SAKANAL Console
24h7d30d
$7.5423
Actual spend (30d)
6,647 calls · 30d window
$1.7325
Counterfactual — burn without SAKANAL
priced for 16.04% of calls — a floor, not a total
$0.0012
Dividend (counterfactual − actual)
= cache + routing + compaction savings
Read the counterfactual honestly. The "burn without SAKANAL" figure is only populated for the 16.04% of calls the gateway could price. The rest have no baseline (routing is shadow-first and the default-model table does not exist), so the true number is higher — and the console will not guess it. On a fleet this small, the Value view is a correctness proof, not a savings case.

Savings by lever

LeverFormulaSaved (USD)Status
Routing Savingscounterfactual(model requested) − actual cost$0.0000LIVE
1 request(s) re-routed in window.
Caching Savingscache_read_tokens / 1e6 × (input_rate − cached_input_rate)$0.0012LIVE
109,594 cache-read token(s) in window. Made-visible, never caused.
Compaction — Dropped × RateDropped_Tokens × Model_Rate = jev_compact_saved / 1e6 × input_rate$0.0000PLACEHOLDER
2,410 dropped token(s) in window — shadow-only, and the target model has no seeded rate, so this reads 0 by the 022 rule.

Master equation (migration 022): Dividend = Σ(counterfactual − cost) = Σ(savings_cache) + Σ(savings_routing) + Σ(savings_compaction)

Go-live gate (migration 022) — 2/4

The green savings number stays off until all four conditions pass.

#ConditionVerdictDetail
G1Every spending model has a verified rate rowFAILmissing: google/gemini-3.1-flash-lite, deepseek-chat, gpt-4o-mini, typesafe/jev-latest, deepseek/deepseek-v4-flash-0731, gpt-4o, qwen/qwen3.8-max
G2Gemini cache_read_tokens non-zeroPASS24,504 tokens
G3default_model_per_feature covers spending featuresFAILtable absent — routing baseline not computable
G4Every rate row has a non-empty sourcePASSall sources set

Rate coverage — spending models vs model_rates

ModelSpend (30d)Rate row
google/gemini-3.1-flash-lite$4.5456missing
google/gemini-2.5-flash$1.7740seeded
deepseek/deepseek-v4-flash$0.4769seeded
openai/gpt-4o-mini$0.4487seeded
google/gemini-2.5-flash-lite$0.1835seeded
deepseek/deepseek-v4-pro$0.0898seeded
deepseek-chat$0.0066missing
gpt-4o-mini$0.0062missing
typesafe/jev-latest$0.0059missing
deepseek/deepseek-v4-flash-0731$0.0026missing
deepseek/deepseek-chat$0.0015seeded
openai/gpt-4o$0.0006seeded
gpt-4o$0.0005missing
qwen/qwen3.8-max$0.0000missing

Missing rates: google/gemini-3.1-flash-lite, deepseek-chat, gpt-4o-mini, typesafe/jev-latest, deepseek/deepseek-v4-flash-0731, gpt-4o, qwen/qwen3.8-max — these models write 0 savings (022 rule: never guess).

Generated 2026-09-28 13:44:03Z · window 30d