Issues

tooling: give acceptance scratch copies and environments bounded lifetimes

Alberto-Codes/typevet#467 · opened

on Oct 29, 2026

2

fix(llama_cpp): give the async autoload reads the early-close retry the sync reads have

Alberto-Codes/typevet#465 · opened

on Oct 25, 2026

build(deps): drop the GHSA-xcq7-p3pf-p88h audit ignore once google-adk allows opentelemetry-api>=1.45.1

Alberto-Codes/typevet#464 · opened

on Oct 25, 2026

process: adopt shared delegation, align sizes and retire fit-label routing

Alberto-Codes/typevet#459 · opened

on Oct 21, 2026

3

ci(release): release-please opens a downgrade release PR in the run that creates the draft release

Alberto-Codes/typevet#458 · opened

on Oct 21, 2026

3

docs: add shared Copilot review baseline with typevet invariants

Alberto-Codes/typevet#456 · opened

on Oct 21, 2026

1

docs(maintainers): port the judgevet triage how-to (priority scale, statuses, searches, release cycle)

Alberto-Codes/typevet#454 · opened

on Oct 21, 2026

2

feat(judgment): add a Mistral V7 [INST] framing and smoke Mistral Small 4 with CPU-offloaded experts

Alberto-Codes/typevet#453 · opened

on Oct 10, 2026

test(live): smoke Nemotron 3.5 Lightning 30B-A3B as the judge on the local llama.cpp router

Alberto-Codes/typevet#452 · opened

on Oct 10, 2026

test(live): smoke Granite 4.2 30B as the judge on the local llama.cpp router

Alberto-Codes/typevet#451 · opened

on Oct 10, 2026

test(live): smoke Qwen3.5-27B as the judge on the local llama.cpp router

Alberto-Codes/typevet#450 · opened

on Oct 10, 2026

test(live): smoke Gemma 4 26B-A4B as the judge on the local llama.cpp router

Alberto-Codes/typevet#449 · opened

on Oct 10, 2026

research(eval): shortlist the next judge model families for the 24 GB llama.cpp rig

Alberto-Codes/typevet#448 · opened

on Oct 10, 2026

2

test(live): smoke Qwen3.8-27B as the vLLM judge on a rented H100

Alberto-Codes/typevet#446 · opened

on Oct 10, 2026

1

epic: run Qwen3.8-27B as a typevet judge on llama.cpp now and on vLLM later

Alberto-Codes/typevet#442 · opened

on Oct 10, 2026

1

perf(judgment): a multi-question judgment costs the sum of single calls on llama.cpp

Alberto-Codes/typevet#428 · opened

on Oct 9, 2026

1

docs(how-to): link the judgevet agent-harness page from the provider how-to

Alberto-Codes/typevet#427 · opened

on Oct 9, 2026

2

research(eval): score nimble, Gemma 4 and Jev as Claude Code dispatch routers on shadow-logged dispatches

Alberto-Codes/typevet#422 · opened

on Oct 9, 2026

1

research: position typevet against Imajev, Plumb and decider with versioned evidence

Alberto-Codes/typevet#418 · opened

on Oct 3, 2026

7

feat(evals): export a doodle duel replay page with round cards and reliability tables

Alberto-Codes/typevet#417 · opened

on Oct 3, 2026

test(live): add CLM-8B on clm-serve as a fourth judge in the DIFrauD wording study

Alberto-Codes/typevet#416 · opened

on Oct 3, 2026

research(adapters): decide whether a contrastive dual-encoder scorer belongs behind JudgmentPort

Alberto-Codes/typevet#415 · opened

on Oct 3, 2026

ci(docs): the plain-English gate skips docs/explanation and docs/tutorials

Alberto-Codes/typevet#408 · opened

on Oct 2, 2026

1

ci(examples): decide and, if feasible, schedule a weekly live example canary

Alberto-Codes/typevet#399 · opened

on Oct 2, 2026

3

test(live): rerun the nimble judge evolution with the Qwen reflector on a second card or a pod

Alberto-Codes/typevet#383 · opened

on Oct 1, 2026

1

Navigation Menu

Repository navigation

Issues

tooling: give acceptance scratch copies and environments bounded lifetimes

fix(llama_cpp): give the async autoload reads the early-close retry the sync reads have

build(deps): drop the GHSA-xcq7-p3pf-p88h audit ignore once google-adk allows opentelemetry-api>=1.45.1

process: adopt shared delegation, align sizes and retire fit-label routing

ci(release): release-please opens a downgrade release PR in the run that creates the draft release

docs: add shared Copilot review baseline with typevet invariants

docs(maintainers): port the judgevet triage how-to (priority scale, statuses, searches, release cycle)

feat(judgment): add a Mistral V7 [INST] framing and smoke Mistral Small 4 with CPU-offloaded experts

test(live): smoke Nemotron 3.5 Lightning 30B-A3B as the judge on the local llama.cpp router

test(live): smoke Granite 4.2 30B as the judge on the local llama.cpp router

test(live): smoke Qwen3.5-27B as the judge on the local llama.cpp router

test(live): smoke Gemma 4 26B-A4B as the judge on the local llama.cpp router

research(eval): shortlist the next judge model families for the 24 GB llama.cpp rig

test(live): smoke Qwen3.8-27B as the vLLM judge on a rented H100

epic: run Qwen3.8-27B as a typevet judge on llama.cpp now and on vLLM later

perf(judgment): a multi-question judgment costs the sum of single calls on llama.cpp

docs(how-to): link the judgevet agent-harness page from the provider how-to

research(eval): score nimble, Gemma 4 and Jev as Claude Code dispatch routers on shadow-logged dispatches

research: position typevet against Imajev, Plumb and decider with versioned evidence

feat(evals): export a doodle duel replay page with round cards and reliability tables

test(live): add CLM-8B on clm-serve as a fourth judge in the DIFrauD wording study

research(adapters): decide whether a contrastive dual-encoder scorer belongs behind JudgmentPort

ci(docs): the plain-English gate skips docs/explanation and docs/tutorials

ci(examples): decide and, if feasible, schedule a weekly live example canary

test(live): rerun the nimble judge evolution with the Qwen reflector on a second card or a pod