docs: present enterprise scope, runtime map, and QA history

README.md

@@ -12,22 +12,49 @@ viewer: false
 
 # MartechBench
 
-**A commercially licensed, stateful RL environment for marketing-automation agents.**
+### Enterprise work, modeled for reinforcement learning.
 
-MartechBench is designed for research teams evaluating and improving agents that operate business software. Agents interpret workplace requests, resolve them through available evidence, change connected product state, and observe downstream effects. Deterministic verifiers score durable business outcomes and preservation of neighboring behavior.
+A commercially licensed RL environment for frontier labs and research teams training agents to operate real business workflows. In the enterprise automation research category alongside AutomationBench, MartechBench focuses on the connected marketing stack: durable state, dependencies, realistic workplace context, and measurable downstream consequences.
 
-## Commercial access
+![MartechBench scope and research use](overview.png)
+
+**1,400+ tasks · 400+ tools · Deterministic verifiers · Multiple valid solutions per task**
+
+[**Discuss licensing with ReadySolve →**](https://readysolve.sh/)
+
+## A change in one system changes the next
+
+Agents must resolve ambiguous requests, find the current owner, and change upstream state in dependency order. Success depends on what happens downstream—and on preserving the workflows around it.
+
+The runtime models connected products, persistent records, permissions, versioned configuration, and time. Tasks demand investigation, coordination, and execution across system boundaries. Verifiers inspect the business state the agent leaves behind; a plausible final answer is not enough.
+
+![Illustrative map of connected runtime capabilities](capability-network.png)
 
-This repository is a product information card. It contains no runnable environment, task samples, runtime source, or downloadable corpus. Product access follows a commercial conversation and is granted deliberately to licensed buyers. There is no public trial corpus or community edition.
+## Built for outcomes, with room for different solutions
 
-For commercial inquiries, [open a discussion](https://huggingface.co/datasets/HumanRuntime/MartechBench/discussions). Share only a brief description of your research use; do not post confidential research details or personal contact information publicly.
+Each task supports multiple valid solutions. Deterministic verifiers grade durable business outcomes rather than a prescribed tool sequence or a single golden configuration. Partial credit captures meaningful progress; protected neighboring behavior and held-out probes test whether the workflow actually works.
 
-## Research use and delivery
+## Quality improves through iteration
 
-The environment is intended for model comparison, capability evaluation, and agent post-training experiments. One versioned Harbor-native buyer release combines a selected taskset with its pinned runtime and verifier artifacts. Additional framework compatibility is documented per release after validation.
+Every task is versioned. Review and repair produce traceable successors, so research teams can pin the task revisions they use and reproduce them later.
 
-Licensed buyers receive an authenticated, commit-pinned Hugging Face download for execution in their own infrastructure. Release manifests identify the purchased task revisions, environment, verifiers, and artifact hashes. Included build materials and container artifacts support auditing and reproduction.
+![Distribution of retained task versions](qa-revisions.png)
+
+The chart shows retained revision counts in the 5 October 2026 corpus snapshot: **89.3% of tasks have multiple versions**. It documents sustained iteration; revision counts are not a count of defects repaired or an independent quality certification.
+
+## Ready for your research infrastructure
+
+| Research use | Delivery |
+| --- | --- |
+| Capability evaluation and model comparison | One canonical Harbor-native buyer release |
+| RL and post-training experiments | Execution in buyer-controlled infrastructure |
+| Reproducible experiments | Pinned task revisions, runtime, verifiers, and artifact hashes |
+| Internal auditing and retention | Licensed build materials and container artifacts |
+
+Corpus scope is approximately 1,400 tasks; the current runtime exposes 426 tools. Exact purchased membership and framework compatibility are specified in each licensed release. Additional integrations consume the same release rather than creating separate editions.
+
+## Commercial access
 
-## Proprietary research asset
+MartechBench is proprietary research infrastructure, available through a commercial agreement. **[Reach out through readysolve.sh →](https://readysolve.sh/)**
 
-The task corpus and software are proprietary and supplied under a separate commercial agreement. Public visibility of this information card grants no rights to the environment. Internal task production, QA workflows, private rollout evidence, and release curation are excluded from delivery.
+This public page contains product information and illustrative visuals only. Task data, runtime code, and verifiers are available to authorized buyers. Task production, internal QA machinery, private rollout evidence, and release curation remain internal.