-LOCK-EXPLICIT) that skips a step (e.g. t1 → t3 directly) and yields
+  nice ratios for a chain whose base path is strange.
+- **Not every strange chain gets a shortcut.** Some strangeness is
+  permanent; the absence of a fix is a valid design choice.
+- **Shortcuts drop only for known chains.** A shortcut recipe enters the
+  drop pool only when both its input items and its output item are
+  already unlocked (in addition to the station level check). The player
+  is never offered a shortcut for a chain they have not built yet. The
+  output-item half of this check already exists in
+  REQ-DEF-SCHEMATIC-DROP; the input half is an open action item (see
+  `README.md`).
+- **Shortcuts are pure rewards, never balance factors.** An item's threat
+  value is the *maximum* across its producing recipes (REQ-THREAT-ITEM), so
+  unlocking a cheaper recipe does not lower the item's threat accounting —
+  the player gains real factory efficiency without their ships being
+  valued cheaper and without enemy wave budgets shifting. Consequently:
+  **balance every chain around its base (expensive) path**; the shortcut's
+  savings define the size of the reward.
+
+### Refactorability
+
+- **Rule (the property):** unlocking the next tier or size of a thing must
+  be a *local edit* of the existing production line — adding assemblers
+  and belts, or replacing a machine or two in place — never a rebuild of
+  the line.
+- **What this buys the player:** foresight pays off in space, not blocks.
+  An experienced player leaves a little slack in the middle of a line,
+  knowing the next size or tier upgrade means tearing out one assembler
+  and a few belts there and inserting the new step — plus maybe swapping
+  a recipe or two elsewhere — while the rest of the line keeps running
+  untouched.
+- **Default technique:** the bigger version introduces one new intermediate
+  that is produced from a subset of the smaller version's inputs (possibly
+  plus one additional low-tier material), and otherwise reuses the smaller
+  version's inputs. Existing lines keep running and feed the new
+  intermediate's assemblers.
+- The property is the rule; the technique is only the default. It may be
+  broken where it fights thematic plausibility, as long as the property
+  still holds.
+
+## Cost archetypes
+
+Every item has two cost knobs: **material quantity** and **cycle time**.
+Both feed the threat value identically (threat = recursive
+production-seconds, REQ-MOD-THREAT), so the split between them does not
+change what an item is *worth* — it changes what kind of **factory
+pressure** it creates:
+
+- **Material-heavy, fast** (e.g. armor plates): simple items; stress belt
+  throughput, splitter logistics, and miner/smelter counts.
+- **Time-heavy, lean** (e.g. shield modules): technically complex items;
+  few inputs — possibly higher-tier ones — but long cycles; stress
+  assembler counts and parallelization.
+
+**Rule:** each module family commits to a clear archetype, so factories
+supporting different fleet doctrines feel structurally different to build.
+
+## Threat model (balancing backbone)
+
+- Threat cost = total recursive production-seconds (REQ-MOD-THREAT). One
+  factory-second equals one threat; player output and enemy wave budgets
+  are denominated in the same currency.
+- **Rule: combat power per threat is roughly constant** across all ships,
+  modules, and tiers. Higher tiers are better per *ship* and per *module
+  slot*, not per invested factory-second — their advantage is
+  concentration (fewer, bigger things; slot geometry per
+  `../content_design.md`) and qualitative capabilities, not a better
+  exchange rate. Deviations from this rule are deliberate and documented.
+- **Difficulty race:** the enemy threat rate (`threat_rate_formula`) is
+  tuned against the factory output (threat/s) achievable by a competent
+  player — slightly below it early, crossing above it eventually. The game
+  is endless; enemy scaling must ultimately outpace any factory, and
+  player skill shifts *when*, not *whether*.
+- **All time scaling lives in the threat rate** — waves get bigger, ships
+  of a given schematic never get individually stronger. There is no ship
+  level dimension: stat formulas are plain values, and per-ship level
+  scaling does not exist. Push scaling on enemy defence stations is the
+  separate, player-triggered difficulty axis and keeps its level formulas.
+
+## Unlock & drop pacing
+
+- **Starting set rule:** the schematics unlocked at game start
+  (`unlock_at_station_level = -1`) must be exactly enough to reach the
+  first push unaided — a functioning block loop, small hulls, a basic
+  weapon, and the salvage loop. Nothing more.
+- The `unlock_at_station_level` ladder mirrors the resource phases:
+  mid-tier hulls/modules/recipes at low station levels, capital content at
+  higher levels. A schematic must not become available before the chains
+  its materials need can be unlocked alongside it.
+- **Schematics can require other schematics.** Beyond the station-level
+  gate, a schematic (ship, module, or assembler recipe) may list
+  prerequisite schematics (`unlock_requires`, REQ-LOCK-PREREQ) that must
+  already be unlocked before it enters the drop pool — e.g. the medium
+  gun requires the small gun; a future Mk2 requires its base version.
+  Station level gates the earliest *when*; prerequisites gate the
+  *order*, keeping drop offers coherent with what the player already
+  owns.
+- **No duplicate drops.** Ship and module schematics leave the drop pool
+  once owned, exactly as assembler recipe schematics already do. There are
+  no schematic level-ups; player power grows through unlock breadth and
+  factory scale only, which keeps power-per-threat exact on both sides.
+  The pool therefore shrinks over a run and late pushes increasingly offer
+  artifacts — intended: the late game is a race for the win condition.
+  Per-item progression may return later as Mk2 upgrade recipes (see Future
+  work), never as free level-ups.
+- **Artifacts trade power for progress.** Artifact options compete with
+  schematic picks in the same choice dialog; the artifact chance must be
+  tuned so that taking one is a real decision (giving up an unlock), not
+  automatic in either direction.
+
+## Scrap & reprocessing economy
+
+- Scrap is the bridge from combat back into the factory, with two sinks:
+  **smelting** (same basic materials as ore — the safe, boring option) and
+  **reprocessing** (probabilistic higher intermediates, including the
+  late-game input — the gamble that eventually becomes mandatory).
+- The reprocessing output pool renormalizes over implicitly unlocked items
+  (REQ-LOCK-REPROCESSING-POOL), so its output quality improves
+  automatically as the run progresses. **Rule:** weights are authored for
+  the *fully unlocked* pool state; early-game behavior falls out of
+  renormalization for free and needs no separate staging.
+- **Rule: ship scrap drops are derived, never authored.** A destroyed ship
+  drops `threat cost × scrap_per_threat` (a `world.toml` key), with the
+  threat cost computed from its actual hull plus installed modules
+  (REQ-MOD-THREAT) — a kitted-out ship drops more scrap than a bare hull
+  automatically. `ships.toml` carries no scrap value. Defence stations are
+  the exception: they keep authored `scrap_drop_formula`s, because pushing
+  rewards are tuned independently of ship production costs.
+- Consequence: the threat value of scrap is the constant
+  `1 / scrap_per_threat` (REQ-THREAT-SCRAP). The former min-`scrap_drop`
+  schematic derivation and its potential circularity are gone.
+- **Rule:** the late-game input's income rate meaningfully gates capital
+  production — unlocking a capital hull must not mean spamming it; the
+  input trickles in slowly enough that every capital ship is a noticeable
+  investment. The tuning target is relative, not absolute: assume a
+  reference player who destroys and salvages roughly the threat the game
+  spawns ("fighting at parity"), and tune `scrap_per_threat`, the
+  reprocessing weights, and capital material costs so that this player
+  affords roughly N capital ships per boss cycle. An absolute income rate
+  would be meaningless (income depends entirely on how much the player
+  fights) and would not self-scale; per boss cycle, the target tracks the
+  threat rate as it steps up.
+
+## Building block economy
+
+- Building blocks are the only global currency and the early game's
+  central constraint. The early game is a bootstrap problem: convert the
+  starting stock into a self-sustaining block loop before the first waves
+  bite.
+- **Rule:** the starting stock suffices for a minimal block loop plus the
+  first shipyard — with a little slack for beginner mistakes, but not
+  enough to skip the loop entirely.
+- **Rule: the growth curve lives here.** A saturated building produces
+  exactly 1 threat/s, so the player's output curve *is* their
+  building-count curve — shaping growth over a run means shaping the
+  block and space economy, there is nowhere else it can live. Intended
+  shape: exponential bootstrap (block-limited) → ramp
+  (expansion-limited) → asymptotic squeeze as expansion costs outrun
+  income, racing the enemy threat rate throughout.
+- **Rule: escalating expansion costs.** Expansion cost is a formula of
+  the number of expansions already purchased, rising steeply enough that
+  expansions eventually outrun any block income. The starting asteroid
+  is deliberately small — filled within the first boss cycle or two, so
+  the early exponential burst is a satisfying ramp, not a balance hole —
+  and from then on the output curve is the expansion curve. Blocks keep
+  a meaningful sink for the entire run, and "grow vs. army" stays a live
+  decision at every moment.
+- **Rule: designed doubling time.** Block production is a positive
+  feedback loop (blocks buy assemblers, assemblers make blocks); its
+  time constant is a designed quantity, never an accident of quantity
+  choice. The block chain's depth and the per-building costs are tuned
+  against a stated target of the form: "a factory spending X% of its
+  capacity on blocks doubles in ~T minutes."
+- **Rule: growth is limited by economy, never by waiting.** Construction
+  times stay short; the serial build queue must not be used as a growth
+  brake. Waiting for placed buildings to become operational — especially
+  at the start of a run — is frustration, not gameplay. All growth
+  limiting comes from block income and expansion pricing.
+- Note: block income has a structural ceiling — blocks enter the stock
+  through the HQ's single belt port, so income is capped at belt
+  throughput regardless of assembler count. Per-building costs should be
+  high enough that this cap can bind late-game (see the condensed-block
+  idea under Future work).
+
+## Numeric guardrails
+
+Constraints that every recipe must respect, independent of tuning:
+
+- **Belt throughput:** belt speed and per-tile capacity cap how fast a
+  single belt can feed an input. A recipe whose per-cycle inputs cannot be
+  sustained by one belt per input at 100% duty cycle is a *deliberate*
+  design (forcing parallel belts/splitters as part of a high-tier puzzle)
+  — never an accident of quantity choice.
+- **Buffer burstiness:** input buffers hold 2× the per-cycle amount
+  (REQ-MAT-INPUT-BUFFER), so large per-cycle quantities create bursty belt
+  demand. Low tiers prefer small quantities with short cycles; big-batch
+  recipes are reserved for high tiers where burstiness is part of the
+  puzzle.
+- **Cycle times scale with tier** monotonically — a higher-tier item never
+  has a shorter total chain time than a lower-tier item of the same role.
+
+## Future work
+
+- **Condensed building blocks** — a drop-unlockable shortcut-style
+  recipe that packs several blocks' worth of value into one belt item,
+  relieving the HQ intake ceiling (see Building block economy) as a
+  late-game reward. The ceiling is the puzzle, the drop is the fix —
+  same philosophy as shortcut recipes.
+- **Mk2 upgrade recipes** — the deferred design for per-item progression,
+  to revisit once the config has stabilized. A duplicate-style drop
+  unlocks a distinct `*_mk2` item whose recipe consumes the Mk1 item plus
+  higher-tier parts. This preserves power-per-threat (the extra power is
+  paid in real production-seconds, since threat is recursive), satisfies
+  the refactorability rule (the Mk1 line keeps running and feeds one new
+  assembler), and keeps balancing one-dimensional (no level variable
+  anywhere). Enemy-side progression happens via `default_modules`
+  variants per era instead of a level formula.

docs/balancing/targets.md

@@ -0,0 +1,104 @@
+# Balancing Targets (base numbers)
+
+The root numbers of the balancing. Everything in `derived.md` is tuned to
+hit these; when rebalancing, **change these first and re-derive — never
+patch derived values directly**. The rules these numbers follow live in
+`rules.md`.
+
+All time targets are in **game time**. The player can pause and
+accelerate, so real session length differs; playtests measure both. The
+time unit is the boss cycle (`world.toml boss_countdown_seconds`, 300 s).
+Destroying a station set advances the boss countdown by
+`boss_advance_seconds` (60 s), so cycles run shorter than nominal when
+pushing actively — targets deliberately ignore that.
+
+## Run shape
+
+1. **Run length** — a winning run takes up to 2 hours of game time: win
+   around boss cycle 20–24. Losing runs end earlier.
+2. **Phase boundaries** — early = cycles 1–5 (iron/copper, small hulls),
+   mid = cycles 6–14 (quartz, medium hulls), late = cycles 15+
+   (voidsteel, capitals). Push cadence: first station set around cycle
+   2–3, roughly one per cycle from mid onward — so the destroyed set's
+   level ℓ is reached around cycle ℓ+2.
+3. **Factory size curve** — producing buildings over time; when
+   saturated, output threat/s equals this count, so this curve IS the
+   player power curve: ~25 when the starting asteroid is full (end of
+   cycle 2), ~60 at the start of mid (cycle 6), ~120 at the start of
+   late (cycle 15), ~150 near the win. `threat_rate_formula` must remain
+   a fraction of this curve; buildings plus belts must physically fit
+   the asteroid plus affordable expansions.
+4. **Threat-cost ladder** — total production-seconds per *fitted* hull
+   (including the typical/default module loadout): drone 10.5,
+   frigate 47, destroyer 99, cruiser 233.5, battlecruiser 354.5,
+   battleship 722.5, dreadnought 1491.5, carrier 1436.5. Every
+   production chain must sum to its ladder value. (The original strawman
+   was 10/40/80/200/350/700/1500; the small end runs ~10–20% hot because
+   fixed chain overhead dominates small hulls — accepted, and the
+   achieved values adopted as the ladder. The ~×2-per-class curve shape
+   is the invariant.)
+5. **Fleet size** — swarm-leaning: ~25 player combat ships as the
+   standing mid-game fleet. Standing fleet = build cadence (4) × average
+   ship lifetime, so this target drives time-to-kill and therefore all
+   combat stat magnitudes.
+6. **Block economy roots** — bootstrap complete (starting asteroid full)
+   by the end of cycle 2; a factory spending ~30% of its capacity on
+   blocks doubles in ~4 minutes early game; one expansion affordable per
+   cycle at ~1/3 of block income mid-game, decelerating to one per 2–3
+   cycles late as escalating costs outrun income.
+
+## Combat anchors
+
+All combat stats derive from these; per-hull HP additionally carries
+empirical trims from arena rounds (values in `derived.md`).
+
+- **Weapon DPS per threat pays a concentration tax that grows with gun
+  size**: small ≈ 0.62, medium ≈ 0.48, large ≈ 0.41 DPS per threat of
+  weapon contribution, compensated by the range ladder 50/70/100 m.
+  Rationale: concentration itself (focus fire, no DPS loss to attrition,
+  range) is worth paying for — with a flat curve, concentrated fleets
+  win equal-threat fights outright (arena round 1).
+- **Hull HP = 15 per threat of hull contribution** as the prior; per-hull
+  HP is the empirical trim knob (guns are shared across hulls, hull HP
+  moves exactly one matchup). The arena consistently prices capitals as
+  *tanks with taxed guns* — capital hulls sit well above the prior.
+- **Armor HP ≈ 37 per threat** — a strong premium over hull HP because
+  armor is pure HP with no capability, and fights snowball: killing
+  removes enemy DPS, surviving merely delays — HP must be cheaper than
+  DPS.
+- **Repair ≈ 0.7 HP/s per threat** — in-combat sustain effectively
+  removes enemy DPS and must be priced like DPS, not like HP.
+- **TTK / fight duration**: parity fights in the 30–60 s band at
+  mid-game scale; capital mirrors ~90 s deliberately; the extreme
+  tank-vs-chip-damage matchup (dreadnought vs destroyer swarm, ~3.5 min)
+  is an accepted outlier.
+- **Mobility is monotone in size** — the smallest hulls are the fastest
+  and nimblest. Sensor ranges (150→350 m) always exceed weapon ranges.
+- **Weapon modifiers are capital economy**: a ×1.2 damage modifier at
+  ~22.5 threat beats adding a gun once a ship carries more than ~68
+  threat of weapons — modifiers pay off on gun-heavy big hulls, waste on
+  small ones. Range modifiers are the strongest and are priced/kept
+  small (×1.3): range is the dominant stat under the orbit AI (free
+  approach fire).
+- **Stations**: a fresh player station holds one early parity wave
+  unaided; the enemy station at level 0 matches the player station
+  exactly and scales per push level. Station range is the dominance
+  lever, not HP (at 4× a small gun's range, two stations annihilated a
+  3× threat swarm through approach fire alone).
+- **Accepted imbalances**: the carrier loses its equal-threat fights
+  until the drone-launching capability exists (the hangar is dead
+  threat) — fix by implementing drones, not stats. A pure smallest-ship
+  fleet modestly loses (~15–25%) to a range-fitted capital — desirable
+  doctrine texture; the fair anti-capital answer is the mixed fleet.
+
+## Pacing anchors
+
+- **Starting set** is the rule-minimum: drone, frigate, small gun,
+  salvager (plus the explicitly unlocked building-block recipe).
+- **Threat rate shape**: below the player's achievable military output
+  (≈ half the factory curve) early, crossing at the late boundary
+  (~cycle 15), overwhelming by ~cycle 24.
+- **Winning = five real decisions**: `artifact_win_count` is set so that
+  across a winning run's ~15–18 pushes (~7 cumulative artifact offers at
+  the current chance formula), the player must choose the artifact over
+  a schematic about five times.