FBED LABS · RESEARCH & API

The science is published.
The tools are real.

The product is opinionated because the evidence is. CASCADE learns from real conversations, stress-tests the next move in simulation, and keeps the proof visible.

9.84M
Parameters
A real neural network, not an LLM wrapper
400M
Training deals
Shannon has played your next call 400 million times
5,000+
Conversations
Parsed at sentence level as ground truth
107
Discovered params
Found by the data, not set by humans
41
Seller actions
Each one calibrated with backfire conditions
75
Agent tools
Shannon is an agent, not a chatbot

SHANNON AGENT · CLAUDE TOOLS PROTOCOL

75 tools. 14 groups. Best in class.

Shannon is not guessing from a notes document. She can inspect the actual deal memory, compare the call to learned patterns, and choose the next useful tool.

cascadeintelligence.ai / api / shannon-tools
SHANNON AGENT · CLAUDE TOOLS PROTOCOL
75 tools across 14 groups
256 tables · 398 RPCs · 28 views
Pipeline8
get_pipeline_summaryget_lead_detailget_pipeline_healthsearch_leads_by_name
Analysis9
get_conversation_scoresget_latest_deltasget_lead_problemssearch_conversations
Coaching8
get_call_prepget_flow_coachingget_pattern_coachinggenerate_call_prep
Scoring5
get_scoring_configget_delta_targetsexplain_score
Patterns7
get_pattern_trendsget_patterns_by_outcomeget_pattern_cooccurrence
Network5
get_network_learningsget_network_statsget_thesis_status
Outcomes8
get_outcome_analysisget_win_loss_summaryrecord_outcometrigger_rescore
ICPX4
get_icp_profileget_icp_analysisget_lead_research
Context6
get_seller_statsget_client_contextget_product_card
Memory4
save_memoryread_memories
Files3
list_filesread_file
Reference2
read_reference_doc
Progress2
update_progress
Debug4
get_system_statusexplain_scoring_formula
Full Claude Tools Protocol spec: 24K tokens · Entity graph with FK chain verification · Submission-ready

ARCHITECTURE

How the science compounds

RLOB
Reinforcement Learning From Observed Behaviour
A third paradigm alongside RLHF and RLVR. No judge, no verifier — the behaviour itself is the training signal. Learns from observed human-to-human interaction in controlled environments.
Closed-loop flywheel
3 production applications
CASCADE, ICPx, and ConversationOS collectively span lead identification through deal closure. Every interaction feeds the central model. The dataset grows AND improves with every conversation.
Simulation-to-reality
Continuous recalibration
Real conversations are compared against simulation predictions. Divergence recalibrates the physics. The recalibrated sim generates better training data. The loop never stops.

Published research notes

NOTE · 04
Complete Analysis
The six discoveries nobody else has found. Why "sounds good" never closes. Why frozen buyers can't decide. Why the seller's need kills 13× more deals.
NOTE · 03
Move Stack
Every move a seller can make, classified and calibrated. 41 actions with personality modifiers — what works on a D-type destroys an S-type.
NOTE · 02
Body Language Analysis
Your buyer's body leaks through their words. 12 signals we detect at 82.6% accuracy. The most valuable: when the buyer computes their own ROI.
NOTE · 01
Empirical Delta Analysis
When 4 of 6 dimensions cross threshold, conversion jumps to 80%. Below 4: 20%. We validated this across 101 real deals.
NOTE · 00
Physics Calibration
The simulation predicted one thing. Reality showed another. We recalibrated 107 parameters until the physics matched.
NOTE · 00
HPC Validation
What happens to a deal between calls? Trust decays. ROI forgets. We modelled the decay curve and validated it against real cadence data.

Better should win.

Build on CASCADE.
The data is real. The tools are open.

Book a technical walkthrough if you need to see how Shannon turns conversation evidence into a next move.

Book a technical walkthrough →Developer docs →Start from $40/mo →