The Krentix Index · 2026-W26 · 2026-06-27
What Krentix can and can’t do this week.
Krentix is the multi-agent reasoning layer behind TradeCaliber. This page publishes the unflattering truth about its capability and cost — refreshed weekly, generated from the live system, no marketing copy. tradecaliber.com · live telemetry
eval 0.0%
phase-c 73.2%
kg 82 nodes
corpus 77 events
embeddings on
Capability snapshot
Knowledge graph
82
biotech-catalyst-trading · v0.2.0
Corpus events
77
4 sources · TF-IDF + OpenAI text-embedding-3-small
Curated archetypes
31
$0-cost canonical answers
Eval pass rate · 30 days
| position-brief.thesis | 46 | 0.0% |
| morning-briefing | 23 | 0.0% |
| explain-this-trade | 23 | 0.0% |
| teach-me.pdufa | 23 | 0.0% |
| comparables.crl_resubmission | 23 | 0.0% |
Weighted pass rate: 0.0% across 138 judged outputs.
Phase C bridge agreement
Agreement rate (target 85%)
73.2%
71 trades replayed · last 2026-04-29 · status: below
Below the gate means the Krentix bridge is not yet authoritative — it advises but does not decide. The Phase C decision-authority module (~8 weeks engineering) closes this gap. Until then, Krentix is read-only on this dimension.
Cost discipline
Process daily cap
$5
REASONING_DAILY_USD ceiling on the whole reasoning surface.
Model router
22
intents across 3 tiers · 73% routed to Haiku-class
Corpus sources
| historical_topline_events | 32 |
| TF7_CRL_Resubmissions | 19 |
| historical_btd_events | 17 |
| historical_adcom_events | 9 |
Generated by scripts/krentix-index-builder.js from live system state. Cached 6h. Page is intentionally plain — no auth, no telemetry beacons.
Powered by TradeCaliber · Krentix is the reasoning layer · 2026-W26