Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
153 changes: 88 additions & 65 deletions docs/creative/06-dogfood.md
Original file line number Diff line number Diff line change
@@ -1,95 +1,118 @@
# GitHub Profile refinement — Step 6 dogfood
# Human Zine — Step 6 dogfood

Status: local acceptance passed on 2026-07-31; public read-back follows release
Status: **completed on 2026-08-09.**

## Task 1 — recognize and choose
The live GitHub Profile passed the executable self-dogfood checks below. This
was operational verification, not participant research: automated self-dogfood
does **not** validate whether real visitors infer personality or creative
judgment from the issue.

### Script

1. Open the Profile source at the top.
2. Read for five seconds.
3. Identify the creator hook, relationship action, and two primary works.
## Script 1 — first visitor, fresh session

### Result
### Script

Pass.
1. Open <https://github.com/BrickerP> in a fresh browser session.
2. Confirm the author is visible in the first viewport.
3. Scroll once by `701px` and choose each work in turn.
4. Return to the Profile after the first work, then inspect the four spreads and
browser console.

- `LOOP / LEDGER` and `Strange loops. Open ledgers.` appear before career
information.
- `OPEN A LINE — EMAIL YUPENG` resolves the distinctive phrase into an
unambiguous `mailto:` action.
- LOOP 01 and LEDGER 02 are separate full-width linked chapters.
### Live result

### Friction found and fixed
- The author was visible in the first viewport.
- One `701px` scroll revealed two visibly distinct works.
- `WATCH FILM →` returned HTTP `200`, title `Beijing — Endless Second Ring`, in
`1772ms`.
- After returning to the Profile, `PLAY ARCHIVE →` returned HTTP `200` with the
title `BrickerP AI Usage Chronicle` in `2304ms`.
- All four SVG requests returned HTTP `200`; there were no broken images and no
console errors.

The first implementation draft used a newly invented invitation and described
the Ledger as making practice “accountable.” Both were replaced with the
approved invitation and the bounded-trace disclosure to avoid expanding the
evidence claim.
Result: **PASS**.

## Task 2 — narrow screen and image-disabled path
## Script 2 — mobile

### Script

1. Read the README in canonical single-column order.
2. Ignore all image contents.
3. Confirm that every essential meaning and destination remains available.
1. Open the live Profile at `320 × 844` and `390 × 844`.
2. Compare `document.documentElement.scrollWidth` with `clientWidth`.
3. Inspect each SVG's completion state and natural dimensions.
4. Exercise the artifact links and the Email, Resume, and GitHub actions.

### Result
### Live result

Pass.
| Check | `320 × 844` | `390 × 844` |
| --- | --- | --- |
| Horizontal geometry | `scrollWidth == clientWidth == 320` | `scrollWidth == clientWidth == 390` |
| Cover | complete; natural `1200 × 630` | complete; natural `1200 × 630` |
| Artifact | complete; natural `1200 × 720` | complete; natural `1200 × 720` |
| Process | complete; natural `1200 × 720` | complete; natural `1200 × 720` |
| Open line | complete; natural `1200 × 360` | complete; natural `1200 × 360` |

- No table or two-column dependency remains.
- Work names, one-sentence meanings, artifact links, email, three breadth
links, resume, and LinkedIn remain in ordinary Markdown/HTML.
- Both image alternatives name the artifact and explain its role without
copying a dynamic total or date.
- Professional facts remain reachable in one disclosure rather than competing
with the creator hook.
- No horizontal overflow or clipped spread content was observed at either
viewport.
- The native work and contact links remained visible and operable. Email kept
its `mailto:` behavior, Resume opened the PDF destination, and GitHub opened
the expected profile destination.
- Browser zoom is **NOT_RUN** because gstack exposes no zoom command; this is a
recorded tool limitation, not a passed zoom claim.

## Task 3 — regenerate and inspect the living card
Result: **PASS** for both available viewport scripts.

## Script 3 — recruiter route

### Script

1. Generate cards from a controlled fixture.
2. Generate cards from the current `usage.json`.
3. Build the production `docs/` artifact.
4. Render LOOP, LEDGER light, and LEDGER dark at `560 × 160`.
1. Open the Profile at `1440 × 1000`.
2. Confirm role and location are available at the top.
3. Use two `PageDown` actions and inspect the contact row.
4. Verify Email, Resume, GitHub, and the Website fallback.

### Result
### Live result

Pass.
- Role and location were visible at the top of the Profile.
- The contact row was visible after two `PageDown` actions.
- Email exposed the expected `mailto:yplmicro@gmail.com` href.
- Resume returned HTTP `200`, `application/pdf`, `50,918` bytes, and `2` pages.
- GitHub resolved to <https://github.com/BrickerP>.
- Website retained the structured `personal-intro` fallback.

- Six focused generator tests passed.
- The production AI Usage build passed.
- Totals, date range, cache amount/share, weekly aggregation, and all four tool
series remain derived from report data.
- Both generated themes contain exactly one vermilion Skyline-origin node.
- LOOP contains one road/vanishing-point node and no drawn control.
- Rendered cards remain legible and visually related without requiring an
edge-to-edge seam.
Result: **PASS**.

### Environment-only retry
## Severity and disposition

The first AI Usage build attempt used a read-only `node_modules` symlink in the
temporary worktree, so TypeScript could not write its build cache. Copying the
already-installed dependencies into the writable temporary worktree resolved
the environment issue; the one subsequent production build passed.
There are no P0 or P1 findings, and Step 6 requires no product fix.

## P0 status
| P2 observation | Evidence | Disposition |
| --- | --- | --- |
| Native links render about `17px` high, below the `44px` touch-target guidance. | They are underlined, keyboard reachable, and operable in both mobile viewports. | Defer; keep GitHub-native links. |
| Contact requires two `PageDown` actions at `1440 × 1000`. | Artifact and process evidence intentionally precede contact. | Defer; preserve the approved editorial hierarchy. |
| AI Usage reports `2 sources delayed`. | The archive discloses its own source freshness rather than hiding it. | Defer as a separate transparency/operations item. |
| The structured Website fallback takes one icon action to open. | Role, location, Resume, Email, and GitHub remain directly available from the Profile. | Defer; do not duplicate professional prose in the zine. |

No unresolved P0 remains in the approved Profile journey.
## Cold-start path

Public publication is a release check, not a substitute for these task scripts:
the Profile README, LOOP card, and both LEDGER themes must still be read back
from their public URLs before completion is claimed.
The existing live flow is the cold start:

## Cold-start path
**Open Profile → one scroll → choose `WATCH FILM` or `PLAY ARCHIVE` → continue
to `OPEN A LINE`.**

No new explanatory prose is warranted; adding it would violate the approved
Human Zine word budget.

Temporary screenshot runtime paths produced during the session are evidence
for that session only. No screenshot binaries are committed to the repository.

## Completion checklist

The Profile itself is the three-line cold start:
- First-visitor live journey: **PASS**
- `320 × 844` and `390 × 844` responsive geometry: **PASS**
- Four live SVGs and browser console: **PASS**
- Artifact/contact links and recruiter route: **PASS**
- P0/P1 remediation: **PASS — none required**
- Browser zoom: **NOT_RUN — gstack has no zoom command**
- Real-human qualitative inference: **NOT_VALIDATED**

> `LOOP / LEDGER`
>
> **Strange loops. Open ledgers.**
>
> I build strange AI loops and keep the ledger open.
All executable Step 6 checks passed. The only unanswered product question is
whether real first-time visitors infer a specific person, creative judgment,
and range from the evidence; that requires Step 7 participants.
84 changes: 57 additions & 27 deletions docs/creative/07-next-loop.md
Original file line number Diff line number Diff line change
@@ -1,40 +1,70 @@
# Next Creative Loop — Open Circuit
# Human Zine — Step 7 next loop

## What the previous loop validated
Status: **ready for the user and real participants as of 2026-08-09. The test
has not run.**

- `LOOP 01` and `LEDGER 02` read as one profile system when they share numbering,
mono identity labels, a dark field, and one restrained handoff node.
- The Profile works better when the two signature artifacts carry the story
instead of more explanatory text.
- The fixed LOOP card is already a stronger expression of the new identity than
the live animation chrome.
## Validated

## What the previous loop disproved
- The live Profile exposes the author in the first viewport and two visually
distinct works after one scroll.
- The four static spreads render without overflow at `320 × 844` and
`390 × 844`, and the native artifact/contact routes remain operable.
- Recruiter facts, Resume, Email, GitHub, and the structured Website fallback
remain discoverable without adding résumé prose to the zine.
- The existing cold-start route works mechanically: open Profile, choose a work,
then continue to `OPEN A LINE`.

- More Profile copy is not the next improvement.
- A dynamic Profile service is unnecessary; LEDGER can update independently.
- Repeating accent dots, decorative grids, dashboard overlays, or more controls
would weaken the artwork.
## Falsified

- A LOOP / LEDGER compatibility layer or shared cross-project visual system is
not needed for the live Human Zine journey.
- More cold-start copy is not the next improvement; it would exceed the approved
word budget without answering the qualitative question.
- Automated link, viewport, and asset checks cannot stand in for evidence about
a real visitor's memory of the author.

## Not validated

- Whether a first-time AI builder can name two different modes of making after
an unaided visit.
- Whether the same visitor infers one non-career author trait and can cite the
specific spread evidence behind it.
- Whether the visit produces a specific next action or question rather than a
generic positive reaction.

## New opportunity

Make the runtime itself recognizable as `LOOP 01` by translating the card's
warm-white carrier and single vermilion handoff node into the physical road.
The Process spread can become stronger evidence of judgment if real visitors
cannot make the intended inference. That opportunity is conditional on a failed
participant test; it is not authorization to add copy or restore shared
branding.

## The one next knife

Run a **five-person, 90-second unaided recall test** with first-time AI builders.

1. Give each participant the live GitHub Profile with no framing or task hints.
2. Allow at most `90` seconds of independent browsing.
3. Hide the page and ask them to name two different modes of making.
4. Ask for one non-career author trait and the exact page evidence that caused
the inference.
5. Ask what specific action they would take or question they would ask next.

## Selected next cut
Success requires both:

**Carrier Line / Open Circuit**
- at least `4/5` participants name two different modes of making and one
non-career author trait with evidence; and
- at least `3/5` form a specific next action or question.

Success means that with the interface mentally removed, a still frame and a
short clip still feel related to the LOOP card.
Do not make a code or UI change unless this test fails. If it fails, the next
design target is the evidence on the **Process spread**—not additional prose and
not shared branding across the works.

## Explicit exclusions
## Backlog, not the next knife

- Profile avatar
- AI Usage data or generation logic
- New public actions
- Route, duration, passage-order, recording, or seam changes
- Maps, progress HUDs, chapter selectors, glow, particles, or repeated nodes
- Native-link touch-target height.
- AI Usage's `2 sources delayed` operations state.
- Two-`PageDown` contact depth.

The user approved this cut and authorized implementation, push, and deployment
on 2026-07-31.
These remain deferred backlog observations and must not displace the single
participant-recall test above.