change-name

Files (9)

• CARD.yml
• GLANCE.yml
• README.md
• SKILL.md
• notice-config.example.yml · examples/
• prestored-change-list.example.txt · examples/
• pdf-prestoration.md · playbooks/
• pdf_add_correction_notice.py · scripts/
• pdf_name_scan.py · scripts/

SKILL.md (first lines)

# CHANGE-NAME 🦋

> *Say my name right — everywhere it's written.*

People change names — transition, marriage, divorce, faith, stage names, safety,
transliteration across scripts. The record does not follow by itself. Names live in
version-of-record PDFs, citation databases, git histories, link annotations, metadata,
and archives engineered never to change. This skill changes them — carefully, honestly,
and with receipts.

**One concrete task ships today:** ninja-editing PDF files. **The context is a socket:**
each venue (git, ACM, ORCID, registries, archives, life admin) plugs in as a playbook in
`playbooks/`. This SKILL.md is self-contained for the PDF task; playbooks extend it.

## The Ethics Gate (read first, always)

A correction is legitimate when **all five** hold. Any four out of five is a forgery
with good manners:

1. **Documented wish or clear standing** — the subject asked (best: find it in their own
   words), or those closest to them act on documented evidence of their identity.
   Standing order: self > subject's recorded request > family/co-authors > community.
2. **Original preserved** — bit-for-bit, sha256-pinned, adjacent to the edition.
3. **Total disclosure** — every change enumerated, **visible on page 1** (see
   [Visible correction notice](#visible-correction-notice-required)), in PDF metadata
   and attachments, and beside the file (README). Metadata-only disclosure fails: a
   renamed or printed copy loses the filename, README, and sibling original.
4. **Honest labeling** — the filename says what the file is (`-memorial-edition`,
   `-corrected`, never a silent replacement).
5. **Canonical fix pursued** — the edited derivative is a bridge to the institutional
   correction (publisher petition, registry update), never a substitute for it.

**Never out anyone.** A scan that finds every occurrence of an old name is also a map of
someone's exposure. Treat SCAN reports as sensitive; publish only what the subject or
their standing-holders approve.

## Protocol: SCAN → DISCUSS → EDIT → VERIFY → PUBLISH

### 1. SCAN — look at the PDF and describe what you find

Run the sister script (degrades gracefully without pikepdf):

```bash
python3 scripts/pdf_name_scan.py TARGET.pdf --find "Old Name" [--replacement "New Name"]
```

It reports, without changing anything:

- **Document metadata** — `/Author`, `/Title`, docinfo, XMP
- **Content-stream text** — including **kern-split occurrences**: TeX-produced PDFs
  store names like `(Freudenber)18(g)`, invisible to naive grep. The scanner inflates
  streams and matches across TJ kern numbers.
- **Link annotations** — URIs containing the old name (profile links, repo links)
- **Font subsets** — per font: BaseFont, CharSet/FirstChar/LastChar, and **whether the
  embedded glyphs can spell the replacement**. (Founding case: the email font subset
  ended at glyph 117 — one code point short of the letter `v`. It could not spell
  "vanessa". Know this *before* you promise an in-font edit.)

Describe findings to the human in plain language: where the name appears, what fonts
carry it, what can be edited in-font and what needs a fallback face.

### 2. DISCUSS — agree the plan

Present the evidence, then agree, explicitly:

- **Which occurrences** change (byline yes; historical citations in *other* people's
  references may be out of scope), and which stay.
- **Replacement strings** — name, email, URLs (beware redirect masking: an old-handle
  URL that redirects still *displays* the old handle; update visible text).
- **Typography** — in-font where the subset allows (recompute kerning), matched fallback
  face where it doesn't (base-14 Helvetica for sans, Times for serif).
- **Labeling** — the edition's filename, the visible page-1 notice (wording and
  placement), and the embedded provenance note / attachment text.
- **The gate** — confirm all five conditions; record the wish/standing evidence in the
  ledger.

## Visible correction notice (required)

Every prestored PDF must carry a **visible correction notice on page 1**. Outside-the-page
disclosure (filename, README, sibling original, PDF `/Note` metadata) is necessary but not
sufficient — those signals vanish when someone renames, re-hosts, or prints the file.

The notice must be **explicit and machine-parseable** — clear enough that a human skimming
page 1 and plain text extraction (`pdftotext`, Ghostscript `txtwrite`) both recover: this is
not the version of record; what changed; where the unaltered original lives; that a canonical
fix is being pursued.

**Required content** (condense for page 1; full list may live in a PDF attachment):

- Honest label (`CORRECTED MEMORIAL EDITION`, `CORRECTED EDITION`, etc.) and date
- **Not the version of record** — cite DOI or publisher identifier when one exists
- What was updated (byline, email, footnote URLs, metadata) and the new identity used
- Standing / wish — link the subject's own words when they exist
- Unaltered original preserved — repository path and **SHA-256** (full hash, or labeled
  prefix with ellipsis)
- Pointer to full change list (embedded attachment or README)
- Canonical correction in progress (publisher petition, registry update, etc.)

**Placement recommendations** (least disruptive first):

| Placement | When to use |
|-----------|-------------|
| **Footer band above the publisher reference / copyright block** | Default. Page 1 already has small legal text there; the notice reads as metadata, not content. Room for 3–4 lines. |
| **Top margin above the title** | When the footer is crowded. One or two lines; keep the full notice in the attachment. |
| **Avoid** | Between author block and abstract; inside the abstract; rotated side-margin text — all read as content tampering or archival stickers. |

**Typography recommendations:**

- **7–7.5 pt** minimum (7 pt floor; do not go to 6 pt)
- Match the document's house style — sans used for affiliations, or base-14 Helvetica
- Full text-block width (~468 pt on US Letter) — wrap across the whole page, not a
  narrow column; fewer lines means less vertical stack and no collision with existing
  header text like "Accepted for DLS"
- **Small caps** on the label line (`CORRECTED MEMORIAL EDITION`)
- Thin rule (0.25 pt) above the notice to separate it from body content
- Monospace or distinct treatment for hashes and URLs aids both reading and extraction

**Example** (4-line footer; adapt labels to the case):

From SimHacker/moollm, MIT.

More from SimHacker/moollm