Collected Wisdom

A site that indexes private Google Docs for their author, and publishes the subset that can be public

site
scoping
working document
The scope of a second personal site built from Google Docs: the source inventory, the two switches that decide what is exposed, how the build mirrors a shared Doc as a PDF, and the acceptance checks for the first release.
Published

September 8, 2026

Scoping, 2026-09-06. No repository exists yet. Five decisions are settled (Section 3); the inventory in Section 4 is the work that unblocks everything else, and it needs Andy rather than an agent.

Revised the same day, after Andy: the Google Docs stay unshared by default and the site earns its place by organizing them for him. Publication is now the exception rather than the goal, which changed Sections 1, 5 and 7.

1. Goal

One site holding the personal writing that has accumulated across Google Docs, organized well enough to find a piece a year from now.

The first reader is Andy. The site is a way of seeing thirty scattered Documents as one collection, with a sentence on each saying what it is. A Document does not have to be publishable to earn a place in that view, and most of them will not be. Where a piece is publishable, the same index entry serves a stranger without any extra work.

2. Out of Scope

Stated once, here.

  • Not a blog. No dates and no feed. An entry is revised in place rather than superseded by a newer post.
  • Not a migration of TrinityMetrics. Both sites stay. The personal material already on TrinityMetrics — the Recommended Resources list, the Now page, the Uke Skywalker link — stays where it is, and the two sites link across.
  • Not a rewrite of the Docs. Google Docs stays the place the writing happens. The site indexes; it does not become the editor.
  • Not a second copy of anything private. No Document body, and no Document that names an identifiable person, is copied into this repository at any tier.

3. Settled Decisions

# Decision Value
1 Name Collected Wisdom. Repository collected-wisdom.
2 Repository Public, on GitHub, published to Pages by a copy of publish.yml.
3 Document sharing Unshared by default. A Document is link-shared only when its contents are deliberately being published.
4 Content model The Google Doc stays canonical. The site carries an index entry per source; only a shared Doc is mirrored as a PDF.
5 Boundary with TrinityMetrics Nothing moves. The two sites link to each other.

The name claims a lot for a page under your own name. “Collected Wisdom” as a site title reads as a verdict on the contents. A subtitle that says whose and why — collected over twenty years, mostly for my own use — costs one line and removes the problem. Decide it when the index page is written; it does not change anything structural.

4. Source Inventory

One row per source. This table is the deliverable that unblocks the build.

Tiers are defined in Section 5. TBD is a legitimate value and should stay until the row is decided rather than being guessed.

# Source Where it is now Subject Tier Doc ID or URL Note
1 Doc on meaning Google Docs Meaning INDEXED needed Stays unshared. Promote to OPEN if a version of it is ever finished enough to publish.
2 Pathways tools Google Docs Meaning INDEXED needed Confirm whether this is career pathways or decision tools; it changes which subject it sits under.
3 Blog post on parenting Published, host to confirm Family OPEN needed Already public, so the tier is settled. If it is on Medium, the site links out rather than copying.
4 Notes on raising kids Google Docs Family LOCAL do not record Presumed to name your children, so neither the contents nor the title go into a public repository.
5 Uke Skywalker ukeskywalker.com Music OPEN https://www.ukeskywalker.com Its own site. Stays where it is; the entry is one line and an outbound link.
6 Hamilton song maps Local files Music OPEN files to upload Confirm the format. PDF renders in a browser on a phone; a spreadsheet or an image set does not.
7+ Others Google Drive — TBD — “There might be others.” A sweep of Drive fills these rows.

Row 7 is the one to work on first. Six named sources do not need a website. What decides whether to build one is how many rows the Drive sweep adds. Twenty rows justifies a site. Eight does not, and the answer then is a single page on TrinityMetrics.

5. The Two Switches

What is exposed is set by two independent switches, and confusing them is how this kind of site leaks.

  1. The Google share setting decides who can read the contents. Unshared means you, and Google’s sign-in wall for everyone else.
  2. Where the entry lives in the repository decides who can read the title and description.

A render-excluded folder is not a privacy mechanism. This site excludes blog/drafts/** from the render, and that is sound because a draft is unfinished rather than secret. In a public repository the excluded files are still readable by anyone who opens the repository, and git keeps every version of them. So the second switch has two usable settings, not three: in the public repository, or on your machine only.

Tier Doc sharing Entry lives in Public sees Use for
OPEN Link-shared Rendered page Title, description, and the contents as a mirrored PDF Anything deliberately published
INDEXED Unshared Rendered page Title and description; the link hits a sign-in wall The default. Organizes the Document for you at the cost of a public title
LOCAL Unshared Gitignored local/ Nothing Anything whose title alone should not be public

INDEXED is the default and covers most rows. You get the collection view, the tags, and one click through to the Document. A stranger gets a title and a sentence, which is the same thing a bookshelf shows a visitor.

LOCAL renders on your machine and stops there. local/** goes in .gitignore and in the Quarto render: exclusion list, so CI never sees it and neither does the repository. quarto preview on your laptop renders the whole site including those entries, and the output directory is gitignored, so nothing about a LOCAL row reaches GitHub. The cost is that LOCAL entries are not backed up and do not follow you to another machine. The contents already live in Google Docs, which is the backup; what is unbacked is the one-line description.

Never widen sharing on a Document whose ID the site has published. An INDEXED entry puts the Doc ID on a public page. That is harmless while the Document is unshared, because the ID is not a credential until link-sharing makes it one. Turning on “anyone with the link” later hands the contents to everyone who has ever read that page. To publish an INDEXED Document, move it to OPEN deliberately, having reread it, rather than by flipping a share setting.

6. Repository, Build and URL

The stack is a copy of what TrinityMetrics already runs, which is why the build is an afternoon rather than a project.

  • A public repository iamstein/collected-wisdom, Quarto website, output-dir of docs, published to GitHub Pages by a copy of .github/workflows/publish.yml.
  • docs/ and local/ both in .gitignore; local/** also in the Quarto render: exclusion list so a CI build cannot fail on a directory it does not have.
  • An entries.yml plus a Quarto listing with categories: true, copying the shape of projects/projects.yml and projects/index.qmd. Subject is a category rather than a folder, so a fourth subject is a new tag rather than a restructuring. Tier is a second category, so the listing can be filtered down to what is publishable.
  • The index page ends with a ranked entry path: three entries, in the order a first-time reader should take them, one clause each on why.

Start at iamstein.github.io/collected-wisdom and buy a domain later. The domain work on this site is recorded in the site-integration specification, Section 3: A records at the registrar, the custom domain in the Pages setting, and no CNAME file when publishing runs through a custom Actions workflow. It took a day, most of it waiting for nameservers, and it can be done at any point without changing the content.

7. Mirroring the Shared Documents

Copy _scripts/export-google-docs.sh from this repository. It fetches each Doc as a PDF into the rendered site after quarto render, and the workflow calls it in one line. Nothing is committed, so the PDFs are never stale and never enter git history.

Its export list carries OPEN rows only. An unshared Document returns Google’s sign-in page with a status of 200, and the script catches that by checking the first four bytes are %PDF and failing the build when they are not. Listing an INDEXED row there would break every build, which is the behaviour you want if a row is ever mistiered, and a reason not to generate the list from the inventory without filtering on tier.

Where a Document is OPEN, the mirrored PDF is published beside a link to the Google original, so a reader on a network that blocks Google Docs still gets it.

8. Acceptance Checks

None run; nothing is built. Run these against the first release.

# Check Result
1 Every inventory row carries a tier, and no row is left TBD not run
2 No file in the repository, in any commit, names a child not run
3 local/ is in .gitignore, and git ls-files local/ returns nothing not run
4 Every Doc ID in the export list belongs to an OPEN row, and downloads as a real PDF not run
5 No INDEXED or LOCAL Document is link-shared, checked in Drive rather than assumed not run
6 Every entries.yml entry resolves to a rendered page, a mirrored PDF, or a live outbound link not run
7 The site renders and publishes from the workflow with no manual step not run
8 TrinityMetrics and Collected Wisdom each link to the other not run

Checks 2 and 3 are the ones that cannot be repaired after the fact. A file committed and then deleted stays in the history of a public repository, so both belong before the first push rather than after it.

9. What Happens Next

In order. The first item blocks the rest.

  1. Sweep Google Drive and fill the Section 4 inventory. Doc ID and a one-line description per source. This decides whether to build the site, per Section 4.
  2. Set a tier on every row. INDEXED unless the title alone is sensitive, in which case LOCAL, or the piece is already published, in which case OPEN.
  3. Confirm in Drive that every INDEXED and LOCAL Document is unshared. The default has changed over the years and an old Document may be shared from a decision you no longer remember making.

The repository, the workflow, the listing and the export script follow from those three and need no further decisions.

Back to top