Collect physical reference calibration #55

New Issue

2026-09-28T05:42:35Z

xavierk commented

2026-09-28 05:42:35 +00:00

Parent

Odin v1: terminal benchmark and health suite specification

What to build

A release reviewer can inspect a reproducible candidate reference corpus for every intended scored protocol and mode.

Acceptance criteria

Use one documented physical x86_64 reference; gather at least five complete eligible Standard runs per candidate mode over two days and two cold boots.
Preserve idle baseline, readiness/cooling, all attempts and exclusions, raw native values, output checks, exact builds/hashes, environment, and the approved median calculations.
Apply corpus stability limits of 5% median absolute deviation/median, or 10% for the browser suite; investigate unstable candidates without trimming valid slow values.
Publish an immutable candidate manifest and raw corpus, including fresh browser p95 values; do not enable numerical score claims before independent validation and freeze.

Blocked by

Decision sources

## Parent [Odin v1: terminal benchmark and health suite specification](https://git.bongbetic.com/xavierk/odin/issues/26) ## What to build A release reviewer can inspect a reproducible candidate reference corpus for every intended scored protocol and mode. ## Acceptance criteria - [ ] Use one documented physical x86_64 reference; gather at least five complete eligible Standard runs per candidate mode over two days and two cold boots. - [ ] Preserve idle baseline, readiness/cooling, all attempts and exclusions, raw native values, output checks, exact builds/hashes, environment, and the approved median calculations. - [ ] Apply corpus stability limits of 5% median absolute deviation/median, or 10% for the browser suite; investigate unstable candidates without trimming valid slow values. - [ ] Publish an immutable candidate manifest and raw corpus, including fresh browser p95 values; do not enable numerical score claims before independent validation and freeze. ## Blocked by - [Compute headline median and eligibility](https://git.bongbetic.com/xavierk/odin/issues/45) - [Run prepared-reference software modes](https://git.bongbetic.com/xavierk/odin/issues/52) - [Install native packages](https://git.bongbetic.com/xavierk/odin/issues/53) ## Decision sources - [Define Odin’s calibration procedure and release gates](https://git.bongbetic.com/xavierk/odin/issues/17#issuecomment-6776) - [Specify Odin’s v1 browser interaction workload](https://git.bongbetic.com/xavierk/odin/issues/20#issuecomment-6927)

xavierk added the ready-for-agent label 2026-09-28 05:42:35 +00:00

xavierk referenced this issue

2026-09-28 05:42:36 +00:00

Validate physical x86_64 systems #56

xavierk referenced this issue

2026-09-28 05:42:37 +00:00

Validate physical aarch64 system #57

xavierk added a new dependency 2026-09-28 05:43:57 +00:00

#45 Compute headline median and eligibility

xavierk added a new dependency 2026-09-28 05:43:57 +00:00

#52 Run prepared-reference software modes

xavierk added a new dependency 2026-09-28 05:43:58 +00:00