Five frontmatter fields turn a note into a checkable claim
YOINK starts with an ordinary folder of markdown files, the same format Obsidian understands. Add a claim, settlement date, reading source, test, and confidence value to a note's frontmatter, and the tool treats it as a prediction. When the date arrives, yoink settle reads a number, applies the declared comparison, and writes the outcome back into that file. The note keeps the thought and the result together.
A claim can read JSON, a dated CSV cell, a public EVM chain, or a manually supplied value tied to a URL. Tests are deliberately narrow: greater than, less than, equality, inequality, and an inclusive range. YOINK rejects vague claims, past settlement dates, unknown source forms, and confidence values at exactly 0 or 1. That constraint is useful because a score built on uncheckable prose would be decoration.
The score makes overconfidence visible in the same vault
Settled claims feed a Brier score, hit rate, stated-confidence average, and the gap between confidence and accuracy. The fixed demo contains 35 notes, 5 settled claims, and 2 open ones. Its author says 75 percent on average, lands on the correct side 40 percent of the time, and gets labeled 35 points overconfident. Those are demo fixtures, not evidence that the program improves forecasting.
YOINK also calls a note dead when it has no valid links, has not been reopened after its creation day, and carries no open claim. In the demo, 21 of 35 notes meet that definition. The rule will annoy people who use a vault mainly as an archive, yet it exposes a familiar second-brain failure: capturing material feels productive even when nothing later points back to it.
What happened when we ran it
Our fresh Python 3.12 sandbox installed commit a262dda in 34 seconds. The 48-file checkout contained about 5,558 lines of source and occupied 8.9 MB. Installation left 38 packages and 41 MB on disk. A build completed successfully in 8 seconds, and pip-audit reported 0 known vulnerabilities.
Pytest failed after 12 seconds with 200 passed and 1 failed out of 201, plus 27 passed subtests. The red case was test_the_headline_reads_the_way_the_screenshot_shows. The program produced the documented demo summary with an extra 1 due now suffix, while the expected string ended at +35 pts overconfident. The log supports calling this fixture or display-text drift; it does not show damage to notes or a failed settlement.
Plain files stay portable, while the ledger is weaker than proof
The vault has no database or search index. Every scan rebuilds links, tags, search results, and pulse data from markdown on disk. YOINK writes Obsidian-style frontmatter and [[wikilinks]], so the folder remains usable in another editor if the program disappears. Its parser handles strings, numbers, booleans, dates, timestamps, and flat lists rather than the whole YAML language.
Reading and settlement history lives separately in .yoink/ledger.jsonl. Each line includes a hash of the prior entry, so editing one event breaks the chain and yoink doctor can point at it. The README is candid about the limit: a person who controls the file can rewrite the entire chain. Use it to catch casual self-editing, not to prove a forecast's timing to an employer, customer, or regulator.
Chain settlement can read balances but cannot sign transactions
The chain adapter supports Ethereum plus Robinhood Chain mainnet and testnet in the documented source. It checks eth_chainId before reading block height, gas price, balance, or transaction count. The implementation contains no private-key handling, signing, or send-transaction method. A claim stays open when its endpoint cannot be read instead of accepting a guessed value.
That design reduces wallet risk, but it still introduces an external dependency for chain claims. An incorrect or unavailable endpoint can postpone settlement. File and CSV sources avoid that network dependency, while manual: requires the user to enter a value and records its URL. YOINK grades the number it receives; it does not independently decide whether a manual source is trustworthy.
The README now labels its animations as drawings
Open issue 1 argues that the animated images look like terminal features the app does not provide. The current README answers part of that complaint directly: its still terminal capture comes from a pseudo terminal, while 3 animations are drawn explanations. It also says the graph shown in an animation is a way to view the vault rather than a separate pane reached with its own key.
GitHub showed 382 stars, 1 open issue, and a last push on September 13, 2026, two days after the repository was created. There is no tagged GitHub release, while pyproject.toml calls the package version 0.1.0 and beta. The MIT license and detailed docs make inspection easy. The failed demo assertion and presentation dispute still put YOINK in the interesting-new-tool bucket, not the dependable-record-system bucket.