What was actually searched, read, sampled, and executed.

Authenticated X corpus

One old 40-post sample became a 1,126-status union.

The previous tracked archive contained 40 main-timeline posts. This audit used the logged-in Posts, Replies, and Media surfaces; official X syndication; year-window live searches; the old tracked sample; DOM IDs; and conversation replies. Collection stopped only after bounded no-new-evidence thresholds. That is broad recoverable coverage, not a claim that deleted, private, or counter-only activity can be reconstructed.

Non-X source universe

Creator-controlled links first; third-party material labeled separately.

High-value creator claims

The claims that materially change our renderer model.

Rows below are concise paraphrases. The linked post is the authoritative wording and media. Scope is explicit so one challenge, asset, or year does not silently become a creator-wide law.

Open-source execution

The public Pogo project was inspected and run, not merely linked.

Failure and recovery log

Collection errors changed the method.

1. The first long Replies crawl used a 10-second shell timeout and was terminated after checkpointing. It was rerun with an appropriate timeout and reached the no-new-evidence threshold.
2. The normalized union regenerated while the sequential claim audit was running. The analyst caught the checksum change, compared the 2025 sources, regenerated annotations, and reread the complete current 144-record 2025 segment before finalizing.
3. The first 2025 X search recorded one rate-limit response after six successful packets. A later clean retry returned eight packets, 121 creator results, and zero response errors; deduplication did not change the 1,126-status union.
4. The first native-retweet search was rate-limited. A clean retry received a valid response and no records; existing retweets remain sourced from authenticated profile packets.
5. The first Sketchfab collector used a nonexistent user-ID route and received HTTP 404. Official API documentation showed that public model filtering uses the username.
6. Unsorted Sketchfab pagination produced 62 rows but only 53 unique IDs. Sorting by descending publication date stabilized cursor pagination: 62 rows, 62 unique models, and 62 successful detail calls.
7. Direct Tumblr post requests returned HTTP 403. Public timeline API responses supplied 112 direct bodies; the live RSS supplied a separately verified 20/20 window.
8. The standalone yt-dlp executable was not on PATH. The installed python -m yt_dlp module succeeded for metadata, thumbnails, and storyboard downloads.
9. Matching-name accounts without a creator-controlled link were excluded. No official YouTube, Vimeo, Reddit, Instagram, Bluesky, Cara, or Mastodon account was confirmed during this pass.

Remaining unknowns

Research narrowed the uncertainty; it did not erase it.

• The eight-count difference against X's profile headline cannot be treated as eight known missing authored posts. X may count unavailable, deleted, or differently classified activity.
• The hourglass's exact Blender file, full material graph, compositor parameters, scene lights, color management, blendshape targets, scroll nodes, and keyframes remain unavailable.
• No retrieved creator statement establishes a universal position on PBR, MeshBasic, Lambert, real-time lights, fog, LUTs, terrain, grass, sky, or day/night architecture.
• The public Pogo repository has no license grant. It is evidence we may inspect, not a source we may copy into Forged by Sin.
• Five third-party videos were sampled through every available uniform storyboard sheet. They were not watched frame-by-frame and are not used as creator-process authority.
• Deleted/private posts, inaccessible archive snapshots, private commissions, and offline files are not recoverable from this public audit.
• No finding authorizes renderer migration or canonical memory mutation. Those decisions remain with the Creative Director.