
A merge is not a promise to build. A Nano Collective whitepaper merges when the thinking is good enough to publish and invite public review - the build decision is a separate gate, made after the 30-day review window closes. https://nanocollective.org



A whitepapers section that accumulates ghost pages stops being a record of what was decided and why. Terminal statuses archive on a clock: Shipped and Paused after 90 days, Declined after 30, Superseded immediately. https://nanocollective.org


Picking `--use-llm` is a guess unless you can see what Turndown would have done. get-md's `--compare` writes both conversions to the same file, so the table that vanished in Turndown but survived the LLM is right there. https://github.com/Nano-Collective/get-md



Adopting a migration library? The worst part is the pre-step: special-casing old data with no version field at all. json-up treats missing `_version` as version 0, so legacy payloads run through the same `migrate()` call. https://github.com/Nano-Collective/json-up


Close the terminal at 6pm, open it at 9am, and the model picks up mid-thought. Nanocoder saves every session every 30s and keeps them for 30 days, so /resume last restores the chat, the model, and the working dir. https://github.com/Nano-Collective/nanocoder


5 numbers per test, not just pass/fail. nanotune benchmark reports total latency, TTFT, generation time, tokens out, and tokens/sec, so a green checkmark actually tells you how the model runs. https://github.com/Nano-Collective/nanotune