docs: two Qwen 2.5 Coder 14B transcripts of find-and-kick, the local model evidence

CHANGELOG.md

@@ -851,6 +851,13 @@ that were not true no longer are.
   HTTP snapshot alternative that needs none. Nobody who is not pointing quackd at a real
   Microduck needs it, and the H.264 and the ICE inside it have never met a duck.
 
+### Local model evidence
+
+- Two transcripts of `find-and-kick` piloted by **Qwen 2.5 Coder 14B on LM Studio**, seeds
+  5 and 6, land in `docs/assets/transcripts/` with a table reading them in
+  `docs/local-llms.md`, from the contributor whose memory feature they were recorded for.
+  The README and `local-llms.md` no longer say "no transcript in this repository".
+
 ## [0.6.0] — 2026-09-04
 
 A run stops starting from nothing. Every release so far built a pilot with no past: the

README.md

@@ -204,7 +204,7 @@ Version 0.7, simulator and mocks. What has been built, and how far each piece ha
 | MCP server (`quackd serve-mcp`) | ✅ Claude Code and Claude Desktop, fleets with `--robots` (eight `robot_*` tools, tested in process against the simulator and the mocks), no Claude Desktop session on record |
 | Memory between runs (`quackd memory`, `remember`) | ✅ one JSONL file per `adapter:backend`, notes and run outcomes into the next prompt, tested end to end offline, 🧪 the `remember` tool itself exercised by one local model on one machine and by no cloud model ([docs/memory.md](docs/memory.md)) |
 | Providers: anthropic, openai, gemini, grok, fake | ✅ implemented, tested offline, real model hero recording pending an API key |
-| Local models (Ollama, vLLM, llama.cpp, LM Studio, any OpenAI compatible server) | ✅ implemented and tested against the OpenAI wire format, 🧪 one live run by a contributor (Qwen 2.5 Coder 14B on LM Studio, two seeds), no transcript in the repo, more welcome |
+| Local models (Ollama, vLLM, llama.cpp, LM Studio, any OpenAI compatible server) | ✅ implemented and tested against the OpenAI wire format, 🧪 two live runs by a contributor (Qwen 2.5 Coder 14B on LM Studio, seeds 5 and 6), never on this machine, transcripts in [`docs/assets/transcripts/`](docs/assets/transcripts/), more welcome |
 | Flock mode (multiple cooperating robots, sim2d) | ✅ deterministic auction and bus, one planner LLM call at most, ground truth checked in tests, 🧪 experimental and simulator only |
 | Heterogeneous flock (a Reachy Mini head and a Microduck, sim2d) | ✅ `reachy-spots-duck-kicks` 10 of 10 seeds, capability aware auction, the spotter judges from its own frames, ground truth vetoes, 🧪 simulator only |
 | LAN discovery (`quackd discover`, `quackd announce`, `quackd[lan]`) | ✅ record format and both commands on fakes in the suite, 🧪 real zeroconf exercised once on one machine, never between two ([docs/lan.md](docs/lan.md)) |
@@ -524,7 +524,7 @@ The physics simulator costs what physics costs. Measured here on one Windows lap
 - `grab` is open loop upstream and unreliable here on purpose. `fetch` says so in its file.
 - A manifest can be smaller than the robot. The LeRobot arm's `real` backend claims no camera and no `pick` until it connects, and even then `pick` appears only when a policy object was injected in code. A rosbridge base over `ws` has no camera verbs unless the address names an image topic, and a ToddlerBot has no `move` unless a walk checkpoint is staged.
 - Default model IDs for OpenAI, Gemini and Grok were not verified at release.
-- Local model quality is unmeasured. The JSON text fallback and the one retry exist because small models often miss native tool calls. One contributor reports `find-and-kick` succeeding against Qwen 2.5 Coder 14B through LM Studio on two seeds, and no transcript from it is in this repository.
+- Local model quality is unmeasured. The JSON text fallback and the one retry exist because small models often miss native tool calls. One contributor ran `find-and-kick` against Qwen 2.5 Coder 14B through LM Studio on two seeds, both successes, one of them reading the previous run's memory; the two transcripts are in [`docs/assets/transcripts/`](docs/assets/transcripts/) and read in [docs/local-llms.md](docs/local-llms.md).
 - Flock mode is simulator only, ships two choreographies and exactly two roles (spotter and kicker), and knows only the Microduck and the Reachy Mini, so an Open Duck cannot join one yet. Separation uses sim ground truth, and two robots share no frame of reference on hardware.
 - LAN discovery and the MQTT bus have each been exercised once, on one machine. Nothing has crossed to a second machine, the MQTT bus is a library with no `--bus` flag, and a flock across machines also needs a clock across machines, which does not exist yet.
 

docs/assets/README.md

@@ -10,6 +10,8 @@
 | `flock.gif` | Flock mode: three ducks split the search, auction the kick, the closest one takes it. Left: the shared world. Right: the claimant's camera. | `quackd record flock-kick --provider fake --seed 3` — the **scripted planner and deterministic coordinator**, no LLM in the loop (docs/flock.md). |
 | `hetero.gif` | The 0.4 headline demo: a Reachy Mini head (the slate square on the wall) spots the ball and judges the kick from its own camera while a Microduck walks in and kicks it, two bodies under one contract. Left: the shared world. Right: the acting robot's camera. | `quackd run reachy-spots-duck-kicks --provider fake --seed 3 --gif-size 320` — **scripted pilots and a deterministic coordinator**, zero planner LLM calls. That run ended `success`: the spotter judged the ball moved 0.51 m and the simulator's ground truth agreed. `quackd record` is pinned to `microduck:sim2d`, so this one comes from `run`, which also writes `run.gif`. |
 | `open-duck.gif` | The 0.5 headline body: an Open Duck Mini v2 finds the ball and walks up to it, because it has no kick. Left: the world from above. Right: the duck's camera. | `quackd run open-duck-scout --robot open_duck:sim2d --provider fake --seed 3 --gif-size 320` — the **scripted pilot**, not an LLM, like every other asset here. |
+| `transcripts/qwen2.5-coder-14b-lmstudio-find-and-kick-seed6-memory-read.jsonl` | A real local model piloting `find-and-kick`: Qwen 2.5 Coder 14B through LM Studio, seed 6, success in 8 steps, with the previous run's episode in the prompt and no `remember` call. | `quackd run find-and-kick --provider lmstudio --model qwen/qwen2.5-coder-14b --seed 6`, copied from `runs/<timestamp>/transcript.jsonl` with the contributor's home directory removed from `duck_path`; `memory.path` is relative because the runs used `--memory-dir` next to the checkout, not `~/.quackd`. Read in [docs/local-llms.md](../local-llms.md). |
+| `transcripts/qwen2.5-coder-14b-lmstudio-find-and-kick-seed5-remember.jsonl` | The same model and duck once `remember` sat in strategy step 5: seed 5, success in 4 steps, one `remember` call with a fact from the verb results. | Same command with `--seed 5`, same scrubbing. |
 | `contributors.svg` | The circle of faces under *Contributing* in the README: everyone who has contributed, humans only, ordered by lines added, every avatar clipped to the same circle. | Generated by [`contributors.py`](contributors.py) from the GitHub contributors API, avatars inlined so the file needs no network. Regenerated by `.github/workflows/contributors.yml` on every push to `main` and weekly, committed only when it changes. |
 | *(missing)* `reachy-spotter.gif` | **Not recorded yet.** A head with no legs finding the ball with its gaze, the simplest illustration that the verb list follows the body. | Would be `quackd run reachy-spotter --provider fake --seed 3`. |