Standing Questions

The questions you asked months ago, and the notes that quietly answered them.

Every vault has open questions in it. Why did we drop the CRDT plan? How long should onboarding take? Is the storage rewrite worth it? You write them down, and then you keep writing — and eight weeks later you write the note that answers one of them, and you never notice, because nothing in Obsidian connects a note to a question you asked in January.

Standing Questions makes a question a first-class object: it has a state, it has sub-questions, and it watches for its own answer.

Standing Questions never edits your notes without being asked. Leads are notify-only by default: it tells you, and you click. The only thing it ever writes on its own is a status: line in a question's frontmatter, when its sub-questions are all answered — and that is a toggle you can turn off.

How it works

Add two lines to a note:

---
type: question
status: open        # open | partial | answered
---

That note is now a question. It appears on the question board, it is watched, and every note you write is matched against it.

Sub-questions

A question can be part of a bigger one. Say so in frontmatter, or inline:

parent: "[[Why did we drop the CRDT plan?]]"

parent:: [[Why did we drop the CRDT plan?]]

Then the states propagate upward:

Propagation only ever moves a question forward. If you wrote status: answered on a parent, a sub-question you never got round to closing will not silently reopen your answer.

A parent:: loop cannot hang it. A → B → A is one typo away, and a loop would make a naive walk run for ever. Loops are detected, reported on the board and in settings, and every question inside one keeps exactly the status you gave it.

Leads

When you save a note, Standing Questions asks: does this answer anything?

If it does, you get a notice — "This may answer a question you asked in January: Why did we drop the CRDT plan?" — with a button. Press it and the lead is written into the question:

## Leads
- [[2026-07-14 Storage benchmarks]] — 0.81 · "…the merge cost scaled with the number of replicas…"

Nothing is written until you press it. (There is an opt-in setting that appends very strong leads without asking. It is off.)

What you get

Free — all of it

• Questions as objects: frontmatter contract, three states, a question board.
• Sub-questions, with status propagated upward, and loop detection.
• Keyword and link leads — matches a new note against your open questions on shared significant words and on links they have in common. The link signal is the one keyword matching normally cannot see: you and this question both link to the same note.
• Works on mobile. No engine, no license, no network, no note limit, no nag screen.

Pro — $29 one-time, unlocks all five Second Read add-ons

• Semantic lead detection. The note that answers "Why did we drop the CRDT plan?" without ever using the word CRDT. Your note is compared against every open question by meaning, not by keyword.

Second Read Pro — $29 one-time. One key unlocks Pro in all five Second Read add-ons (Note Decay, Standing Questions, Effort Index, Prior Art, Unwritten). Buy here: https://buymeacoffee.com/vaultspotlight/e/560213 — the key is emailed to you automatically, within seconds (delivery is fully automated); paste it into Settings → Standing Questions → License. It is verified offline (Ed25519) — no account, no server, no subscription.

Pro needs the local semantic engine (below). One key unlocks Pro in all five Second Read add-ons: Note Decay, Standing Questions, Effort Index, Prior Art, and Unwritten. Licenses are verified offline with an Ed25519 signature built into the add-on. No account, no server, no network request.

Threshold, and why it is set where it is

The lead threshold is a cosine similarity, and it defaults to 0.75.

That is deliberately conservative. Below about 0.6, notes that merely share a subject start matching — and an add-on that interrupts you about those gets muted within a day, and a muted add-on has no features. At 0.75 a lead is nearly always a real one. It is a slider; if you would rather see more, lower it.

Disclosures

Please read these before installing. They are the things this add-on does that you cannot see.

The Pro features download and run a program — and only if you ask

Semantic lead detection needs a local semantic engine: a self-contained program that runs on your computer, embeds your notes locally, and opens no network connections of its own.

• Nothing is downloaded unless you click "Download engine" in settings, and confirm a dialog that names the exact URL, the version, the SHA-256 checksum, and the directory it will be installed into, and states plainly that a program will be run.
• The download is verified against a checksum built into this add-on before anything is extracted, before anything is made executable, and before anything is run. If it does not match, the file is deleted and nothing executes.
• It is never updated silently. If a newer engine exists, you get an "Update engine" button. You click it, or you do not.
• This add-on accesses files outside your vault. The engine is installed in your system's application-data folder — %LOCALAPPDATA%\second-read-engine on Windows, ~/Library/Application Support/second-read-engine on macOS, ~/.local/share/second-read-engine on Linux — and its index lives there too. It is never written into your notes folder. (Two reasons: Obsidian Sync and Dropbox replicate .obsidian/plugins/**, and you do not want 100 MB of it on your phone; and a note-taking app that drops an executable into your documents folder is exactly the pattern that makes security software escalate.)
• The engine runs only while Obsidian is open. It is killed when the last Second Read add-on unloads, it dies when Obsidian's process ends, and it exits by itself after ten idle minutes.
• Source: github.com/israerusan/second-read-engine. It is open, and the release is built in public CI.
• Desktop only. On mobile there is no engine and no download button — and the whole free tier still works.
• If your system refuses to run a downloaded program (antivirus quarantine, a noexec mount, Flatpak confinement), point the add-on at an engine binary you installed yourself: Settings → Semantic engine → Path to an existing engine. The download is a convenience, not the mechanism.

What is sent where

Nothing leaves your machine. The engine receives the text of your notes over a pipe — not a network socket — and returns numbers. The license check is a signature verification done offline. This add-on makes exactly one kind of outbound request in its entire life: downloading the engine from GitHub, after you click the button.

What it writes

• A status: line in a question's frontmatter, when its sub-questions have moved it. Toggleable, and it only ever writes when the value actually changes.
• A ## Leads line in a question's note — only when you press "Add lead", unless you opt in to auto-append.
• Nothing else, ever.

Commands

No default hotkeys — bind your own in Settings → Hotkeys.

Install

Community plugins (recommended): open Settings → Community plugins, search Standing Questions, and install it — one click, auto-updates.

Manual install: copy main.js, manifest.json, and styles.css into <vault>/.obsidian/plugins/standing-questions/ and enable it in Settings → Community plugins.

Development

npm install
npm run sync:shared     # vendor the shared Second Read core
npm run lint            # typecheck + the review bot's own eslint ruleset
npm test                # lint + drift check + the whole suite
npm run build           # production bundle
npm run install:vault -- <path to a vault>

src/shared/ is vendored from obsidian-plugin-core and must never be edited here — npm test fails on drift.

License

MIT. See LICENSE.