Industrial Genome Platform

Component Architecture

Full stack from public signal ingestion through fingerprinting, classification, Studio UI, and commercial outreach engine.

April 2026 ~/projects/genome

System Architecture

Signal Source

FRED

St. Louis Fed CSV API. 40+ series. Free, no key. Industrial production, PPI, capacity utilization, freight.

signals/fred.py

Signal Source

Sector Bundles

16 industry-specific output signal sets. Differentiates entities by sector sensitivity to the same macro inputs.

signals/sectors.py

Signal Source — Planned

SEC EDGAR

Entity-specific financials. 10-Q/10-K revenue, COGS, inventory, capex. Enables company-level vs. sector-level fingerprinting.

signals/edgar.py

Signal Source — Planned

EIA / Baker Hughes

Energy production, rig count. For industrials with energy exposure: ag equipment, heavy equipment, chemicals.

signals/eia.py

↓

Signal Store

Supabase PostgreSQL — project mngmqcmzscltyplnspvx

Tables: companies · signal_events · fingerprints · judgments. TimescaleDB upgrade path for hypertable partitioning when needed. Shared with ops-maturity and Market Intelligence projects.

api/db.py · api/persist.py

↓

Core IP — Algorithm Layer

System ID Engine

Six fingerprint algorithms: latency (cross-correlation), gain (% change ratio), damping (detrended autocorr), oscillation (FFT + Nyquist cap), saturation (slope-change), volatility transmission (std ratio).

core/system_id.py

Core IP — Classification

Fingerprint + Archetype

Builds structured fingerprint from algorithm outputs. Priority-ordered rule set classifies into 7 archetypes. Peer cohort comparison by industry.

core/fingerprint.py · core/archetypes.py

Core IP — Inference

Judgment Engine

Fingerprint → constraint node → trajectory inference → probability-weighted implications → recommended actions with priority, impact, and time-to-effect.

core/hidden_state.py

↓

API Layer

FastAPI — localhost:8000

POST /analyze (fingerprint + judgment pipeline, auto-FRED fallback if no signals provided) · GET /entities (Atlas list) · GET /entity/{id} (history + judgment) · GET /cohort/{industry} (peer data). Persists fingerprints and judgments to Supabase.

api/main.py · api/persist.py

↓

Consumer — UI

Genome Studio

Next.js 15 app. Atlas scatter (gain × damping, colored by archetype). Entity detail: Judgment / Fingerprint / Trajectory / Report / Docs tabs. Analyze form for ad-hoc sequencing.

studio/ · localhost:3002

Consumer — Commercial

Outreach Engine

Targeting → genome reports → Gemini/Ollama message generation → email send (dry-run default) → response classification → strategy learning. CLI and batch runner.

commercial/ · cli.py

Consumer — Planned

Operating Command

COO-facing decision support. Real-time constraint detection, intervention simulation, action prioritization. Requires entity-specific signal ingestion (SEC EDGAR).

TBD

Repository Structure

genome/

core/ # The IP layer — system ID algorithms

system_id.py # latency, gain, damping, oscillation, saturation, vol_tx

fingerprint.py # FingerprintResult builder + signal bundle

archetypes.py # Priority-ordered rule classifier → 7 archetypes

hidden_state.py # Judgment engine → implications + actions

signals/ # Public signal ingestion

fred.py # FRED CSV API client + dedup

macro.py # Default macro input/output bundle

sectors.py # 16 industry-specific output signal bundles

api/ # FastAPI backend

main.py # Routes: /analyze /entities /entity /cohort

db.py # Supabase client singleton

persist.py # fingerprint + judgment persistence

commercial/ # Commercial outreach engine

targeting/service.py # PE firm targets + company selection

reporting/service.py # Genome report generation + persistence

outreach/service.py # Message generation (Ollama/Gemini)

outreach/sender.py # Email send (SendGrid stub, dry-run default)

outreach/response_parser.py # Inbound reply classification

orchestrator/engine.py # run_single / run_cycle / update_strategy

orchestrator/cli.py # CLI: run / single / report / targets

studio/ # Next.js 15 frontend

app/page.tsx # Atlas: scatter + entity list

app/entity/[id]/page.tsx # Entity detail (5 tabs)

app/analyze/page.tsx # Ad-hoc analyze form

app/docs/page.tsx # Documentation index

components/JudgmentPanel.tsx # Archetype + constraint explanation + actions

components/FingerprintRadar.tsx # Recharts radar chart

components/AtlasScatter.tsx # Recharts scatter, click-to-select

lib/taxonomy.ts # Archetype explanations + D1–X3 constraint labels

migrations/ # Supabase SQL migrations

001_initial_schema.sql

002_judgments.sql

003_add_industry.sql

004_commercial_schema.sql # targets, prospect_companies, outreach_*, strategy_state

Algorithm Details

Algorithm	Method	Key design decisions
estimate_latency	Cross-correlation peak across lags 0–12	Series aligned by timestamp (_align()) before correlation. Different FRED series have different observation counts — inner join on shared dates, fallback to positional truncation.
estimate_gain	Median ratio of output % change to input % change	Uses % changes (not levels) for unit-agnosticism (WTI $/bbl vs INDPRO index). Confidence = min(1.0, n_moving_periods / 12). Filters near-zero input moves (threshold 1e-4).
estimate_damping	1 − mean(\|autocorr\|) of detrended residuals	Detrend first (linear polyfit), then autocorr of residuals at lags 1–4. Uses absolute value — both positive persistence AND negative alternation indicate low damping. High autocorr = low damping.
estimate_oscillation	FFT dominant frequency + spectral power fraction	Nyquist cap: period > observation_window/2 is a trend artifact, not a cycle. Eliminates spurious 152-week "cycles" from 35-point series. Returns None if cap triggered.
estimate_saturation	Slope-change detection via piecewise fit	Identifies where output growth flattens relative to input growth. Returns threshold value and confidence.
estimate_volatility_transmission	std(dy%) / std(dx%)	Uses % changes for unit-agnosticism. Ratio > 1 = system amplifies shocks. < 1 = absorbs. Key fragile/resilient discriminator.

Database Schema

fingerprints

entity_id, entity_nametext

industrytext

gain, damping, latency_daysnumeric

oscillation_period_weeksnumeric

volatility_transmissionnumeric

saturation_thresholdnumeric

confidencejsonb

archetypetext

sequenced_attimestamptz

judgments

entity_id, fingerprint_idtext / uuid

industrytext

system_type, constraint_nodetext

trajectory, trajectory_horizon_weekstext / numeric

confidence, signal_strengthnumeric

implicationsjsonb[]

recommended_actionsjsonb[]

peer_comparisonjsonb

narrative_summary, report_texttext

targets

name, typetext (pe_firm / ma_advisor)

aum_bn, industry_focus[]numeric / text[]

statusnew|enriched|active|paused

outreach_messages

target_id, company_id, report_iduuid refs

subject, bodytext

statusdraft|sent|opened|replied

tone, messaging_styletext

outreach_responses

message_id, contact_iduuid refs

classificationinterested|not_now|unsubscribe|referral|no_signal

signal_strength0.0–1.0

next_actionfollow_up|demo|proposal|none

strategy_state (single row)

messaging_styleinsight_led | question_led

tonedirect | consultative | brief

max_daily_sendsinteger

performance_metricsjsonb

preferred_industriestext[]

Services & Deploy

Service

Stack

Start command

URL

API

FastAPI + uvicorn

uvicorn api.main:app --reload

localhost:8000

Studio

Next.js 15

npm run dev (studio/)

localhost:3002

Ollama (LLM)

llama3.2:3b local

ollama serve

localhost:11434

CLI

Python 3.12

.venv/bin/python3 -m commercial.orchestrator.cli

—