I built Eripion Product Coroner from zero as an AI-capable product decision-intelligence system: product brief intake, parser and validation layer, deterministic 900-agent society across four markets, scale-free social graph, culture and category mechanics, optional AI provider mode, sensitivity analysis, mechanism ablations, validation dashboard, run history, shareable reports, Markdown export, and local-first security safeguards.
Know before the market knows. AI-capable product decision-intelligence for stress-testing launch assumptions. I built the full system from zero: structured brief builder, deterministic 900-agent simulation engine, optional AI mode, segment readouts, validation dashboard, and shareable report workflow.
Quick read
Engine: 900 deterministic agents across Georgia, Germany, China, and the USA.
Model: scale-free social graph, homophily, cultural modifiers, category mechanics, optional AI mode.
Outputs: adoption pressure, churn risk, advocacy, paid conversion, next experiments.
Validation: 53/53 metrics in or near benchmark range with visible uncertainty bands.
About
Product Coroner for AI-capable product-assumption stress tests before launch.
The report surface is only the final layer. I built the product workflow, simulation engine, agent model, validation layer, AI-capable runtime option, and decision outputs so a founder can turn a brief into an auditable stress test before spending real build or launch budget.
Engine900-agent society
AIOptional provider-assisted mode
OutputReports · history · Markdown
Validation53/53 benchmark checks
01
Product System
I built Eripion Product Coroner from zero as an end-to-end decision product: brief intake, validation, simulation runtime, report generation, examples, history, sharing, and export surfaces.
AI-Capable Engine
The core is a deterministic 900-agent society with optional AI assistance, snapshot simulation steps, scale-free social graph mechanics, cultural modifiers, category behavior, and segment readouts.
Evaluation Layer
I built the validation and decision layer around benchmark ranges, uncertainty bands, sensitivity analysis, mechanism ablations, adoption pressure, churn risk, GTM actions, and next experiments.
AI + Simulation Architecture
Inspectable, not prophetic.
Eripion Product Coroner does not claim that the market will do X. I built it to show which segment understood the product, where churn appears, what trust proof is missing, and which mechanism moved adoption or retention across deterministic and optional AI-assisted runs.
The benchmark system is a structural sanity check, not a prediction-accuracy claim. The app keeps raw, adjusted, and uncertainty-band outputs visible so the validation story is auditable.
Proof Surfaces
Product BriefStructured input instead of vague founder intuition.
Decision ReportSegment-level adoption, churn, advocacy, and next experiments.
Validation DashboardBenchmark ranges, uncertainty bands, and parameter provenance.
Build Scope
Built from zero.
Scope here means the actual build ownership: I designed and implemented the product workflow, engine logic, agent data model, graph mechanics, AI-capable runtime option, validation layer, reporting system, local persistence, API surfaces, export flow, and security posture.
Full Product
Built from zero
I built the application layer, brief workflow, API routes, state model, examples, report pages, run history, shareable outputs, Markdown export, and proof surfaces as one coherent product system.
Next.jsTypeScriptAPI routesreporting
AI + Engine
Agent simulation core
I built the deterministic engine and AI-capable mode around 900 generated agents, market-level cultural modifiers, category mechanics, archetype bias controls, homophily, and scale-free social pressure.
I built the benchmark comparison, raw and adjusted metrics, uncertainty-band checks, parameter provenance, sensitivity analysis, and mechanism ablation layer so outputs remain inspectable.
benchmarksuncertaintyablation53/53
Posture
Local-first and bounded
The default path is deterministic and free. Optional Anthropic mode is configurable, while local locking, persisted rate limiting, no-store API responses, and security headers protect the demo surface.
Eripion Product Coroner is designed to make assumptions explicit, expose segment-level failure modes, and turn uncertain launch decisions into testable hypotheses. The AI capability is intentionally secondary to the deterministic engine, so the project can show both modern AI integration judgment and reproducible simulation architecture without black-box prediction claims.
01Positioned Eripion Product Coroner as decision support, not a prediction engine or replacement for real user research.
02Kept deterministic local mode as the default so evaluations are reproducible, inspectable, and free to run.
03Added AI capability as a configurable support layer instead of letting a model become the single source of truth.
04Made uncertainty visible through raw, adjusted, and range-based outputs instead of presenting one confident score.
05Documented production gaps clearly: auth, workspace ownership, durable storage, and broader browser E2E coverage would come before multi-tenant launch.
