Roadmap discussion for 2026 and 2027

This issue will contain some ideas for the roadmap for this year and the next one.

# 2026

## Ongoing development

* #1626

* #1625

## TODO

* #1623

* **Agentic preview stabilization:** add symbolic transaction injection, event and log inspection, trace queries, function avoidance, capability discovery, cancellation and structured error reporting. Record every agent-directed campaign change so that an agent-assisted run can be audited and reproduced. [Q4]

* **Continuous fuzzer benchmarking:** maintain reproducible campaigns measuring coverage, time to first failure, number of distinct failures, executions per second, memory consumption, solver cost, corpus growth and reproducibility. Compare Echidna versions and representative EVM fuzzers on synthetic and real-world contracts. [Q4]

* **Basic Testing Hooks:** we want to offer so basic hooks to allows users to run code before and after sequences, and such. [TBD]

# 2027

* **Extended agentic support:** evolve workers into specialized but cooperating roles, such as a coverage analyst, sequence planner, symbolic or formal-verification explorer, corpus curator and failure triager. Let agents propose campaign strategy changes within explicit permissions and resource budgets. Keep the interface model-agnostic, replayable and fully optional so that Echidna remains usable without an AI model.

* #1627

* **Improved performance using the hevm concrete-execution branch:** use the faster concrete path for ordinary transactions while retaining seamless fallback to symbolic execution and advanced cheatcode handling. Differentially test the fast and reference paths for identical state, logs, return data, reverts, gas behavior and exceptional halting.

* **Dependency-aware transaction-sequence generation:** use static and dynamic storage read/write sets, external calls, value flow, events and calldata dependencies to construct useful transaction sequences. Preserve dependencies while splicing, crossing over, extending and shrinking sequences.

* **Adaptive power scheduling:** distribute mutation energy according to rare edges, difficult branches, state novelty, property progress, execution cost and per-worker productivity. Allow workers to use complementary schedulers and exchange discoveries during a campaign.

* #1624

* **Stateful coverage metrics:** supplement instruction and branch coverage with storage-state transitions, call-graph transitions, event combinations, balance-flow states and property-specific progress. Use these metrics for scheduling without treating every raw storage value as a unique state.

Start by reviewing the roadmap items and linked issues #1623, #1624, #1625, #1626, and #1627, then compare their scope with the proposed agentic, fuzzing, testing-hook, and performance work. This issue is done only when a specific roadmap item has been turned into a scoped implementation issue with clear acceptance criteria.

Written by the indexing model from the issue text.