Appendix B: Lumen Glossary

Agent A specialized AI sub-system in Lumen that handles one PM domain. Each agent reads specific inputs (context slots), does its analysis, and writes outputs (context slots) for downstream agents. Examples: SignalMonitor, DiscoveryOS, DecideWell. You never invoke agents directly — the Orchestrator sequences them.

Churn attribution type: The classification of why a customer churned. Lumen classifies into four types: product churn (usage dropoff precedes cancellation), pricing churn (cancellation spike coincides with a price change), competitive churn (exit interviews cite a specific competitor), and involuntary churn (payment failure, no usage dropoff). The recovery strategy is different for each type.

Churn cohort fingerprint: A label written by SignalMonitor that describes the behavioral pattern of a churned cohort. Used by GrowthIQ to match new at-risk accounts against prior churn patterns. Example: "2026-03 Team plan, roadmap editor regression cohort."

Consensus mode: A Scale-tier DecideWell capability that solicits structured input from multiple agents (and optionally stakeholders) before issuing a recommendation. Used when a decision affects multiple teams or has conflicting evidence from different data sources.

Context slot: A named piece of information that flows between agents during a workflow run. Agents read required slots (blocking if absent) and optional slots (improving output quality if present). Example: pmf_score_by_segment is written by SignalMonitor and read by DiscoveryOS and RoadMap.

Crisis severity score: A 0–10 score produced by SignalMonitor that measures the urgency of a churn or PMF regression event. Score ≥ 7 triggers an automatic routing to W2 (PMF Recovery). Score 5–7 triggers targeted intervention. Score < 5 is monitoring mode.

DataLayer: The governance agent responsible for AI ethics clearance and regulatory compliance. The only Lumen agent that operates at Level 3 oversight. Issues GRANTED, CONDITIONAL, or BLOCKED status for AI/ML features. A BLOCKED status stops the workflow entirely.

DecideWell: The decision-structuring agent that always requires Level 2 PM approval before writing a decision memo. Evaluates multiple options against evidence quality, capacity constraints, and explicitly excludes previously tried interventions. Available with consensus mode at the Scale tier.

Discovery OS (DiscoveryOS): The customer discovery agent that synthesizes interview transcripts and PMF signals into an opportunity solution tree. Checks for interview saturation (threshold: 20 per segment) and participant bias (power-user skew, recency bias).

Evidence quality: A per-agent rating of how much data backed a recommendation. Values: HIGH (all criteria met — sufficient sample size, recent data, source quality), MEDIUM (one criterion missing), LOW (two or more criteria missing). The Orchestrator applies a lowest-wins rule: the overall report grade equals the lowest individual agent grade.

GrowthIQ: The revenue retention agent that decomposes NRR into expansion, contraction, and churn. Ranks at-risk accounts by ARR × renewal proximity × health score. Surfaces product-qualified leads for expansion. Available at the Growth tier.

HypothesisLab: The experimentation agent that designs statistically rigorous experiments. Outputs a hypothesis, sample size (via power analysis), minimum detectable effect, and time estimate. In crisis workflows, design the fastest possible intervention (2-week maximum).

Knowledge Graph (KG): The persistent data store for all Lumen outputs, stored in Supabase. Contains 15 entity types, including Decision, Experiment, UserSegment, and GovernanceEvent. Entities link through edges. Enables outcome tracking, cross-session comparison, and historical context for future workflow runs. Only active when Supabase is connected.

LaunchPad: The GTM readiness agent that audits launch readiness across seven dimensions. Produces a scored readiness report against launch-type thresholds (feature: 7.0/10, product: 8.0/10, pricing: 9.0/10, market entry: 8.0/10). Runs twice in W5 — open audit before launch and close audit for Day 1/7/30 monitoring.

MarketIQ: The competitive intelligence agent that maps the competitive landscape across four moat dimensions: switching costs, network effects, data advantages, and brand. Available at the Growth tier.

MCP (Model Context Protocol): The standard that allows Claude Code to connect to external tools and data sources. Lumen uses 10 MCP servers: PostHog, Supabase, Upstash, Slack, HubSpot, Stripe, Figma, GitHub, Sentry, and Postman.

Minimum detectable effect (MDE): The smallest meaningful change in a metric that an experiment is designed to detect with statistical significance. HypothesisLab calculates this from the baseline metric, sample size, and power level (typically 80%). A smaller MDE requires a larger sample size and longer run time.

NarrativeEngine: The messaging agent that translates PMF signals and competitive positioning into a complete narrative suite: product one-liner, category definition, key claims, proof points, and audience messaging matrix. Flags claims that require legal review. Available at the Growth tier.

NorthStar: The metric-definition agent that identifies the single North Star metric best correlated with customer value and revenue retention. Cascade the North Star into team-level OKRs and flags cross-team conflicts. A North Star change triggers a Level 2 gate.

NRR (Net Revenue Retention): Measures revenue retained from existing customers, including expansion and contraction. NRR = (Starting MRR + Expansion − Contraction − Churn) / Starting MRR. Above 100% means existing customers are growing. Below 100% means they are shrinking even before new customer revenue.

OpsCommand: The operating rhythm agent that generates executive updates, board briefings, and weekly PM ritual templates. Escalates to Level 2 for board-level communications or crisis declarations. When Supabase is connected, and a prior run exists, it leads the update with the delta from the last report. Available at the Growth tier.

Orchestrator: The master agent that is the entry point for all Lumen workflows. Sequences sub-agents, enforces tier gates, manages oversight gates, aggregates evidence quality, and generates the final report. You interact with Lumen through the Orchestrator — you never call sub-agents directly.

Outcome tracking ID: A unique identifier assigned to every DecideWell decision memo and stored in the Knowledge Graph (when Supabase is connected). Used to link a decision to its 30/60/90-day follow-up comparison. Example: outcome-helix-w1-team-plan-recovery-20260313.

Oversight gate: A mandatory pause in a workflow where the PM must respond before the workflow continues. Four levels: L0 (automated, no pause), L1 (advisory, 24-hour override window), L2 (approval required — APPROVE / DECLINE / MODIFY, 48-hour window), L3 (governance approval by a named authority, 72-hour window).

PARTIAL mode: The W1 PMF Discovery operating mode when PostHog is not connected. SignalMonitor is skipped, and the PMF Scoring section is empty. EventIQ and DiscoveryOS still run with available data. The report is marked PARTIAL at the top.

PMF score: A 0–100 score produced by SignalMonitor measuring product-market fit for a specific segment. Score interpretation: below 30 = PMF NOT FOUND, 30–40 = WEAK PMF, above 40 = CONFIRMED PMF. Requires PostHog to produce. Cannot be substituted with CSV uploads.

PQL (Product-Qualified Lead): An existing account (typically on a lower pricing tier) that is showing usage patterns consistent with the next tier up. GrowthIQ surfaces PQL alerts in W6 as expansion opportunities. Acting on PQLs is typically the highest-leverage short-term NRR improvement.

PriceLogic: The pricing agent that designs or validates pricing models by analyzing segment willingness-to-pay, competitor pricing, and NRR impact. Always runs an existing customer protection check. Advisory (Level 1). Available at the Growth tier.

Roadmap: The roadmapping agent that translates opportunity trees, capacity data, and OKR cascades into a quarterly roadmap with committed, likely, and exploratory tracks. Flags customer commitment conflicts and capacity overruns. Advisory (Level 1). Available at the Growth tier.

SetupGuide: The setup agent that runs automatically at the start of every session. Validates MCP connections, reads lumen-setup.json, and seeds the context object with project configuration. Operates in four modes: first-run, reconnect, pass-through, and single-MCP.

SignalMonitor: The PMF health agent that scores product-market fit by segment, detects churn precursors, calculates crisis severity, and classifies churn type. The primary consumer of PostHog data. Cannot run without PostHog — marks W1 as PARTIAL if absent.

Tier gate: A restriction that limits certain agents or capabilities to specific subscription tiers. Starter, Growth, Scale, and Enterprise each unlock additional agents. The Orchestrator enforces tier gates at the start of each workflow run.

UNAVAILABLE The status shown when an agent was skipped because a required MCP was absent. Different from LOW evidence — UNAVAILABLE sections are excluded from the overall evidence quality aggregation and do not count against the report grade.

UXLayer: The interaction design agent that produces wireframe-level interaction specs for product features. Flags WCAG accessibility violations. For AI features, always include a human-override interaction pattern consistent with DataLayer clearance conditions. Available at the Scale tier.

Workflow: A named sequence of agents that runs from end to end to answer a specific PM question. Lumen has six workflows: W1 (PMF Discovery), W2 (PMF Recovery), W3 (Strategy), W4 (Feature Validation), W5 (GTM Launch), and W6 (Churn Analysis). Each workflow is triggered by a slash command.