Skip to content

Shunt submission hand-off — RouterArena (sub_10) - #205

Open
KookaS wants to merge 3 commits into
RouteWorks:mainfrom
KookaS:shunt-sub_10-submission
Open

Shunt submission hand-off — RouterArena (sub_10)#205
KookaS wants to merge 3 commits into
RouteWorks:mainfrom
KookaS:shunt-sub_10-submission

Conversation

@KookaS

@KookaS KookaS commented Sep 4, 2026

Copy link
Copy Markdown

Shunt (https://github.com/KookaS/shunt, @KookaS, open-source) routes every
prompt through its real session_cascade router engine (pinned Shunt commit
7ee66af): each prompt is passed to RouterEngine.decide, then mapped to the
chosen arena model. RouterArena's stateless single-turn / no-verification regime
provides no verified-outcome channel (no repo to re-test), so Shunt's
verified-failure escalation ladder never fires and the router deterministically
stays on its cheapest live model — deepseek/deepseek-v4-flash (809/809
sub_10, 420/420 robustness).

This is an honest deployability probe: it exercises Shunt's shipped routing
machinery end-to-end against RouterArena's harness, and its score should be read
as deepseek-v4-flash's standalone accuracy/cost under that harness — not as
evidence of Shunt's routing value, which requires the verified-outcome channel
RouterArena does not provide. No RouterArena data was used to train or tune
anything. Reproducibility instructions are in REPRODUCE.md.

@KookaS

KookaS commented Sep 4, 2026

Copy link
Copy Markdown
Author

/evaluate

…ion reproducible + CI-green

- router module: SHUNT_PIN constant + seed-dir-first resolution
  (router_inference/router/shunt_seed/ preferred, env/artifact root fallback),
  typed _PRIOR (fixes mypy Collection[str] index errors on 3.9-3.12), no
  unused _import_shunt assignment (ruff F841); import registered in __all__
  (ruff F401).
- shunt_bench_prior: _embed_many callable narrowing (mypy 'None' not
  callable), drop zip(strict=) for py3.9 mypy, SPDX header, no codespell hits.
- bundle router_inference/router/shunt_seed/ (sample + embeddings verbatim,
  results filtered to the menu models) + REPRODUCE.md + .codespellrc so a
  clean checkout at the pin regenerates byte-identical predictions.
- predictions files untouched.
…nch_prior)

Prediction choices are now all deepseek/deepseek-v4-flash, produced by Shunt's
real session_cascade engine at the pinned commit; answers reused from the
recorded cache (no re-inference).
@KookaS

KookaS commented Sep 6, 2026

Copy link
Copy Markdown
Author

/evaluate

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant