Round 5 (with fight durations): TTK anchor validated - mirrors run
22.7/71.2/94.6 s (drone/cruiser/battleship), most fights within
22-72 s, destroyers-vs-dreadnought the accepted 214 s outlier. All
matchups converged: mirrors 5-10%, swarm vs cruisers +15%, frigates
vs battleship +13% (dropped on its own, confirming the noise floor),
dreadnought +3%, glass vs armored +10%, two-to-one 78%, stations 44%.
Two two-round signals actioned per the convergence policy:
battlecruiser hull 2200 -> 2400 (mixed fleet won +20%/+18% twice) and
repair_amount 12 -> 9 (~0.7 HP/s per threat; escorted team held +24%
twice with all frigates surviving). Everything else frozen; further
refinement moves to real-game playtests. Removed the arena
end-condition action item (fixed upstream).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DyCu8vwChKMbLJQ3xosYEN
Round 3 (narrow lanes; arena geometry is part of the fixture from
here): dreadnought closed to -11%, battleship improved to +22%, the
drone swarm flipped to +14% over cruisers under full engagement, glass
beat armored for the third consecutive round, mirrors/repair/
two-to-one healthy, stations cracked at 51% by the 3x swarm under the
new geometry (watch, not touched).
Round-4 knobs: armor_plates 1000 -> 1200 (~37 HP/threat; persistent
glass signal, also lifts armor-carrying cruiser/BC/destroyer/DN
loadouts), battlecruiser hull 2500 -> 2000 (net -300 EHP after its
armor gain), battleship 7000 -> 6300 (loadout has no armor, clean
-10%), dreadnought/carrier 19000 -> 22500 (opposing destroyers gain
armor too, so the larger step). Anchors doc notes the emerging shape:
the arena prices capitals as tanks with taxed guns.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DyCu8vwChKMbLJQ3xosYEN
Round 2 (with the new EHP-margin logging): mirrors healthy and the
round-1 side bias was noise; drones-vs-cruisers (10%) and
mixed-vs-battlecruisers (14%) near parity; battleship over at +33%
while dreadnought under at -37%; glass-vs-armored narrowed to 11%
(watch); repair escort fair at 11%; two-to-one decisive; the 3x swarm
now cracks the station set at 13%.
Round-3 knobs: weapon_stabilizer range multiplier 1.5 -> 1.3 (range is
the strongest stat under the orbit AI, and the stabilized battleship
was the one loadout still overperforming), plus per-hull HP trims with
the 15 HP/threat anchor demoted to a prior and per-hull HP documented
as the empirical trim knob: battlecruiser 2700 -> 2500, battleship
7500 -> 7000, dreadnought/carrier 15500 -> 19000.
Action items: mirror-bias and EHP-logging items resolved (the latter
implemented upstream); new item to verify the arena end condition seen
idling in round 1 with all ships dead.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DyCu8vwChKMbLJQ3xosYEN
Round 1 findings: concentrated fleets won all four equal-threat
cross-tier matchups flawlessly; glass destroyers beat armored; repair
escort beat raw numbers flawlessly; two stations annihilated a 3x
threat swarm through 200 m approach fire; team 1 won all three mirror
matches (possible sim bias, recorded as action item 6, plus action
item 7 for HP-margin logging).
Knob changes, deliberately few for clean attribution: concentration
tax on m/l gun DPS (railgun_m damage 17 -> 14, railgun_l 70 -> 52),
armor_plates 640 -> 1000 HP (~31 HP/threat: killing removes enemy DPS
while surviving merely delays, so HP must be cheaper than DPS), repair
25 -> 12 HP per cycle (~1 HP/s per threat: in-combat sustain removes
enemy DPS and must be priced like DPS), station range 200 -> 120. Gun
ranges untouched; if concentration still wins flawlessly in round 2,
the range ladder is the next suspect. Anchors and findings documented
in content_design.md; fitted threat values are unaffected by stat
changes, so the arena suite counts stay valid.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DyCu8vwChKMbLJQ3xosYEN
Derive all combat stats from documented anchors (content_design.md,
'Combat stat anchors'): weapon DPS ~0.6 per threat with a mild size
discount traded for range (railgun m/l damage 17/70), hull HP at 15
per hull threat (drone 60 up to capitals 15500), armor at a 20/threat
premium (armor_plates 640 HP), repair at ~2 HP/s per threat (25 HP per
cycle), salvager range fixed to 60 m with cargo 20, and a monotone
mobility ladder (the placeholder drone was the least agile ship).
Stations anchored to 'a fresh station holds one early parity wave';
enemy level 0 matches the player station and scales per push. HQ 5000
HP. scrap_despawn_seconds 30 -> 120 so capital-sized scrap drops are
collectable.
Rewrite balancing.toml as the canonical 12-arena suite: class mirrors,
equal-threat cross-tier matchups with counts from the verified threat
values, a two-to-one decisiveness check, armored-vs-glass and
repair-escort doctrine matchups, and station assault. The carrier
arena is expected to lose until drone launching exists (hangar is dead
threat) - documented, not stat-compensated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DyCu8vwChKMbLJQ3xosYEN
recipes.toml: full v2 rewrite - iron/copper/quartz mining, smelting
incl. the value-losing scrap sink, reprocessing (4 scrap / 4 s,
voidsteel at 20% of the full pool), tiered intermediates, hulls,
module prefabs, and the three drop-only shortcut recipes.
modules.toml: lasers renamed to railguns (implementation unchanged),
prefab materials and numbers-pass production times; combat stats stay
placeholders for the arena pass.
ships.toml: hull-item materials, numbers-pass base production times,
and geometry-validated default_modules loadouts for every ship (waves
now spawn armed).
visuals.toml: item entries for the new palette (quartz, silicon,
copper_coil, control_chip, capacitor_bank, hardened_steel,
ceramic_plate, voidsteel, voidsteel_plate, railgun modules); retired
titanium/alloy/laser items removed.
world.toml: scrap_per_threat = 0.25 per the numbers pass.
threat_report.py: commit an item's threat only once every eligible
recipe is computable - the previous first-resolved-wins behavior let
shortcut recipes underprice items (same flaw exists in
ThreatCostCalculator, recorded as action item 9). Verified: all
default_modules placements valid, verify_recipes and verify_layouts
pass, and the report reproduces the numbers-pass tables exactly
(fitted: 10.5/47/99/233.5/354.5/722.5/1491.5/1436.5). Doc tables
updated for the three geometry-corrected loadouts.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DyCu8vwChKMbLJQ3xosYEN
