Tim Ferriss — How to Remember What You Read

Why this is in the vault

This is the retention half of the ingestion-economics stack — Ferriss’s “How to Speed Read” doubles WPM, but speed without retention is just expensive recreation. The video opens with the clearest single-sentence framing the vault has on the topic: most reading is “just-in-case” (recreation, low retention) rather than “just-in-time” (retention-purpose, high index-quality). This is exactly the lens RDCO needs to apply to vault ingestion: most newsletter and YouTube ingestion is happening just-in-case (bias toward archiving everything), when it should be biased toward just-in-time (retain what serves an active question).

The tactical content is also concrete — Ferriss demonstrates a hand-built front-matter index pattern (page numbers + descriptions, with a phrasing-tag variant) and a multi-pass spaced-revisit cadence that builds a personal canon by attrition. This is the human-analog of what the qmd + graph reingest pipeline does for the vault.

Core argument

The unit of value in reading is not the finished book — it’s the indexed, retrievable insight. Most readers consume books just-in-case (“I might need this someday”), which produces minimal retention because there’s no purpose to the reading. Just-in-time reading — reading because you have a problem the book might solve — produces dramatically higher retention because the brain has somewhere to file the insight.

Ferriss’s tactical system:

1. Build a front-matter index in every book. Two patterns, often combined:
   • Page number + brief description of what’s interesting on that page
   • “PH” tag for phrasing — a separate index of beautifully-written passages worth re-reading for craft
2. Underline on the page when you index a phrasing pass. Marginalia is non-optional — write in the book even if it’s a hardcover. The cost of damaging the artifact is trivial compared to the cost of forgetting.
3. Use stars and circles to re-rank over time. Stars = potential next actions or experiments to try. Circles (or a second star) added on a re-read a week later = items that survived re-reading. Three-pass attrition (initial → stars → circles) builds a personal canon — the items that resonated repeatedly across multiple revisits.
4. Re-read the index, not the book. Ferriss flips through the front-matter index of every adjacent book before starting a new one in the same territory. The body is disposable; the index is the survival layer.
5. Spaced revisit cadence. 1 week, 1 month, 1 quarter. Each pass collapses the index further. Anki/SRS is optional — the physical index does most of the work because re-reading the index forces the recall.

Different books get different note systems. Steinbeck (literary fiction) gets a phrasing-heavy index with starred experiments. Goldberg (writing how-to) gets a chapter-tag index for repeat-applicable techniques. Marcus Aurelius (canonical philosophy) skips the phrasing index entirely and gets star-only mining for principles.

Mapping against Ray Data Co

• The “just-in-case vs just-in-time” frame is a sharp critique of current vault ingestion. RDCO is currently optimized for capture — newsletters, YouTube, web articles flow into ~/rdco-vault/06-reference/ with assessment notes. The retention/usage rate is unknown but likely low. Most ingestion is just-in-case. Ferriss’s argument is that this is structurally low-retention by design. Counterpoint: the vault isn’t only the founder’s memory; it’s the COO agent’s working memory too, and qmd + graph reingest gives the agent retrieval that the human can’t match. So the just-in-case bias may be correct for the agent even when wrong for the human.
• The vault’s qmd lex+vec+hyde stack IS the retrieval layer Ferriss is hand-building manually. What Ferriss builds in front-matter pencil indexes, the vault builds via the typed graph + the qmd index. The vault is a couple orders of magnitude more powerful than handwriting. The lesson transfers in the other direction though — the star + circle attrition pass is missing from RDCO. The vault has no “this entry survived a re-read” signal. Adding a revisit_score or attrition_pass field to vault frontmatter would let /vault-health surface high-survival entries.
• The “different book, different note system” principle applies directly to RDCO’s ingestion skills. Currently /process-newsletter, /process-youtube, and /process-inbox produce structurally similar assessment notes. Ferriss’s point is that the note system should match the source type. A Stratechery deep-dive deserves a different extraction template than a Tim Ferriss tactical video. This is a concrete refactor: each ingestion skill should have a per-source extraction profile, not a one-size-fits-all assessment template.
• Spaced revisit cadence is a missing /vault-health subroutine. Ferriss’s 1-week / 1-month / 1-quarter pattern could be implemented as: every Monday, surface 5 entries from 1 week ago, 5 from 1 month ago, 5 from 1 quarter ago, prompt founder to mark which still resonate. Survivors get a revisit_score++. This is the RDCO version of Ferriss’s circle-marking ritual.
• “Re-read the index, not the book” is the qmd-first principle. Already partially implemented — /research-brief and /morning-prep use qmd queries instead of full-vault reads. Worth elevating to an explicit principle in SOUL.md: parent context never reads a vault doc raw if a qmd query against the index would surface the relevant section.
• The “personal canon by attrition” is the missing concept-page layer. RDCO has 1490 reference docs but few concept pages. Ferriss’s attrition model suggests the concept pages should be built from the survivor entries — the references that have been cited repeatedly by other entries in the vault. The graph reingest already produces citation counts; using them to surface concept-page candidates would be a clean implementation.

Open follow-ups

• Add a revisit_score field to vault frontmatter. Initially zero; bumped each time the founder marks an entry as still-resonant during a /vault-health revisit pass. Lets the graph surface “personal canon” entries.
• Build /vault-revisit subroutine in /vault-health. Weekly cadence: surface 5 entries each from 1 week / 1 month / 1 quarter ago, prompt for resonance check. Two-minute exercise for the founder. Generates the attrition signal.
• Refactor ingestion skills with per-source extraction profiles. /process-newsletter currently uses one template for all senders; should have per-sender profiles (Stratechery deep-dive vs Every concept piece vs Lenny’s deep-research vs Cedric Chin tactical). Same for /process-youtube — Tim Ferriss tactical vs Practical Engineering pedagogical vs IndyDevDan demo-driven.
• Concept-page candidates from graph citation counts. Build a /surface-concepts subroutine: query the graph for reference docs cited 5+ times by other docs but with no corresponding concept page. Surface as candidates for founder review.
• Make “qmd-first, raw-read second” explicit in SOUL.md. Parent context should never read a vault doc raw if a qmd query against the index would return the relevant section. This is the Ferriss “re-read the index” principle for the COO agent.
• Tim Ferriss as a recurring channel — pre-build the index page. This is the third Ferriss video in today’s batch (Speed Read + Sharpen Thinking + Remember What You Read form a tight reading-writing-retention triad). Once the cluster reaches 5+ entries, build ~/rdco-vault/05-concepts/tim-ferriss-index.md to prevent orphan accumulation.

Related

• ~/rdco-vault/06-reference/transcripts/2026-04-19-tim-ferriss-how-to-remember-what-you-read-transcript.md — raw transcript
• ~/rdco-vault/06-reference/2026-04-19-tim-ferriss-how-to-speed-read.md — companion video; speed without retention is wasted throughput, this is the retention half
• ~/rdco-vault/06-reference/2026-04-19-commoncog-reading-quickly-reading-lots.md — Cedric Chin’s strategic case for high-volume reading; Ferriss provides the per-book retention discipline that makes Chin’s volume sustainable