DoE per release: run, analyse (R, LaTeX), publish to the .com XWiki #158

New Issue

2026-11-08T13:29:03Z

rajames commented

2026-11-08 13:29:03 +00:00

Ruled

Captain Bob, 2026-11-08: "a pipeline that will run and publish a DoE of some small kind we can use to support patent claims to a SMALL degree. It should use R and LaTeX to publish an analysis similar to what we do with experiments and reports. Difference being they will be published each release on .com's XWiki with charts and tables in the XWiki dialect. This will be a milestone too."

What is there today

R analyses and LaTeX reports: experiments/bare_metal/analysis/ (analyse_bare_metal.R, analyse_multirep.R, ...; report/bare_metal_doe_report.tex, stadium_relaunch_report.tex), and the CSV extraction every QEMU boot already runs (scripts/extract_doe.sh -> experiments/bare_metal/runs/).
But it measured v3, whose physics runtime goes with v3 (#156), and no DoE has run since 2026-09-16 (#141; the CSV columns always 0, #142). The per-release DoE measures v4 -- its heat model (v4/src/heat.c) and what the boot records.

Built (to do)

The design, fixed before it runs (written in the tree, versioned): the claim each run supports, its factors and levels, the response, the replicates, the analysis -- small enough to run inside a release on today's server (one QEMU at a time, foreground; the release's time budget to be set).
The run: a scripts/nightly.sh doe stage, after acceptance, from the release's own image: the design's boots, their CSVs committed as logs always are.
The analysis in R: one script from the CSVs to tables and charts -- effect sizes with confidence intervals, not p-values alone; every run reported, failures included.
The report in LaTeX (PDF, as the existing reports), attached to the release, org.starship:documentation:<version>, classifier doe.
The .com page, in XWiki 2.1: the same tables (|= headers) and charts (rendered by R and attached, or XWiki's {{chart}} macro over a page table -- to be chosen), one page per release under a DoE section, published by the release workflow (a .com credential as a repository secret; REST, XML, as site/org publishes).

Open, for rulings when the milestone starts

Which claim(s) the first design supports, and its factors.
The time budget per release, and whether the DoE runs for alpha releases or only from a set point.
Charts as R images or XWiki's chart macro.
Where the pages sit on the .com (its own section; linked from Documents, #157).

Care

Patent support "to a SMALL degree": the design is published before its results, the analysis is fixed, and the report states its limits (QEMU under TCG, one server, the sample size) -- so what it shows can be relied on for exactly what it says.

### Ruled Captain Bob, 2026-11-08: "a pipeline that will run and publish a DoE of some small kind we can use to support patent claims to a SMALL degree. It should use R and LaTeX to publish an analysis similar to what we do with experiments and reports. Difference being they will be published each release on .com's XWiki with charts and tables in the XWiki dialect. This will be a milestone too." ### What is there today - R analyses and LaTeX reports: `experiments/bare_metal/analysis/` (`analyse_bare_metal.R`, `analyse_multirep.R`, ...; `report/bare_metal_doe_report.tex`, `stadium_relaunch_report.tex`), and the CSV extraction every QEMU boot already runs (`scripts/extract_doe.sh` -> `experiments/bare_metal/runs/`). - **But it measured v3**, whose physics runtime goes with v3 (#156), and no DoE has run since 2026-09-16 (#141; the CSV columns always 0, #142). **The per-release DoE measures v4** -- its heat model (`v4/src/heat.c`) and what the boot records. ### Built (to do) 1. **The design, fixed before it runs** (written in the tree, versioned): the claim each run supports, its factors and levels, the response, the replicates, the analysis -- small enough to run inside a release on today's server (one QEMU at a time, foreground; the release's time budget to be set). 2. **The run:** a `scripts/nightly.sh doe` stage, after acceptance, from the release's own image: the design's boots, their CSVs committed as logs always are. 3. **The analysis in R:** one script from the CSVs to tables and charts -- effect sizes with confidence intervals, not p-values alone; every run reported, failures included. 4. **The report in LaTeX** (PDF, as the existing reports), attached to the release, `org.starship:documentation:<version>`, classifier `doe`. 5. **The .com page, in XWiki 2.1:** the same tables (`|=` headers) and charts (rendered by R and attached, or XWiki's `{{chart}}` macro over a page table -- to be chosen), one page per release under a DoE section, published by the release workflow (a .com credential as a repository secret; REST, XML, as `site/org` publishes). ### Open, for rulings when the milestone starts - Which claim(s) the first design supports, and its factors. - The time budget per release, and whether the DoE runs for alpha releases or only from a set point. - Charts as R images or XWiki's chart macro. - Where the pages sit on the .com (its own section; linked from Documents, #157). ### Care Patent support "to a SMALL degree": the design is published before its results, the analysis is fixed, and the report states its limits (QEMU under TCG, one server, the sample size) -- so what it shows can be relied on for exactly what it says.

rajames added this to the DoE per release milestone 2026-11-08 13:29:03 +00:00

rajames referenced this issue from a commit

2026-11-08 13:29:16 +00:00

FABRIC-3 XLVIII.12: a DoE per release, its own milestone (#158) [skip ci]

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: rajames/LithosAnanke#158