Personal Knowledge System for Developers: Capture, Triage, Review

The diagnosis first: you have a capture problem, not a memory problem. Storing 20 articles/week is easy; retaining any of it is impossible unless you build a recall loop. Storage format matters less than the pipeline. So the system is designed around review, not around the vault. COMPARISON Folder hierarchies Good for: projects, working files, anything with a single obvious home. Bad for: ideas and articles. Forces one category per note, breaks when topics overlap, and retrieval requires remembering where you filed it. The failure mode is "I know I saved that somewhere." Verdict: use for organization of projects only, not as the knowledge spine. Tags Good for: cheap multi-dimensional labeling, fast to apply. Bad for: structure. Vocabularies decay into tag spam, tags carry no context or relations, and recall still depends on you remembering the tag exists. Retrieval by tag = search with extra steps. Verdict: keep as a thin layer (2-4 tags per note max), never the primary structure. Backlinks Good for: the only system that captures context and relations. Notes accumulate connections, the graph surfaces serendipitous links, and retrieval becomes navigation instead of recall. Bad for: raw capture. Backlinks do nothing if you just dump full articles. They require atomic notes and linking discipline. Verdict: the spine for synthesis — the place distilled ideas live and connect. Spaced repetition Good for: the only mechanism that actually moves material into long-term memory. Forces active recall (recognition is not retention). Works on atomic facts, not articles. Bad for: whole documents. You cannot SRS a 2,000-word article; you can SRS the 3 ideas inside it. Verdict: the retention engine. Applied to distilled atoms, never to raw articles. Core finding: folders store, tags label, backlinks connect, SRS retains. You need all four, but in different layers. The mistake is picking one. RECOMMENDED STACK Capture: Readwise Reader (browser extension + app). One click saves an article; you highlight as you skim; highlights auto-sync. Handles the 20-article firehose with zero manual filing. Storage: Obsidian. Local markdown, backlinks, graph view, git-versionable. Everything is plain text you own. Review: Anki via the Obsidian_to_Anki plugin. Cards live as markdown in your vault, so the note and its cards are one artifact. (Alternative if you want zero plugins: Obsidian's Spaced Repetition plugin.) Weekly synthesis: Obsidian weekly note + backlinks. The distillation happens here, in the vault. Pipeline: Readwise Reader (capture) -> Obsidian (distill + link) -> Anki (retain). Three tools, one direction of flow. THE CAPTURE-TRIAGE-REVIEW LOOP (weekly budget, ~3.5 hr + 1 hr monthly) 1. Capture — 40 min/week (2 min/article, spread ~6 min/day) Skim every article as it arrives. Highlight at most 2-3 lines per article — the claim, the number, the trick. Never read fully in this phase; reading is for triage. Readwise syncs highlights automatically. No other action. 2. Triage — 30 min/week (one sitting, e.g. Sunday) Process the 20 highlight sets. Delete ~80% (highlights that no longer matter). File ~15% as reference (link-only note: title, URL, 1-line summary, 2 tags). Distill the remaining ~5% (1-3 articles) into atomic notes. 3. Distill — 60-90 min/week (same sitting) For each kept article: write one atomic note per idea (a single claim in your own words), link it to existing notes, and generate 1-3 Anki cards from the note. This is where retention is actually created. If you skip this step, the article was never read. 4. Review — 10 min/day (~70 min/week) Anki reviews (the cards from step 3). Re-read the current weekly note once. That's it. Daily review is the only non-negotiable step; it is what converts capture into memory. 5. Monthly review — 60 min/month Prune dead notes, merge duplicates, archive stale references, promote frequently-linked notes into a proper index note. Prevents the vault from rotting. Total: ~3.5 hr/week. If you only have 2 hours: cut capture to highlights-only (no full skims) and cut triage to 20 min — never cut daily review. WHAT TO DELIBERATELY NOT CAPTURE (and why) 1. Full articles. You cannot review 20 articles a week, and retention requires recall, not storage. Keep the link and the highlights; the body lives in Readwise if you ever need it. Storing full text is how you build a graveyard you never visit. 2. News, release notes, version bumps, ephemeral facts. They decay in weeks and cost maintenance forever. Capture the durable principle underneath, not the event. A knowledge system is for compounding knowledge, not newsfeeds. 3. Tutorials you already applied once. If you used it, it's in your code and your muscle memory. Duplicating it into the vault is dead weight that will never be re-read. 4. Anything re-findable in 10 seconds: API docs, language references, Stack Overflow answers. Store the pointer, not the content. Your system should hold what search cannot find — your synthesis, your decisions, your connections. 5. "Interesting" things you have no plan to review. Capture is cheap; maintenance is expensive. Every note is a liability: it must be linked, reviewed, and pruned. Unmaintained notes rot and bury the good ones. 6. Content captured out of obligation ("I should remember this"). The system's cost is review time, and every item you capture commits you to reviewing it. Only capture what you will actually put through the loop. The one rule that makes it work: nothing enters the vault unless it will be distilled, and nothing is distilled unless it will be reviewed. If it won't survive triage, don't capture it at all.

Why this specific combination fixes your problem: every other PKM setup fails at the same point — capture is frictionless, so the inbox grows, and review is the first thing dropped. This stack makes review the cheapest possible step (10 min/day of Anki) and makes capture the expensive one (you must distill or it dies in triage). The bottleneck moves from "remembering to file" to "deciding what's worth keeping," which is exactly where it should be. Retention is not a storage feature; it is the output of a loop you cannot skip. First-week action plan, if you want it running today: 1. Install Readwise Reader + Obsidian + Anki, and the Obsidian_to_Anki plugin. ~30 min. 2. Create three folders in Obsidian: Inbox, Notes, Reference. That's the whole hierarchy. 3. For the next 20 articles, do capture + triage only. Delete most, distill one article into 2-3 atomic notes, generate 5-10 cards. 4. Do the 10-minute Anki review every day for a week, no exceptions. 5. After 7 days, run the monthly-review pass early: prune, link, archive. Then judge whether the system holds — not by how much you stored, but by how many cards you can still answer. The metric that matters is not articles read or notes created. It is recall rate on your Anki cards and the number of times a vault note connects to a problem you are actually solving. Everything else is decoration.