diff --git a/PEER_CLI_SCENARIOS.md b/PEER_CLI_SCENARIOS.md
index f89e984..8336ba1 100644
--- a/PEER_CLI_SCENARIOS.md
+++ b/PEER_CLI_SCENARIOS.md
@@ -22,9 +22,124 @@ for deterministic local runs; mDNS/macvlan remains an environment smoke path.
 | S12 | Transfer serving gates | A peer has a non-catalog, missing-sentinel, active-operation, or `local/` path request. | The serving peer declines metadata/data; covered by unit tests where timing is too small for a stable CLI race test. |
 | S13 | Exact transferred-file equality | Repeat small and large downloads, then compare every transferred regular file against its source with SHA-256 manifests. | Source and receiver manifests match exactly for each transferred file; no extra or missing files appear in the downloaded game root. |
 | S14 | Large multi-peer chunked download | `fixture-alpha/alienswarm` contains a renamed RAR `.eti` larger than 100 MB. A second peer downloads it, then a third peer downloads `alienswarm` from both peers. | The third peer's downloaded files match the source by SHA-256; `download-chunk-finished` events show the large `.eti` chunks coming from both peers with byte counts balanced within one chunk. |
+| S15 | Three-way version skew | Three peers advertise the same catalog game ID. Peer A has `version.ini=20250101`, peer B has `version.ini=20250201`, and peer C has `version.ini=20250301`; each version has distinguishable file contents. An empty client connects to all three and downloads the game with `install=false`. | `list-games` shows one row for the game with `peer_count=3` and `eti_game_version=20250301`. The `got-game-files` descriptor set and transfer source are peer C's newest version only; no chunks come from A or B. The receiver's `version.ini` and SHA-256 manifest match C exactly. |
+| S16 | Latest-version fanout with stale peers present | Peer A has an older version of a game. Peers B and C both advertise the same newest version with matching file manifests; use a large file when proving chunk split. | The aggregated row still counts all ready peers, but eligible transfer peers are only B and C. Large-file chunks may split between B and C; peer A contributes no manifest majority vote and no file chunks. |
+| S17 | Latest-version conflict rejection | Peer A has an older version. Peers B and C both advertise the newest version, but their latest-version file sizes conflict. | Validation considers only the latest-version peers, so A cannot rescue the majority. The download fails with `download-failed`, and no committed target `version.ini` remains. |
+
+## Version-Skew Contract
+
+Use S15-S17 to pin down what "newer" means when several peers have the same
+game ID:
+
+- Version comparison uses the eight-digit `version.ini` string, so use sortable
+  `YYYYMMDD` values in manual fixtures.
+- `list-games` aggregates by game ID. The game appears once; `peer_count`
+  counts all ready peers with that ID, including peers that only have older
+  versions.
+- The aggregated `eti_game_version` must be the newest ready version.
+- The descriptor set emitted to the download path, file-size validation, and
+  transfer planning are latest-only. Older-version peers may be queried by a
+  generic detail request, but their descriptors must not supply download
+  descriptors, majority votes, or chunks once a newer version exists.
+- If exactly one peer has the latest version, that peer is the only transfer
+  source. If several peers tie on the latest version, validation and chunk
+  fanout happen among that latest-version set only.
+- Capture proof with the `list-games` row, `got-game-files` descriptors,
+  `download-chunk-finished` source addresses, and source/receiver SHA-256
+  manifests.
+
+For a manual run, prefer a catalog game ID already served by the fixture lab,
+such as `cnc4`, then create temporary `just peer-cli-run` game roots with
+different `version.ini` contents. The existing alpha/bravo/charlie fixtures
+cover duplicate-source and shared-game cases, but not the three-version skew
+until a dedicated fixture or temporary games root is prepared.
 
 ## Run Log
 
+### 2026-05-18 - Full Matrix Manual Docker Pass
+
+- Build/setup: `just peer-cli-image` passed. Local `just peer-cli-build`
+  needed `RUSTC_WRAPPER=` because the host `kache` wrapper failed with a
+  read-only filesystem error; `RUSTC_WRAPPER= just peer-cli-build` passed.
+- Temporary skew/conflict fixtures were created under the ignored
+  `.lanspread-peer-cli/full-fixtures/` tree using `rar a -idq -m0` against
+  `/dev/urandom` payloads and then renaming the archives to `.eti`.
+  `find .lanspread-peer-cli/full-fixtures -name '*.eti' -exec unrar t -idq {} \;`
+  passed.
+- S1 startup scan: `just peer-cli-alpha` emitted `cli-started`,
+  `local-library-changed`, and `local-peer-ready`; `alienswarm`, `bf1942`, and
+  `ggoo` were `downloaded=true`, `installed=false`, `availability=Ready`.
+- S2 clean direct connect: with only alpha and bravo running, alpha connected to
+  bravo at `10.66.0.3:42776`; `wait-peers` returned `peer_count=1`, and
+  `list-peers` showed exactly one bravo peer with four games.
+- S3 clean remote aggregation: an empty `clean-s3-client` saw exactly alpha and
+  bravo. `list-games` showed `ggoo peer_count=2`; `alienswarm`, `bf1942`,
+  `bfbc2`, `cnc4`, and `cnctw` each had `peer_count=1`.
+- S4 single-source no-install: `full-empty-client` downloaded `bfbc2` from
+  bravo with `install=false`. Events included `got-game-files`,
+  `download-begin`, `download-finished`, and local `installed=false`. Host
+  verification: `diff -r crates/lanspread-peer-cli/fixtures/fixture-bravo/bfbc2
+  .lanspread-peer-cli/full-empty-client/games/bfbc2` passed and `local/` was
+  absent.
+- S5 auto-install: `full-empty-client` downloaded `cnctw` with default install.
+  Events included download finish, `install-begin`, and `install-finished`;
+  `local/fixture-payload.txt` existed. Host verification diffed the downloaded
+  files against `fixture-bravo/cnctw` excluding `local/` and `.lanspread.json`.
+- S6 manual install/uninstall: after S4, `install bfbc2` created `local/` and
+  marked `installed=true`; `uninstall bfbc2` removed `local/` and preserved the
+  downloaded root files. Host verification diffed the preserved files against
+  `fixture-bravo/bfbc2` excluding `.lanspread.json`.
+- S7 duplicate-source download: `full-empty-client` downloaded shared `ggoo`
+  from alpha/bravo with `install=false`. Chunk events used alpha for
+  `version.ini` and bravo for `ggoo.eti`; host `diff -r` matched both
+  `fixture-alpha/ggoo` and `fixture-bravo/ggoo`.
+- S8 ambiguous metadata rejection: `full-s8-a` and `full-s8-b` both advertised
+  `ggoo` version `20260101` but with different `.eti` sizes (`1,048,746` and
+  `2,097,323` bytes). The client saw `peer_count=2`, then `download ggoo`
+  emitted `download-failed`; no target `ggoo/version.ini` was committed.
+- S9 missing game: `download does-not-exist` emitted `no-peers-have-game` and
+  returned a command error; `.lanspread-peer-cli/full-empty-client/games` had no
+  `does-not-exist` directory.
+- S10 shutdown cleanup: alpha saw bravo before shutdown with one remote peer and
+  bravo-only remote games. After bravo `shutdown`, alpha emitted `peer-lost`;
+  `list-peers` returned `[]` and `list-games` returned an empty remote list.
+- S11 same identity reconnect: restarting bravo reused peer ID
+  `019e347d901e70c19adf5b9fd313fce4` at new address `10.66.0.3:41764`.
+  Alpha `list-peers` showed exactly one bravo entry at the new address.
+- S12 transfer serving gates: this remains covered by unit tests because the
+  CLI cannot stably race raw transfer requests against non-catalog, missing
+  sentinel, active-operation, and `local/` path states. `RUSTC_WRAPPER= just
+  test` passed, including `local_download_available_gates_on_catalog_operation_and_sentinel`,
+  `get_game_response_respects_serve_gates`,
+  `file_transfer_dispatch_respects_serve_gates`, and
+  `local_relative_paths_are_never_transferable`.
+- S13 exact transferred-file equality: the S4 small transfer and S14 large
+  transfer both passed host `diff -r` against the original source game
+  directories, proving exact file equality beyond event flow.
+- S14 large multi-peer chunked download: `full-empty-client` first downloaded
+  `alienswarm` from alpha and diffed cleanly against `fixture-alpha/alienswarm`.
+  A fresh `full-s14-client` then saw `alienswarm peer_count=2` and downloaded
+  from both alpha and `full-empty-client`. Large `.eti` chunk totals were
+  `67,108,864` bytes from alpha and `58,721,049` bytes from the staged peer,
+  balanced within one `32 MiB` chunk. Final host `diff -r` against
+  `fixture-alpha/alienswarm` passed.
+- S15 three-way version skew: peers A/B/C advertised `cnc4` versions
+  `20250101`, `20250201`, and `20250301`. The client saw one row with
+  `peer_count=3` and `eti_game_version=20250301`; all chunks came only from C
+  at `10.66.0.4:60290`. Host `diff -r` against C passed.
+- S16 latest-version fanout with stale peer present: A advertised stale
+  `20250101`; B/C both advertised latest `20250301` with a `134,217,906` byte
+  `.eti`. The client saw `peer_count=3`; chunks came only from B/C
+  (`67,108,873` and `67,109,042` bytes respectively), with stale A contributing
+  zero. Host `diff -r` matched both B and C.
+- S17 latest-version conflict rejection: A advertised stale `20250101`; B/C
+  both advertised latest `20250301` but with conflicting `.eti` sizes
+  (`1,048,748` and `2,097,325` bytes). The client saw `peer_count=3` and latest
+  `20250301`, then `download cnc4` emitted `download-failed`; no target
+  `cnc4/version.ini` was committed.
+- Gates after manual runs: `just fmt`, `RUSTC_WRAPPER= just test`, and
+  `RUSTC_WRAPPER= just clippy` passed.
+
 ### 2026-05-17 - Exact Transfer And Large Multi-Peer Chunking
 
 - Fixture update: `fixture-alpha/alienswarm/alienswarm.eti` was rebuilt with