Vale rule: flag mid-sentence "Infrastructure as Code" overcapitalization (IaC sweep)

[workprentice]

Follow-up to #20653 / PR #20952.

PR #20952 adds Pulumi.Overcapitalization, a Vale rule flagging generic concepts capitalized as if they were proper nouns. Its first draft also matched capitalized mid-sentence "Infrastructure as Code" ("the Infrastructure as Code" -> "the infrastructure as code"), but that token had a boundary bug (a bare [a-z] anchor doesn't reliably sit at Vale's auto-wrapped \b, so it silently matched only the rare case where the phrase follows a word ending right at punctuation). Fixing the bug surfaces the real violation surface: 161 hits across 78 files, almost entirely blog posts written over several years.

That's a real, separate editorial sweep and shouldn't be rushed into #20952, which is scoped to the already-validated "Stack" token. This issue tracks doing it properly:

1. Add the fixed token back to styles/Pulumi/Overcapitalization.yml: [a-z]+ Infrastructure as Code\b(?! [A-Z]).
2. Run vale content/ tree-wide and hand-audit the ~161 hits across ~78 files (mostly blog) for genuine proper-noun exceptions (workshop/talk titles, Pulumi's own "Universal Infrastructure as Code" branded phrase, external product names) before assuming they're all true positives.
3. Fix the true positives, add narrow .vale.ini opt-outs or TokenIgnores entries for the rest, following the precedent in Add Vale rule for overcapitalized "Stack" (issue #20653) #20952.
4. Ship at zero tree-wide findings, same bar as the Stack token.

Also in scope: widen scope: paragraph to cover headings and list items

Pulumi.Overcapitalization currently uses scope: paragraph, which Vale excludes headings and list items from. That was a deliberate, not-yet-resolved narrowing at merge time for #20952 -- not an oversight -- and it means the rule's most common real-world triggers (an H2 like "Managing your Stack", a list item like "- Create a new Stack") go unscanned today.

Widening it needs a re-tune, since the "zero false positives against 2,241 files" result for the Stack token was measured under paragraph scope. The correct shape is not a bare scope: text or scope: heading: either would false-flag Title-Case H1s, which are correct per AGENTS.md (H1 = Title Case, H2+ = sentence case) -- this repo already hit and solved that exact problem for Google.Headings, which is disabled in .vale.ini for the same reason. Mirror the enumerated scope list styles/Pulumi/HeadingSentenceCase.yml and styles/Pulumi/CrossReferenceHeadings.yml already use: heading.h2 through heading.h6, plus list for list items.

Widening the scope also resolves whether the content/docs/iac/operations/continuous-delivery/argocd.md opt-out in .vale.ini is covering anything real: at PR #20952's head, the Stack token's only match on that page is ## Define a Stack custom resource, an H2 -- so if scope: paragraph excludes it today, the opt-out currently covers nothing. Resolve the opt-out and the scope question together rather than independently.

---

🧠 This issue was created by workprentice.

[workprentice]

Adding scope-shape guidance from @CamSoper's review of PR #20952 (#20952 (comment)), since the IaC token this issue tracks shares the same scope: paragraph question as the Stack token:

Don't widen scope to text or add bare heading. Either would flag Title-Case H1s, which are correct per AGENTS.md (H1 = Title Case, H2+ = sentence case). This repo already hit that wall: Google.Headings is disabled in .vale.ini for exactly that reason, per the header comment on styles/Pulumi/HeadingSentenceCase.yml. If this rule needs to see headings at all, copy the enumerated shape HeadingSentenceCase.yml and CrossReferenceHeadings.yml already use:

scope:
  - heading.h2
  - heading.h3
  - heading.h4
  - heading.h5
  - heading.h6

plus list if list items should be covered.

The argocd.md Stack opt-out and this scope question are linked. At PR #20952's head, the determiner+Stack token matches exactly one line on that page: an H2 (## Define a Stack custom resource). If scope: paragraph excludes headings (unconfirmed -- vale couldn't be invoked in the review sandbox), that opt-out covers nothing. Worth re-checking both together when this issue lands rather than assuming either independently.

Also: PR #20952 was rescoped since this issue was filed to drop the IaC token entirely (a boundary bug made its "zero findings" claim false -- see the PR's commit history), so this issue is now the only place that token lives. The scope-shape choice belongs here.

[workprentice]

Opened #20983, which ships the mid-sentence "Infrastructure as Code" Vale token and clears every content/docs/** hit. Leaving this issue open for the remaining historical sweep — here's where things stand for whoever picks it up next.

Remaining surface (re-measured against master + the new token, per-directory):

Directory	Files	Hits
content/blog	80	180
content/events	~40	57
content/gads	~30	33
content/what-is	~15	24
content/case-studies	~12	14
everything else (tutorials, partner, solutions, pulumi-up, community, whitepapers, topics, testimonials, pricing, partners, enterprise, cloud-engineering, about, b)	~15	~17

Scope-behavior findings, from a hand-built repro fixture (not just a tree-wide count — see the boundary-anchor lesson from the earlier Stack token work): scope: paragraph on the Vale existence rule reaches body paragraphs and link text inside them, but not YAML front matter, headings, list items, blockquotes, or table cells. So even after the full sweep, front-matter titles/descriptions and heading text with this phrase won't be caught by the rule itself and would need either hand-fixing or a separate, carefully-scoped rule (do not widen scope: paragraph — see the header comment on Overcapitalization.yml for why that reflags correct Title-Case H1s).

Why this PR doesn't attempt the full sweep: most of the ~300 remaining hits need a human call on branded phrasing ("Universal Infrastructure as Code" and similar), quoted customer testimonials, and conference/talk titles — not something to push through in one review pass. Recommend tackling it by directory (blog first, since it's 60% of the surface) rather than all at once.

[workprentice]

Opened #21125, covering the content/blog slice (130 fixes across 65 files, 2 deliberate proper-noun exemptions, 3 files skipped due to in-flight PR collisions — see PR body for details).

Remaining surface after that PR lands, by directory (re-measured 2026-08-25 against the current tree, tree-wide vale run for the Overcapitalization IaC token):

Directory	Approx. hits
content/events	~57
content/gads	~33
content/what-is	~24
content/case-studies	~14
long tail (tutorials, partner, solutions, pulumi-up, community, whitepapers, topics, testimonials, pricing, partners, enterprise, cloud-engineering, about)	~17

Recommend the next slice be content/events (largest remaining directory), same directory-at-a-time approach, checking for in-flight PR collisions before editing. The 3 blog files skipped in #21125 (infrastructure-as-code-tools/index.md, walkthrough-of-adopting-infrastructure-as-code/index.md, aws-cdk-vs-pulumi-why-sst-switched/index.md) should also get a follow-up pass once their respective PRs (#21065, #21095, #21066) land.