Primary Sources

The narrative pages on this site — the five chapters, the era timeline, the team page — are written from a small set of structured source documents that live under docs/ps1/archaeology/ in the project repo. Those files are what get edited when the truth changes; the prose is downstream.

If you want to verify a date, an era boundary, a deprecated tool’s last touched commit, or a status-surface meaning that’s been demoted, the primary sources below are the right place to look. They are also, on their own, a reasonable read — written in the same plain register as the prose, without the narrative connective tissue.

The files

timeline.yaml

timeline.yaml breaks the PS1 branch into five eras with date ranges, summary sentences, dispositions (preserve / archive / promote), and — for the foreground playback era — a milestone commit list with SHAs and messages. It’s the structured source for the era timeline page. Open this when you want the exact commits that walked FISHING 1 from “scaffold” to “pixel-perfect with full SFX,” or when you need to know which part of the project history is officially demoted to archaeology.

team-perspective.yaml

team-perspective.yaml catalogues the six roles the author played across the project’s phases — operator-editor, embedded porter, instrumentarian, validator-skeptic, archaeology engineer, bespoke scene engineer — with the strongest evidence file for each. It also lists the project’s shared values in their original yaml-list form, which is where the website’s voice was anchored. Open this when you want to see which docs and skills correspond to which role, or when you want the canonical phrasing of the project’s ethic without the narrative wrapper.

tools.yaml

tools.yaml catalogues every script in the repo with a policy: keep_live, keep_but_demote_from_primary_docs, or archive_candidate. Each entry has a role hint and, where relevant, a last-touched commit and date. It’s the source of truth for “is this script still part of the operator loop, or is it historical?” The currently-live scripts include build-host.sh, build-ps1.sh, make-cd-image.sh, rebuild-and-let-run.sh, capture-host-scene.sh, export-scene-foreground-pilot.sh, build-scene-foreground-pack.py, wav2vag.py, and a handful of others that drive the per-scene bring-up loop.

blog-source-map.md

blog-source-map.md is a pointer map — not raw data, but a curated list of the branch’s strongest narrative sources, with one-line annotations. It points at the early branch voice, the validation skepticism agent prompt, the OFFLINE_SCENE_PLAYBACK_PIVOT research note that announced the methodology shift, the FISHING 1 pixel-perfect plan, and the water animation worklog. Open this when you want to read the project’s own prose at the moments when the direction was changing.

assumptions.yaml

assumptions.yaml is the project’s honest list of things it used to believe. Each entry has the assumption, why it made sense at the time, what changed, and the replacement insight. The assumptions catalogued include “broad automation would be the primary truth surface,” “generic restore and replay contracts would be the finish line,” “if the static analyzer said the content fit then runtime memory would be fine,” and “fgpilot was just a temporary experiment name.” Open this when you want to understand why the project went down particular roads before it turned around.

memory-constraints.md

memory-constraints.md is a short prose document that explains the two facts that shaped every decision: 2 MB main RAM, 1 MB VRAM, native 640x480. It walks through the math on why full-screen prerender for every scene was a non-starter (a single 16-bit frame is about 614400 bytes), why static fit numbers proved weaker than they looked once runtime allocation timing entered the picture, and how the wave-asset path ended up dictating preload order. Open this when you want the quantitative version of the chapter-4 narrative.

reporter-notes.md

reporter-notes.md is a short essay framing the cleanest read of the branch as one team that kept changing jobs, not standards. It identifies the April 12–22 window as the decisive turn from “trying to prove a universal replay theory” to “tightening a bespoke line that actually closes.” Open this when you want the framing distilled to five paragraphs.

status-surfaces.yaml

status-surfaces.yaml catalogues every “N of 63” number this project has ever published — 25/63, 27/63, 63/63, 57/63, 60/63, 1/63, 2/63 — with the date each was claimed, what it meant at the time, the source documents and commits, and the demotion reason. The 63/63 from 2026-03-29 is explicitly preserved as a historical false summit. Open this when you want to confirm that yes, the project’s status numbers really did go down before they went up, and the reasons are documented.

---

There are other supporting files in docs/ps1/archaeology/ — binary-library-index.csv, binary-library-manifest.json, the vision-artifacts/ and regtest-references/ corpora, the retired script and source archives. Those are evidence rather than narrative; the index above is the curated set that the website is written from. The full directory listing is at docs/ps1/archaeology/.

Drill-downs

For each of the larger evidence corpora, there is now a per-directory narrative page that walks through the contents in plain prose. These were written so a reader can understand each archive’s purpose and shape without opening every file.

• The binary library — the indexed record of every PS1 build the project produced (6,902 entries, originally 118 GB of payloads, deleted on 2026-04-22 with the index preserved).
• Regtest reference set — the sixty-three frozen scene captures the regtest harness compares fresh runs against, with a per-scene table.
• Host-script review — the bundled snapshot of the host-side identification and expectation pipeline as it stood on 2026-04-04.
• Retired scripts — fourteen scripts from the binary-library regression era, each with a paragraph explaining what it did and why it was retired.
• Retired source code — three minimal C bring-up programs from late 2025 that proved the GPU and CD-ROM I/O before the full game integration began.
• Vision-classifier artifacts — the reference-bank self-check across 63 scenes, the pipeline manifest, and a frozen FISHING 1 regtest run from the visual-detection-spec era.