Profiles — the allocation matrix
From the Meridians Wiki · Public · Maintained · joint
Each column is one profile; read it top-to-bottom to see every category's tier. B=Budget,
Bal=Balanced, Perf=Performance, Front=Frontier. Balanced is the shipped default
(DEFAULT_PROFILE_ID = "balanced-smart") and is the original hand-tuned allocation.
Smart profiles
Intelligent per-category curves — substrate/reasoning work is the last to drop and first to rise; latency-bound and pure-volume work is the first to drop and is capped below Frontier.
| Category | Budget | Balanced | Performance | Frontier |
|---|---|---|---|---|
| generate | Bal | Perf | Perf | Front |
| plan | B | Bal | Perf | Front |
| extract | B | Bal | Perf | Front |
| predict | B | Bal | Perf | Front |
| write | B | Bal | Perf | Front |
| review | B | B | Bal | Perf |
| synthesis | B | B | Bal | Perf |
| gameTheory | B | B | Bal | Bal |
| interaction | B | B | Bal | Perf |
| action | B | Bal | Bal | Bal |
| intuition | B | Bal | Bal | Perf |
| question | B | Bal | Bal | Bal |
Design notes:
- Budget protects
generateat Balanced — dropping the substrate-committed scene generator to the cheapest model poisons everything downstream. "Cheap but not stupid." - Performance lifts substrate + reasoning to Performance; latency-bound
actionstays at Balanced (a slow move freezes the play clock). - Frontier puts substrate + reasoning + conversation on Kimi K3, but caps
action,gameTheory, andquestionbelow it — the frontier tier's cost/latency isn't worth it there.
Singular profiles
Naive "pin every category to one tier" modes.
| Profile | Every category → |
|---|---|
| All Budget | Budget |
| All Balanced | Balanced |
| All Performance | Performance |
| All Frontier | Frontier (Kimi K3) |
Custom allocations layer per-category overrides on top of any base profile (a Director pins a few categories in the AI Profiles modal; the rest follow the base).