KeiSeiKit-1.0/_primitives/_rust/kei-model-router
Parfii-bot 4d79049eff feat(kei-model-router): registry-driven, three-layer DNA
Removes hardcoded Claude-only Model enum. Pricing constants now read
from _blocks/registries/models.toml at startup; provider/model lookup
goes through a typed Registry returned by registry.rs.

New API surface:
  - Registry::load(dir) → (providers, models, profiles)
  - pick(profile_id, &Registry) → Result<(provider_id, model_id)>
  - cost_micro_cents(model_id, in, out, &Registry) → Option<u64>
  - next_model(model_id, &Registry) → Option<&Model> (ascending cost,
    same provider, skip deprecated)

Files:
  - registry_types.rs      new   107 LOC  (Provider/Model/Profile structs)
  - registry.rs            new   152 LOC  (TOML load + lookups)
  - pricing.rs             rew   127 LOC  (registry-backed, no constants)
  - escalate.rs            rew   181 LOC  (registry-backed ladder + skip deprecated)
  - select.rs              rew   131 LOC
  - select_kernel.rs       new    74 LOC  (Constructor-Pattern split)
  - select_posterior.rs    new   178 LOC  (Constructor-Pattern split)
  - posterior.rs           rew   197 LOC
  - calibrate.rs           rew   175 LOC
  - lib.rs                 rew    53 LOC
  - main.rs                rew   163 LOC  (CLI updated to new API)
  - Cargo.toml             dep   added toml 0.8

Verification (orchestrator-side, RULE 0.13 §Verify-before-commit):
  - cargo check                 → clean
  - cargo test --release        → 58 passed / 0 failed / 0 ignored
  - LOC limit (Constructor)     → max 197 / limit 200
  - largest fn cmd_select       → ~27 LOC / limit 30

DNA-INDEX.md regenerated by kei-registry hook (primitive count
144 → 150 reflects the 6 new/split modules).

=== STATUS-TRUTH MARKER ===
shipped: functional
stubs: 0
cargo-check: PASS
behaviour-verified: yes
follow-up-required:
  - select.rs `estimated_cost` still embeds inline cost constants
    mirroring models.toml; if non-Anthropic providers need dynamic
    pricing in select-time estimation, thread Registry through.
  - External callers of old `cost_micro_cents(Model, ...)` signature
    will break — intentional, no external callers in this workspace.
2026-05-13 21:23:53 +08:00
..
src feat(kei-model-router): registry-driven, three-layer DNA 2026-05-13 21:23:53 +08:00
Cargo.lock feat(kei-model-router): registry-driven, three-layer DNA 2026-05-13 21:23:53 +08:00
Cargo.toml feat(kei-model-router): registry-driven, three-layer DNA 2026-05-13 21:23:53 +08:00
README.md KeiSeiKit-public — clean state 2026-05-01 12:09:03 +08:00

kei-model-router

Model selection (Haiku 4.5 / Sonnet 4.6 / Opus 4.7) for Claude Code Agent spawns. Empirical-posterior decision rule keyed on task-class DNA + Beta posterior + cost minimization, with kernel-smoothing for unseen task classes.

Math

Decision rule:

m*(d̂) = argmin_{m ∈ M} { c(d̂, m) | P[q(d̂, m) ≥ q*] ≥ 1  δ }

Empty feasible set → fallback (top tier) per RULE -1 NO DOWNGRADE.

Posterior: q(d, m) ~ Beta(α₀ + n⁺, β₀ + n⁻) with uniform prior. n⁺ counts rows where outcome='functional' AND escalation_depth=0; n⁻ everything else.

Kernel-smoothed transfer for unseen task classes:

K(d, d') = α_role · 1[role=role'] +
           α_caps · |caps ∩ caps'| / |caps  caps'| +
           α_scope · 1[scope=scope'] +
           α_body · jaccard_bigram(body, body')

Verified pricing

[VERIFIED: https://platform.claude.com/docs/en/docs/about-claude/pricing 2026-04-30]

Model Input/MTok Output/MTok
Haiku 4.5 $1.00 $5.00
Sonnet 4.6 $3.00 $15.00
Opus 4.7 $5.00 $25.00

Opus 4.7 uses a new tokenizer that may produce up to 35% more tokens on identical text — multiply quote accordingly when comparing against Haiku/Sonnet on the same input.

CLI

kei-model-router pricing                  # print pricing table
kei-model-router select <agent> [--prompt P]
kei-model-router calibrate                # re-fit kernel weights
kei-model-router --help

Orchestrator integration (Path B — runtime)

Per RULE 0.13, the orchestrator owns Agent spawning. Before spawning a non-trivial agent the orchestrator can consult the router and pass an explicit model parameter:

kei-model-router select code-implementer-rust \
    --prompt "Add multi-tool integration test for parser"
# → model: sonnet (if posterior built up); model: opus (initial fallback)

Then in the orchestrator's Agent invocation:

Agent({ subagent_type: "code-implementer-rust", model: "sonnet", ... })

Until posterior data accumulates the router conservatively returns top-tier (Opus). As outcome column fills via agent-fork-done.sh STATUS-TRUTH parsing, posterior diversifies and cheaper tiers begin to qualify.

Assembler integration (Path A — compose-time, deferred)

Rebaking the model into generated .md files at assemble time is deferred. Current default model: opus in 55/55 manifests is safe; adopt Path B (orchestrator discipline) until ledger has ≥100 outcome-tagged rows per common task class.

Cubes

File LOC Concern
pricing.rs 167 Verified per-MTok constants (microcents)
dna_class.rs 113 DNA component extraction (role/caps/scope)
complexity.rs 178 τ-estimator (heuristic regex+role+length)
posterior.rs 197 Beta posterior from ledger + Wilson lower b.
kernel.rs 134 Substrate similarity kernel for unseen DNAs
escalate.rs 73 Retry-ladder bookkeeping
select.rs 197 Decision rule (argmin cost s.t. q_lb ≥ q*)
calibrate.rs 193 Offline LOO weight refit (grid search)
main.rs 142 CLI dispatch
lib.rs 35 Module barrel + re-exports

All cubes within Constructor Pattern budgets (≤200 LOC, ≤30 LOC/fn).

Schema dependency

Requires kei-ledger schema v9+ which adds:

tokens_in INTEGER
tokens_out INTEGER
stubs_count INTEGER DEFAULT 0
outcome TEXT CHECK (outcome IN ('functional','partial','scaffolding','fail',NULL))
escalation_depth INTEGER DEFAULT 0
task_class_dna TEXT GENERATED ALWAYS AS (...) VIRTUAL
INDEX idx_agents_task_class ON agents(task_class_dna)

Hooks

  • agent-fork-logger.sh (PreToolUse:Agent, advisory) — writes 'running' row with DNA at spawn.
  • agent-fork-done.sh (PostToolUse:Agent) — closes row + parses STATUS-TRUTH MARKER from agent's tool_response → updates outcome, stubs_count, tokens_in, tokens_out.

Lock

2026-04-30. Phases 1-9 of kei-model-router rollout complete (Phase 9 orchestrator-discipline; assembler refactor deferred until ≥100 outcome-tagged rows accumulate).