# CATALYST 2.0 Product Strategy

Date: May 13, 2026

CATALYST 2.0 should be positioned as an enterprise research-to-decision operating system, not a theorem prover or miracle science console. The strongest commercial shape is: generate auditable candidate packets, ground them with Browser Run evidence, preserve them in Cloudflare artifacts, and require explicit approval gates before any real-world mutation.

## Market Signal

- AI infrastructure power is a board-level constraint. The IEA estimates global data center electricity consumption at about 415 TWh in 2024, or roughly 1.5% of global electricity use, and projects about 945 TWh by 2030 in its Base Case. Accelerated servers, mainly driven by AI, are projected to grow electricity consumption at about 30% annually. Source: https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
- Materials informatics is still small enough for a focused wedge but growing quickly. Grand View Research estimates global material informatics at USD 134.6M in 2023 and USD 390.8M by 2030. Research and Markets cites AI in materials discovery at USD 0.97B in 2026 and USD 2.77B by 2030. Sources: https://www.grandviewresearch.com/industry-analysis/material-informatics-market-report and https://www.researchandmarkets.com/reports/6227043/ai-in-materials-discovery-market-report
- The research frontier is moving from passive prediction to autonomous, workflow-aligned materials science. Recent 2026 work frames the next phase around active learning, RAG, data provenance, validation, uncertainty, reproducibility, and experimental feedback. Sources: https://arxiv.org/abs/2601.00742 and https://arxiv.org/abs/2605.09624

## Ideal Customer Profile

1. Advanced materials R&D teams
   - Battery, semiconductor, aerospace, catalyst, coating, and specialty chemical groups.
   - Pain: expensive wet-lab iteration, fragmented evidence, candidate overload, weak go/no-go discipline.
   - CATALYST wedge: MaterialForge packets with ranked candidates, constraints, procurement notes, and falsifiable bench gates.

2. AI infrastructure and platform operations
   - Platform, cloud cost, sustainability, and inference operations teams.
   - Pain: power availability, carbon reporting, model routing, GPU utilization, latency/cost tradeoffs.
   - CATALYST wedge: GridForge plans that turn model class, SLA, watts, and carbon intensity into measured routing tactics.

3. Government labs and enterprise research offices
   - National labs, defense innovation programs, corporate advanced R&D, and regulated science teams.
   - Pain: auditability, reproducibility, approval gates, and vendor-neutral evidence preservation.
   - CATALYST wedge: Cloudflare-native evidence ledgers with Browser Run grounding, R2 packet storage, Workers AI agent reasoning, and Durable Object sessions.

## Product Doctrine

- Do say: "CATALYST shortens the path from hypothesis to falsifiable experiment packet."
- Do say: "The system is designed for evidence triage, candidate ranking, and governed agent execution."
- Do not say: "CATALYST proves unsolved mathematics, guarantees stable molecules, or replaces lab validation."
- Keep the breakthrough language aimed at possibilities: better search, faster experiment loops, lower power waste, more auditable agents.

## Use CATALYST Correctly

1. Frame a bounded decision: a material property, power objective, spectral triage target, or agent task.
2. Generate an auditable packet: shortlist, evidence, assumptions, risks, and validation gates.
3. Ground external claims: use Browser Run markdown and screenshots to capture source evidence.
4. Validate externally: run lab protocols, measurement windows, confounder reviews, or human approval gates.
5. Promote only what survives evidence review into production workflows.

## Cloudflare Implementation Notes

- Browser Run Quick Actions are appropriate for fast markdown and screenshot grounding. Cloudflare documents markdown and screenshot Quick Actions and requires a custom token with `Browser Rendering - Edit`. Source: https://developers.cloudflare.com/browser-run/quick-actions/
- Wrangler is correct for storing secrets. `wrangler secret put` creates and deploys a Worker version with the encrypted secret. Source: https://developers.cloudflare.com/workers/configuration/secrets/
- Account-wide/global keys should be treated as temporary legacy fallback. Prefer a scoped Browser Run API token for production. Cloudflare’s current API-token flow recommends creating custom tokens with explicit permission groups. Source: https://developers.cloudflare.com/fundamentals/api/get-started/create-token/

## 2.0 Product Surfaces

- MaterialForge Pro: the first impression and paid wedge.
- GridForge: AI power operations and cost/carbon optimization.
- Catalyst Agent: Workers AI reasoning with Browser Run grounding, R2 packets, and approval-gated connectors.
- ExoSignal: evidence and confounder triage for spectral anomalies.
- Reality Foundry: hypothesis compiler for product and research strategy.
- Research Validation: archive of benchmark simulations, clearly separated from production claims.