H-6736: Support campaign-based QA measurement, update low sample and outlier exclusion logic

[CiaranMn]

🌟 What is the purpose of this PR?

The PR is primarily aimed at supporting a production campaign-based measurement of QA hold, whereby a single observation is produced per production campaign, rather than an observation per batch. The rationale being that the actual time for QA testing starts when the campaign is finished, and earlier batches in the campaign aren't on hold for QA, they're on hold for the entire production to be complete.

Because this massively reduces the number of observations for production -> QA release, the PR also includes some rethinking of the 'exclude low samples' logic (most campaigns become 'low sample'). Instead, a 'low sample' is now < 5 (was < 10), and we introduce a new 'limited' sample tier at 5-9 which isn't excluded when the box is checked, but is shown with a badge.

Finally, the 'exclude outliers' logic is updated to only apply when calculating the mean. It previously applied to everything, but the median is already resistant to outliers, and the P95 (which drives planning parameter mismatch observations) should include the tail (and excluding outliers from the P95 when you have a population of e.g. a lot of 2 day QA release observations, and then a sizeable minority of 5 day QA release, leads to less informative results).

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

This PR:

• does not modify any publishable blocks or libraries, or modifications do not need publishing

📜 Does this require a change to the docs?

The changes in this PR:

• are internal and do not require a docs change

🕸️ Does this require a change to the Turbo Graph?

The changes in this PR:

• do not affect the execution graph

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
hash	Ready	Preview, Comment	Jul 29, 2026 3:43pm

2 Skipped Deployments

Project	Deployment	Actions	Updated (UTC)
hashdotdesign-tokens	Ignored	Preview	Jul 29, 2026 3:43pm
petrinaut	Skipped		Jul 29, 2026 3:43pm

[cursor]

PR Summary

Medium Risk
Changes core timing grain for QA, outlier semantics, and sample filtering across many derived metrics; behavior is well-tested but users may see different means, counts, and table visibility than before.

Overview
Adds campaign-grain QA hold timing: one observation per production campaign from campaign_rows (batch detail_rows stay as evidence), with UI copy and counts labeled as campaigns where timing_grain === "campaign".

Exclude outliers is renamed and narrowed to exclude outliers from mean — Tukey fences populate mean_observations and recompute means (including monthly means and batch-timeline / what-if means) while raw observations, sample size, median, percentiles, min/max, and yield/consumption series stay untouched. Period trends and opportunity comparisons use filtered values only for mean deltas; P95-driven planning signals stay on the full series.

Low-sample handling moves the hide threshold to < 5 observations, introduces a limited (5–9) tier with badges (not hidden by “exclude low samples”), and wires confidence/opportunity labels through shared sample-confidence helpers.

Site overview windowing now runs filterGraphNodeByDateRange (outlier + range in one path). In-app docs and settings text are updated to match.

^{Reviewed by Cursor Bugbot for commit 50eb1a4. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit d092c78. Configure here.}