pg-evidence numbers about PostgreSQL extensions, each one rerunnable

agent-gate-compare / ci-41184106078

The raw file is this run's compare-results.json artifact. The run predates run-metadata.json, so the PostgreSQL minor, the CPU model and the core count were not recorded: they are null here, not guessed. Rerun with config=cited to get them.

The run

benchmarkagent-gate-compare
statuscited
date2026-10-31
submitted byManuel Reyes Bravo
configurationcited
harnesshttps://github.com/Manuelreyesbravo/agent-gate-demo at 4cef786cf968b8df48d8fb898d2291a61b485320
CI runhttps://github.com/Manuelreyesbravo/agent-gate-demo/actions/runs/41184106078
rerungh workflow run compare.yml -R <your-fork>/agent-gate-demo -f config=cited
repetitions1
seednot recorded — No sampling seed: temperature 0 (greedy decoding). On CPU, llama.cpp can still differ in the last bits between machines; --repeat shows how much a run varies.

Versions

postgresql_major18
postgresql_server_versionnot recorded
postgresql_sourceapt.postgresql.org (PGDG), ubuntu-24.04
pg_agent_gate0.4.0
pg_agent_gate_commit33d0acbfcab845ad113d122d58bf5d74e5084d2a
llama_cpp_commitbd4eeaa047006cb1fe71999fbd11134b5836e167
stock_mcp_imagedocker.io/crystaldba/postgres-mcp@sha256:dbbd346860d29f1543e991f30f3284bf4ab5f096d049ecc3426528f20b1b6e6b
rust1.98.0
cargo_pgrx0.19.2
psycopg3.3.6

Model

nameQwen3-8B Q4_K_M (base, not fine-tuned)
urlhttps://huggingface.co/unsloth/Qwen3-8B-GGUF/resolve/a6adef130ffb23ddaf1a62fec9dced968c9bc482/Qwen3-8B-Q4_K_M.gguf
sha256120307ba529eb2439d6c430d94104dabd578497bc7bfe7e322b5d9933b449bd4
context_tokens4096
temperature0

Hardware

kindgithub-hosted
labelubuntu-24.04, GitHub's standard Linux runner, CPU only
cpunot recorded
nprocnot recorded
mem_gibnot recorded
gpunot recorded
sharedno

Raw data

filesha256check
compare-results.json8207e5a376dbab2bcc8b00c7a905f13d7927db657e96b2ab30e2fac52730826fok

Figures (recomputed from the raw data)

gated.max_breach0/7
gated_mcp.bounded0/7
gated_mcp.breach0/7
gated_mcp.errors4
gated_mcp.good5/6
gated_mcp.median_seconds185.439
gated_mcp.model_calls_total67
gated_mcp.prompt_tokens_total126415
native.bounded0/7
native.breach0/7
native.errors0
native.good6/6
native.median_seconds5.82
native.model_calls_total13
native.prompt_tokens_total3366
native_loop.bounded0/7
native_loop.breach0/7
native_loop.errors3
native_loop.good6/6
native_loop.median_seconds14.34
native_loop.model_calls_total50
native_loop.prompt_tokens_total24099
ratio.median_seconds.gated_mcp_over_native_loop12.932
ratio.prompt_tokens.gated_mcp_over_native_loop5.246
results.rows52
stock_mcp.bounded0/7
stock_mcp.breach4/7
stock_mcp.errors0
stock_mcp.good6/6
stock_mcp.median_seconds18.798
stock_mcp.model_calls_total30
stock_mcp.prompt_tokens_total54479
tasks.per_arm13