· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
CUDA Q1 learning dynamics
Full-TRAIN and development checkpoint controls, spatial query accessibility, optimization telemetry and bounded next-run recommendations.
Explore research by domain and date. Each campaign links its narrative report, mission or availability note, and interactive figures. Dates describe the findings unless explicitly labeled as a publication or follow-up.
Issues map · Evidence archive · Literature reviews
16 campaigns · newest dated findings first within each domain.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Full-TRAIN and development checkpoint controls, spatial query accessibility, optimization telemetry and bounded next-run recommendations.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. The dated Mac Q1 report, with its original configuration and research context.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. The retained R1 supervision comparison and its declared evaluation limits.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Pair ordering, signed bias and top-action regret under recorded-state support.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Checkpoint contrasts and operational-teacher calibration from the failed joint attempt.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Optimization traces, horizon residuals and receipt-bound cloud readback.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Completed Q1 and interrupted QH findings, kept as distinct experimental outcomes.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Retained generation evidence and its post-hoc population and measurement limits.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. The bounded selector screen, including support limits and retained negative outcomes.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Target-centred trajectory geometry and discrete motion diagnostics.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Matched geometry in the subsequent camera-stable validation.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Geometric evidence from the original negative renderer-equivalence checkpoint.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Fleet layout, transport and inspection-performance evidence.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Measured hardware and workload comparisons with source-bound timings.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Retained conceptual synthesis of acquisition, measurement and collection decisions; local campaign references are explicitly unavailable.
· Scientific snapshot; synthesis 15 October, domain/date restoration 16 October 2026
Answer-first report at its domain/date owner, with archived versions, reusable evidence and thesis coverage. Reconstructed historical checkpoint of nine held branches and a Mac hardware gate; current PR status remains on GitHub.
Open the GitHub issues map to explore the captured issue relationships, dependencies and linked pull requests. The map shows its capture time and any gaps in relationship data. GitHub owns the live status; the site refreshes after main pushes and on its daily schedule.
The experiment archive contains the supporting receipts and source readmes. The literature reviews explain the methods behind the study.
The original PR458 budget report archive preserves its HTML but lacks the companion assets needed for a complete hosted view. Its README points to the original bundle. It remains accessible as historical source evidence rather than appearing here as a working report.
The rollout-recipe history through 27 September also retains its source, but some referenced figures are unavailable. The newer conceptual report above includes the retained figures and marks unavailable local campaign references.
---
title: "Research campaigns"
phase: thesis
audience: public
status: current
owner: jan
format:
html:
toc: true
number-sections: false
---
Explore research by domain and date. Each campaign links its narrative report,
mission or availability note, and interactive figures. Dates describe the findings
unless explicitly labeled as a publication or follow-up.
[Issues map](../../_research/issue-map.html) ·
[Evidence archive](https://github.com/JanDuchscherer104/ARIA-NBV/tree/main/experiments) ·
[Literature reviews](../literature/index.qmd)
{{< include ../../_generated/research-index.md >}}
## Project map
[Open the GitHub issues map](../../_research/issue-map.html) to explore the
captured issue relationships, dependencies and linked pull requests. The map
shows its capture time and any gaps in relationship data. GitHub owns the live
status; the site refreshes after main pushes and on its daily schedule.
The [experiment archive](https://github.com/JanDuchscherer104/ARIA-NBV/tree/main/experiments)
contains the supporting receipts and source readmes. The
[literature reviews](../literature/index.qmd) explain the methods behind the study.
## Incomplete historical publications
The original [PR458 budget report archive](https://github.com/JanDuchscherer104/ARIA-NBV/tree/main/experiments/archives/pr458-publications)
preserves its HTML but lacks the companion assets needed for a complete hosted
view. Its README points to the original bundle. It remains accessible as historical
source evidence rather than appearing here as a working report.
The [rollout-recipe history through 27 September](https://github.com/JanDuchscherer104/ARIA-NBV/blob/main/.omx/context/local-rollout-recipe-history-through-20260927.html)
also retains its source, but some referenced figures are unavailable. The newer
conceptual report above includes the retained figures and marks unavailable local
campaign references.