Knowledge systems / Search / Provenance / Product engineering / Publishing · 5 min read
A bookmark is not a research system
What I changed in Kiln to keep ideas useful without losing the evidence, the uncertainty or the original source.
Muhammad Musa
Saving was the easy part
A saved link is a promise to return. It is not yet a useful note. The interesting part might be a diagram halfway through a video, a tool mentioned without a link, or a technique that only works under conditions the creator never explains. Collect enough of those promises and the library becomes another inbox.
I built Kiln to close that gap. A source can become a practical guide with citations, files, commands and project ideas. The important design constraint is that the guide must not replace its evidence. I want to be able to ask where a statement came from after the polished explanation has made it feel obvious.
Split capture from research
The earlier workflow tried to acquire and understand the media itself. That made authentication, frame selection and transcription part of every research task. The current workflow starts from a Markdown dossier: the original URL, transcript or visual observations, named tools, claims and gaps are already captured.
Kiln saves an immutable revision before it changes the database. An identical completed revision is not new work; a changed revision becomes pending. From there, research can concentrate on checking current sources and producing something useful. A source’s confident wording is evidence to inspect, not permission to run its commands.
Merge the idea without erasing its history
Two sources can discuss the same practical problem in different language. Keeping both as isolated cards duplicates the work; merging everything with a shared tag loses useful distinctions. I compare purpose and subject, using specific terms, tools, repositories and original URLs to find candidates. A broad category alone does not establish that two items belong together.
When sources do become one topic, the originals remain attached. Membership can be reversed, and earlier item links still resolve. That makes organization a decision I can revisit instead of a destructive cleanup operation. Disagreements and different methods should survive the synthesis when they matter to the task.
Keep finished research separate from finished work
A good tutorial can suggest ten projects. It cannot make all ten projects real. Kiln tracks the owner’s outcome separately from whether research is complete, and building a proposed project requires a separate queue decision. That boundary is small in the interface and important in the data model.
The same distinction applies to unresolved evidence. Missing sources or a failed artifact leave an item pending with a retryable error. I would rather retain a useful incomplete record than disguise a missing step as a successful pipeline run.
A publish button is only the beginning
The local workspace and public library serve different purposes. The public site is a static snapshot. It receives an allowlisted set of fields and artifacts, while private notes, raw intake, credentials and machine paths remain local. Sharing a common interface does not require sharing all the underlying capabilities.
After deployment, the workflow compares the live files with the exported hashes. A successful export or push is not sufficient proof that the reader is seeing the intended version. This is the same lesson a small visual website change can teach: verify the thing the person receives, not only the command that produced it.
The result I want from Kiln is modest but useful. Find an idea, understand how to apply it, see what is uncertain, and follow the trail back. The library earns its place when that takes less effort than reconstructing the research from the original bookmarks.