clarity-inventory

name	clarity-inventory
description	Phase 0 of the stranger test — build the shared evidence base (section map, per-section copy with content hashes, fold crop, CTA census) that every scored stage consumes. Use before any clarity-* stage skill; never score anything here.

clarity-inventory — the evidence base

You are not auditing yet. You are building the single shared view of the page that all six stage auditors will cite, so their findings reconcile. Six auditors scraping independently produce six slightly different pages and findings that can't be deduplicated — one inventory, six consumers.

Procedure

1. Render honestly. Load the target at a desktop viewport (1440×900) and a mobile viewport (390×844) if browser tools are available. Screenshot both, plus a crop of exactly the first fold (0 to viewport height, no scroll). If only fetch is available, extract HTML and mark the run --fast: note that fold geometry, scroll cues, and visual hierarchy findings will be UNVERIFIED.
2. Map the sections in scroll order. A section = a visually or semantically bounded band. For each, record: anchor (id, or a generated sec-N), position (which screenful it starts on), heading, and its full normalized text (whitespace-collapsed).
3. Hash each section: sha256 of the normalized text, first 12 hex chars. This is the finding anchor_hash — it lets a future run tell "same copy, finding stands" from "copy changed, re-audit".
4. CTA census: every clickable action on the page — label, visual weight (filled / outlined / bare link), position, destination, and whether it is above the fold. Mark the primary per screenful (the most visually dominant). Two primaries on one screenful is a fact to record, not judge.
5. Number census: every quantitative claim on the page (prices, counts, percentages, times) with its section anchor — the trust and pricing stages audit specificity and consistency from this list.
6. Entry matrix: record every significant path that brings visitors, one row per channel — paid ad copy, search snippet, the README, a shared link's unfurl (fetch the og:title/og:description; advocacy traffic arrives through that card), and the page's own <title> + meta description as the direct/fallback row. For each row: the exact entry copy, and the awareness stage (Schwartz) it delivers — a paid ad tends to bring problem-aware visitors, a friend's link most-aware ones. Mark the primary channel (best available traffic evidence; otherwise the title/meta row). The promise stage audits message match per row; its gate binds on the primary.

Output

Write audit/inventory.md (in the audited repo, or the working directory for a remote site):

# Inventory — <target> — <date>
viewport: 1440×900 · mode: full|fast

## Entry matrix (the promises visitors arrive holding)
| channel | entry copy (verbatim) | awareness stage | primary? |
|---------|-----------------------|-----------------|----------|

## Sections (scroll order)
| # | anchor | screenful | heading | hash | words |
|---|--------|-----------|---------|------|-------|

### <anchor> — full text
<normalized copy>

## CTA census
| label | weight | section | above fold? | destination | primary? |

## Number census
| value | claim | section |

Keep the copy verbatim — auditors quote evidence from this file. Do not summarize, do not editorialize, do not flag anything. Inventory findings are a category error.