diff --git a/doc/leios_tests.md b/doc/leios_tests.md
new file mode 100644
index 000000000..9319b8e14
--- /dev/null
+++ b/doc/leios_tests.md
@@ -0,0 +1,105 @@
+# Running Leios Tests
+
+The Leios testrun exercises the experimental Leios feature of `cardano-node` on the
+experimental Dijkstra era. It is a regular regression run with a different setup, plus the
+dedicated Leios tests in `cardano_node_tests/tests/test_leios_blocks.py`.
+
+## The Leios Setup
+
+The setup lives in [`runner/env_leios`](https://github.com/IntersectMBO/cardano-node-tests/blob/master/runner/env_leios), except `NODE_REV`, which the launcher and the CI workflow default:
+
+| Variable              | Value                                  | Why                                                     |
+| --------------------- | -------------------------------------- | ------------------------------------------------------- |
+| `NODE_REV`            | `leios-prototype`                      | The node branch that has Leios.                         |
+| `TESTNET_VARIANT`     | `leios_fast`                           | Starts in Dijkstra, block rate slow enough to see EBs.  |
+| `PROTOCOL_VERSION`    | `12`                                   | Dijkstra protocol version.                              |
+| `ENABLE_TX_FIREHOSE`  | `true`                                 | Tx load generator, needed to fill the mempool.          |
+| `TX_TPS`              | `15`                                   | Tx rate ceiling of the load generator.                  |
+| `DESELECT_FROM_FILE`  | `scripts/deselected_leios_tests.txt`   | Skips the tests that are known to fail here.            |
+
+## Running in CI
+
+Workflow [`04 Regression tests with Leios`](https://github.com/IntersectMBO/cardano-node-tests/actions/workflows/regression-leios.yaml)
+triggered manually from the `Actions` tab (`Run workflow`) or with `gh`:
+
+```sh
+gh workflow run "04 Regression tests with Leios" --ref master
+```
+
+Inputs:
+
+| Input                       | Default            | Description                                            |
+| --------------------------- | ------------------ | ------------------------------------------------------ |
+| `node_rev`                  | `leios-prototype`  | `cardano-node` revision.                               |
+| `cli_rev`                   | (empty)            | `cardano-cli` revision, optional.                      |
+| `allow_unstable_error_msgs` | `true`             | Let tests pass with unstable error messages.           |
+| `skip_deselect`             | `false`            | Run also the tests that are known to fail on Leios.    |
+
+## Running Locally
+
+### Full Leios Regression Run
+
+Run the commands below from the repo root.
+
+Inside a container. Preferred, as no local Nix is needed.
+
+```sh
+./runner/runc.sh -- ./scripts/test_leios.sh
+```
+
+Or directly on the host. Local Nix is needed.
+
+```sh
+./scripts/test_leios.sh
+```
+
+Any variable from `runner/env_leios` can be overridden, but **only as a `VAR=VALUE`
+argument** - the env file is loaded on top of the environment, so an exported value would be
+overwritten:
+
+```sh
+./scripts/test_leios.sh TX_TPS=10
+```
+
+The exceptions are `NODE_REV`, `MARKEXPR` and `ALLOW_UNSTABLE_ERROR_MESSAGES` - there an
+already exported value wins, so the node branch can be switched without touching the setup:
+
+```sh
+NODE_REV=my-leios-branch ./scripts/test_leios.sh
+```
+
+To run also the tests that are known to fail on Leios, clear the deselect file (or point it
+to your own list):
+
+```sh
+./scripts/test_leios.sh DESELECT_FROM_FILE=
+```
+
+Only `VAR=VALUE` arguments are accepted; anything else is rejected with a usage message.
+Results land in `run_workdir`.
+
+## Analyzing Failures
+
+### Locally
+
+In Claude Code:
+
+```text
+/analyze-failures
+```
+
+### In CI
+
+On a failed testrun the workflow runs the analysis automatically and surfaces it in
+
+* the **run summary** of the workflow run, and a foldable `Preliminary failure analysis`
+  group in the log of the `Read failure analysis into env` step,
+* `failure_analysis.md` in the **`testrun-files`** artifact.
+
+## Known Failures
+
+`scripts/deselected_leios_tests.txt` lists the tests that are known to fail in this setup,
+grouped by cause, each group with the upstream issue link (missing PlutusV2 cost model in
+Dijkstra genesis, `transaction assemble` rejecting a Dijkstra `TxWitness`, missing required
+signers in the Plutus script context, new Dijkstra protocol params, mandatory Leios BLS
+signing keys, ...).
diff --git a/src_docs/source/how-tos/540_leios_tests.rst b/src_docs/source/how-tos/540_leios_tests.rst
new file mode 100644
index 000000000..0abc43cb2
--- /dev/null
+++ b/src_docs/source/how-tos/540_leios_tests.rst
@@ -0,0 +1 @@
+.. mdinclude:: ../../../doc/leios_tests.md