docs: standardize README and restructure docs/

[willtheorangeguy]

Wave 2 content pass. README 232 -> 74 lines, with an 8-page docs/ folder.

Where things went

Was	Now
The ~88-row table of contents	docs/builds.md
How To Use, the three representations	docs/usage.md
Automated Price Updates	docs/scraper.md (from the root SCRAPER_README.md)
Download, branches	docs/architecture.md

New: faq, troubleshooting, roadmap. Removed ## Changelog, ## You may also like..., ## Contributors, and the Discord badge. Credits table kept.

docs/legal/PRIVACY.md and TERMS.md lowercased via a temp filename; docs/images/ deleted.

Defects recorded in docs/internal/known-issues.md

The hero GIF never rendered. It used a /blob/ URL, which serves an HTML page rather than image bytes — the same fault found in Moms-Canning-Timer. Both images now come from .github/icons/.

A build-index row pointed at the wrong month. January 2018 Turbo linked to /2018/March/Turbo.md, while 2018/January/Turbo.md exists and is a different build. Anyone following the index for January's Turbo silently got March's parts and prices. Corrected.

Current prices carry no timestamp. scraper.py rewrites them daily, but nothing records when a file was last updated — so if the workflow has been failing for a month, stale prices look exactly as authoritative as fresh ones. A "prices as of" line written by the scraper would fix it in one change; roadmap.md calls it the most valuable item there.

Two inconsistent link forms in the index. Most rows were root-relative (/2021/...), nine were not. Both resolved from a README at the repository root, so the inconsistency was invisible — until the table moved into docs/, where nine rows broke and the rest did not. Normalised.

CC BY 4.0 over build lists published by the magazine, which were not this repository's to grant. Part lists are closer to fact than expression so the practical risk is low, but the statement is inaccurate — same pattern as FBI-Application-Guide. A provenance notice would be more honest, and the README now carries an ## Attribution section.

76 PCPartPicker links use willtheornageguy rather than willtheorangeguy. It is consistent across all 76, which suggests it is genuinely the account name rather than a typo — but worth confirming once, because if it is wrong every PCPartPicker link in the archive is dead.

Explained rather than listed

Why three representations exist: PCPartPicker is live and buyable but cannot hold a historical price; the Markdown diffs and works offline; the website browses. Only the latter two preserve the printed price, and that comparison is the archive's whole point. The CAD/USD split reflects where each is read.

The git history is already a price series. Because the scraper rewrites prices in place daily, every commit records what those parts cost that day — a genuinely interesting dataset that nothing surfaces.

One upstream, no fallback. Every current price comes from PCPartPicker; a layout change there stops all price updating.

Tooling note

Fixing this repo surfaced a gap in the sweep's own verifier: it tested relative links without URL-decoding, so any path containing %20 was reported broken. It now decodes first — which is what let the nine genuinely broken rows be separated from 57 false positives.

Verification passes: 74 lines, no orphan nav anchors, no forbidden sections, every relative link resolves.

Co-Authored-By: Claude Opus 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01TtaJBmFDK3GcSyuhSuZ84R

🤖 Generated with Claude Code

https://claude.ai/code/session_01TtaJBmFDK3GcSyuhSuZ84R