An experiment has no name: three runner scripts are the interface; no [project.scripts], no list

[dim-sous]

What is wrong

There is no console script and no __main__.py (pyproject.toml has no [project.scripts]). Work is launched as:

python comparison/run_v5_comparison.py [--full | -n N | ...]
python comparison/run_phase4_gate.py   [--hours H --out PATH --skip-validate]
python comparison/run_synthetic_day.py

Those three filenames are the complete set of questions the repo makes askable. Each script builds its own nine parameter objects, its own strategy list (STRATEGY_FACTORIES in run_v5_comparison.py; greek_milp_bidding is not in it and has its own runner), and its own output naming. An experiment is literals at a call site — it has no name, so it cannot be listed, reused by a second figure, parameterised from the CLI, or recorded. This is RCWS #55, verbatim.

The cost is the one the developer named: "I am forgetting commands and capabilities." A CLI demands recall; a list offers recognition. Every capability that has to be remembered is, functionally, a capability that does not exist.

What it should be

RCWS's platform, sized for one simulator:

• A registry of named experiments with a one-line finding each, discoverable in code — the three runners become entries in it, not the interface.
• One CLI, a small verb set. RCWS carries seven (list run sweep show compare rerun emit-config); BESS probably needs four to start: list, run, show, compare. list is the one that pays immediately.
• Dotted overrides: run NAME --set path=value, read off the config schema from run_simulation takes 15 positional parameters and 9 config dataclasses from one file, with no runtime schema — every caller is welded to the core's shape #5, refusing unknown keys.
• The runtime guard: an experiment declares its expected minutes; long ones refuse without --force.
• Names are claims: a command names the ARTEFACT it produces, never the activity (RCWS [A8]).

Done when

• uv run bess list prints every registered experiment with its note and expected runtime.
• uv run bess run NAME [--set …] [--name …] writes a record (A run does not describe itself: results/ holds 27 files named by version, with no config, seed, conditions or git stamp — two runs of one version overwrite each other #6).
• The three scripts in comparison/ are gone or reduced to thin registry entries.
• [project.scripts] declared.
• Proof: run of the canonical synthetic day reproduces the current runner's numbers.

Phase 3 of the platform epic. Depends on #5 and #6. The Lab (phase 4) is its own epic, opened when this lands — as RCWS #70 was.