Scenarios and their SMAC lineage

Scenarios and their SMAC lineage

No OpenBW, no Blizzard assets, no parity claim

This coworld implements SMAC-style micro combat natively on the coworld-ctf engine. To be explicit, because the source idea framed the game as a mod of an OpenBW wrapper:

• There is no OpenBW, no StarCraft: Brood War binary, no BWAPI, no Blizzard data files, no Blizzard art, no .rep replays and no OpenBW rendering anywhere in this repository or in the image it builds.
• There is no dependency on Metta-AI/coworld-bw, on PySC2, on the SMAC package or on JaxMARL/SMAX. The idea itself sanctions the dependency-free route.
• No bit-exact parity with SMAC, SMACv2 or SMAX is claimed or tested. No numerical, action-space or reward-curve parity. The unit classes (ranger, blade, swarm) are this repo's own and the numbers are tuned here.
• The replay is rendered by our own static WebAssembly viewer, never by OpenBW.

What IS carried over is the shape of the scenarios: a small fixed squad against a scripted enemy army of fixed composition, ranged units that kite, melee units that body-block, focus fire as the decisive skill, and a shared win reward with damage shaping — and the shape of their maps. The plains and corridor boards are original geometry authored for this engine in the layout of the source scenarios' battlefields (an open field for the line fights, one choke doorway for the corridor); no map file or terrain data is imported from anywhere.

The four shipped scenarios

A scenario is a value of three config fields — roles (always five entries), enemyRoles (1..24 entries) and mapPath (plains for the line fights, corridor for the corridor). Nothing else changes: the seat count, the clock, the weapons and the scoring are identical, and every damage term divides by that scenario's own starting pool, so a harder scenario is genuinely harder to score on rather than differently scaled.

default — Micro — two and three

roles = 2 x ranger, 3 x blade. enemyRoles = the same five. Enemy pool 480 hp against our 480.

Lineage: SMAC 2s3z. Fought on the open plains. The smallest squad in which focus fire, kiting and body-blocking are all live decisions at once — two ranged units that must be protected and three melee units that can do the protecting. This is the variant the league ranks.

Full-contact squad dps is 34.6 against 480 enemy hp, i.e. about 14 s of perfect contact, which in practice lands battles at 25-45 s inside the 60 s clock.

outnumbered — Micro — five against six

roles = 5 x ranger. enemyRoles = 6 x ranger. Enemy pool 360 hp against our 300.

Lineage: SMAC 5m_vs_6m. Fought on the open plains. Nobody can body-block, because nobody has the health to. The only way to win the trade is to kill faster than you are killed: one target at a time, and back off while the weapon cools. Every unit out-ranges nothing, so positioning is entirely about who is inside whose 380 px.

corridor — Micro — the corridor

roles = 5 x blade. enemyRoles = 20 x swarm. Enemy pool 600 hp against our 600.

Lineage: SMAC corridor. Fought on the corridor map: the field pinches to one 104 px doorway at the board's centre. Twenty-five bodies on the board. The swarm units are fast (86 px/s) and fragile (30 hp), and they pick the closest thing they can see and walk at it — straight into the doorway. A blade that steps out past the choke is surrounded by four of them and dies; five blades holding the doorway kill them one at a time. This is the scenario that tests whether a commander can make its unit STOP.

heavy — Micro — outgunned

roles = 2 x ranger, 3 x blade. enemyRoles = 3 x ranger, 4 x blade. Enemy pool 660 hp against our 480.

Lineage: SMAC 3s5z_vs_3s6z. Fought on the open plains. Outgunned by 38 %. A victory here needs perfect focus fire from the opening volley; a full-time draw with 70 % damage banked is a real result and scores like one (0.30 x 0.70 + 0.10 x survival). The variant exists to make the damage-shaping term matter on its own, rather than only as a consolation for a lost battle.

Adding a scenario

Add a variant to coworld_manifest_template.json with a new scenario id, a new enemyRoles array and a mapPath (plains or corridor; a new layout is one authored map function in src/smac/arena.nim). That is the whole change: no new art, no new tests beyond tests/test_manifest.nim, which already builds and steps EVERY variant's game_config (the collab-cooking scar — testing only the certification fixture is how a 25-unit roster ships broken).

The seat-count pin bounds every scenario at five units on our side, and enemyRoles is bounded at 24 entries by config_schema, so no variant can schedule an unbounded army.

Out of scope for v1

SMACv2's randomised unit types per episode and its free-form start positions are v0.2 — roles and enemyRoles are already the right shape for a generator to write into, and the seeded +-24 px spawn jitter is the whole of the randomisation today. The other twenty-odd SMAC scenarios are enemyRoles values away and cost nothing but variants. SMAC's built-in-AI difficulty ladder is not modelled: this ships one script, and it is published in docs/RULES.md so a prompt can be written against it.