improving pitching

docs/src/business.md

@@ -1,21 +1,39 @@
 # Business overview
 
-PHANTOM targets **platform operators and researchers** who need to:
+Dynamic pricing extracts margin by exploiting the information asymmetry between buyer and seller. When a user browses a flight or hotel across multiple sessions, each interaction accumulates demand signals that push the quoted price upward. That is the mechanism working as intended.
 
-1. **Observe** session-level behavior and price quotes together (trajectories and policies—not just clicks).
-2. **Separate** human-driven demand signals from agent-mediated reconnaissance where possible (distinguishability and contamination \alpha in the thesis).
-3. **Evaluate** pricing policies that remain useful when **Cost of Information (COI)** is under pressure from automated querying (formal COI framework and theorem in the thesis PDF).
+LLM agents break it. An agent can conduct reconnaissance—across dozens of isolated sessions, at machine speed—and then execute a purchase through a clean session that looks like a first-time visitor. The platform sees a low-engagement session and quotes a floor price. The margin that should have been captured, the **Cost of Information (COI)**, vanishes. At scale this is not a theoretical concern; it is a structural leak in any session-based pricing system.
 
-## What this product is not
+**PHANTOM is a research platform for studying and defending against that leak.**
 
-- A drop-in fraud API that returns “bot score” for every request without your event schema.
-- A certified compliance guarantee for regulated pricing: it is a **research stack** with configurable experiments.
-- A hosted SaaS: you run the stack (or adapt components) under your infrastructure policy.
+## Who it is for
 
-## Self-service story (ideal path)
+| Role | What they get |
+|---|---|
+| Pricing and revenue researchers | A controlled lab with instrumented human and agent sessions, behavioral kernel estimation, and contamination simulation at configurable levels |
+| Platform engineers evaluating agent risk | A concrete pipeline from behavioral event logs to a per-session agent-probability signal, ready to feed into an existing pricing provider |
+| RL practitioners | A Distributionally Robust RL gym built on a Wasserstein ambiguity set, with benchmark tiers and sweep tooling out of the box |
 
-A team connects their **catalog** (today: Supabase-backed flows in this repo), streams **interaction events** through the ingest path, runs **labeled or weak-labeled** human vs agent sessions, estimates **behavioral kernels**, varies **contamination** in simulation, and **trains or benchmarks** robust policies via `engine/`. Steps and caveats are in [Setup](platform-setup.md) (same content as root `SETUP.md`).
+## Core capabilities
 
-## Thesis link
+**Behavioral fingerprinting.** PHANTOM logs interaction trajectories at the event level (action, item, timestamp) and fits separate Markov transition kernels for human and agent sessions via MLE. Per-session divergence scores (Δ_H, Δ_A) and a learned agent-probability signal f(τ) are computed on partial trajectories in real time, giving the pricing layer a continuous signal rather than a binary bot flag.
 
-Problem statement, contributions, and research questions: **Introduction** and abstract in the [thesis PDF](https://pub-d5b94a3c29fd40c6b3881946e463fdb7.r2.dev/thesis-latest.pdf).
+**Contamination simulation.** The contamination generator G(α) mixes real human trajectories with synthetic agent trajectories at a configurable ratio α. This lets you evaluate pricing robustness across the full spectrum from purely human traffic to fully automated demand, without needing live agent traffic in production.
+
+**Robust policy training.** The defense gym trains pricing policies against the worst-case demand distribution within a Wasserstein ball around the generator's empirical distribution. The reward function penalizes COI leakage (weighted by agent probability) while bounding UX degradation for legitimate users.
+
+## The path from logs to defense
+
+A team: connects their catalog and ingest path → streams interaction events through Kafka → labels or weak-labels sessions → estimates behavioral kernels → varies α in simulation → trains and benchmarks robust policies. The full walkthrough is in [Setup](platform-setup.md).
+
+## Scope and honest caveats
+
+This is a **research stack**, not a hosted service:
+
+- It ships two demo verticals (`hotel`, `airline`); a new catalog requires engineering work on events and reward features.
+- Kernel estimates are research-grade until validated on your traffic distribution.
+- There is no built-in compliance layer for regulated pricing markets.
+
+The thesis PDF contains the formal proofs, the COI erosion theorem, and the full DR-RL formulation. The code operationalizes those constructs—every term in the reward function maps to something computed from your logs.
+
+**Thesis PDF:** [thesis-latest.pdf](https://pub-d5b94a3c29fd40c6b3881946e463fdb7.r2.dev/thesis-latest.pdf) — Introduction and Chapter 3 cover the problem statement, contributions, and formal model.

docs/src/index.md

@@ -1,21 +1,23 @@
-# PHANTOM platform documentation
+# PHANTOM
 
-Welcome. This site mirrors the **operator and research** documentation for the PHANTOM repository: a research platform for studying **dynamic pricing** under **LLM-mediated browsing and transaction orchestration**, with ties to the academic thesis.
+LLM agents are quietly eroding the pricing power of dynamic pricing systems. They conduct reconnaissance across isolated sessions at machine speed and execute purchases through clean sessions that quote floor prices. The margin that should have accumulated never does.
 
-## Start here
+PHANTOM is a research platform for measuring, simulating, and defending against that erosion. It provides behavioral fingerprinting of human vs agent sessions, a contamination generator for controlled experiments, and a Distributionally Robust RL gym for training pricing policies that hold up under automated demand.
 
-| Document | Audience |
+---
+
+## Where to start
+
+| Document | What it covers |
 | --- | --- |
-| [Setup](platform-setup.md) | Full walkthrough: Docker/web/ingest, kernels, contamination, RL training, and audit—content from `SETUP.md` in the repo. |
-| [Configuration reference](configuration.md) | Env vars, paths, and Makefile entrypoints in one place. |
-| [Roadmap & implementation notes](roadmap.md) | What is turnkey vs research-grade; thesis vs code. |
+| [Business overview](business.md) | The problem, capabilities, and who this is for |
+| [Setup](platform-setup.md) | Full bring-up: Docker stack, ingest, behavioral kernels, contamination, RL training |
+| [Architecture](architecture.md) | Service map and data flow |
+| [Configuration reference](configuration.md) | Env vars, paths, and Makefile targets |
+| [Roadmap & notes](roadmap.md) | What is turnkey vs research-grade |
 
-## Canonical sources in the repo
+## Key references
 
-- Thesis PDF: [thesis-latest.pdf](https://pub-d5b94a3c29fd40c6b3881946e463fdb7.r2.dev/thesis-latest.pdf)
-- Root onboarding: single file [`SETUP.md`](https://github.com/velocitatem/PHANTOM/blob/main/SETUP.md) (included on this site via snippets—edit that file to change content).
-- Quick start and command tables: [`README.md`](https://github.com/velocitatem/PHANTOM/blob/main/README.md)
-
-## Academic project page
-
-The research landing page (figures, abstract, links) is the site root on GitHub Pages: [velocitatem.github.io/PHANTOM/](https://velocitatem.github.io/PHANTOM/). Open **Documentation** in the Project Links menu there to return to this subsite.
+- **Thesis PDF:** [thesis-latest.pdf](https://pub-d5b94a3c29fd40c6b3881946e463fdb7.r2.dev/thesis-latest.pdf) — formal model, COI erosion proof, DR-RL formulation
+- **Repo root:** [`SETUP.md`](https://github.com/velocitatem/PHANTOM/blob/main/SETUP.md) | [`README.md`](https://github.com/velocitatem/PHANTOM/blob/main/README.md)
+- **Academic landing page:** [velocitatem.github.io/PHANTOM/](https://velocitatem.github.io/PHANTOM/)