Khoj Research — everything, in one place

The curator composes relationships and intervals, not objects — it can value the gap as much as the node, and treat removal as a creative act.

AddWarburg's "good neighbour" (the book that sits best beside a cluster); shakkei "borrowed-view" tier; the anchoring counterpoint (make a one-sided cluster two-sided); wunderkammer cross-domain juxtaposition; yohaku — tag some gaps "leave blank."

SubtractAAMD deaccession (redundancy / low idea-unit yield, never resale value); "kill your darlings" (over-loved dense zones); the piece that no longer converses (zero bridge-betweenness); prune for the whole, not the specimen.

SerendipityChance operations (Cage/Duchamp); the exquisite corpse; constraint as a generator (Oulipo / Sol LeWitt); the generative mis-shelving.

Recommender "beyond-accuracy" objectives + the cognitive science of discovery — computable functions for novelty, coverage, and expected learning.

AddActive learning / uncertainty sampling; query-by-committee (buy where your own models disagree); Bayesian surprise (KL prior→posterior); Marginal Value Theorem patch-leaving; low-coverage regions.

SubtractRedundancy / low marginal self-information (Vargas & Castells); popularity-bias overweighting; over-exploited patches; Schmidhuber's boredom (zero compression progress).

SerendipityUnexpectedness × relevance (surprise with a relevance gate); calibrated recommendation (Steck); learning-progress curiosity (the edge of current threads).

Subset selection in embedding space with provable guarantees — the one lens that supplies a tunable dial between coverage and dispersion.

AddSubmodular greedy (1−1/e ≈ 0.63); facility-location / max-coverage; gap-geometry (largest-empty-ball via Voronoi, KDE); persistent-homology holes (TDA); MMR as the cheap ranker.

SubtractCoreset / submodular pruning; k-medoids; CUR leverage scores; the outlier-protection rule — cull the redundant, shield the low-density outlier (the lens's signature inversion).

SerendipityRandom-walk-with-restart; Determinantal Point Processes (DPP sampling); entropy-regularized / temperature τ; bandits (Thompson/UCB/Gittins); the secretary rule.

The only lens that measures the shape of the coverage itself and treats gaps as objects with a size, location, and scale.

AddBox-counting the coverage (empty boxes adjacent to occupied ones, at the scale that sags); feed the tail against preferential attachment; grow by an L-system rule.

SubtractPrune monofractal spikes (over-dense clumps); cut redundant self-similar copies; the Devil's-staircase test for dead plateaus (zero thread-participation).

SerendipityZoom-level leaps (near in pattern, far in catalogue); strange-attractor random restart; the un-taken forking-path branch.

The antilibrary's own Eco→Taleb bloodline — each book a financial position, the library a portfolio; the scarce resource is attention, not shelf space.

AddThe barbell (canonical core + a wild-bet sleeve, hard ratio); buy the option not the certainty (rank by convexity); value of waiting (Buy-now vs Watchlist); diversify against correlation; the secretary rule for a fixed budget.

SubtractVia negativa; the anti-Lindy hype cull; redundant-given-a-dominating-book; low-optionality dead weight; defeat endowment/sunk-cost (reframe as opportunity cost, demote-don't-delete, actively push a removal quota — Klotz).

SerendipityInstitutionalize the tail (the wild-bet sleeve is the engine); convexity tolerates being wrong (never penalize the sleeve's hit-rate).

The library as a living community whose health depends on diversity, not size — and the source of the only field-tested subtract method (CREW/MUSTIE).

AddColonize vacant niches (Hutchinson / low-density regions); the keystone introduction (rank by Δ bridge-centrality — "the one book that reorganizes everything"); evenness enrichment (Pielou's J′); ecotone / edge seeding; r-mode vs K-mode.

SubtractMUSTIE weeding (Misleading · Ugly · Superseded · Trivial · Irrelevant · Elsewhere); cull invasive monocultures; thin same-niche competitors; protect rarity — the sole occupant of a niche is exempt.

SerendipitySeed rain one hop out; the ecotone (boundary) surprise; seed-bank germination of dormant unread books; patch depletion.

Signaturesvādhyāya (self-study / lifelong recitation); śruti vs smṛti (the heard-eternal vs the remembered-mutable); adhikāra (fitness to receive); anadhyāya (codified fallow days).

BehaviourRedundancy-encoded recitation (error-correcting chant); the book as sacred (Sarasvatī); guru–śiṣya transmission conferring the right to use knowledge; Ranganathan's "a library is a growing organism."

Khoj takesADD by readiness (like a paramparā), not popularity; KEEP a core for lifelong return; SUBTRACT the perishable (smṛti) before the durable (śruti); SERENDIPITY = re-cut facets + deliberate fallow days.

Signaturecángshū (藏書, "to hide/store books" as cultivation); shūxiāng (書香, the fragrance of inherited learning); dúshū zhǒngzi (the seed-corn of reading).

BehaviourVermilion collector's seals + colophons accrete on a book (provenance often outweighs text); "sunning the books"; keeping collections whole (Tianyi Ge); cóngshū (keep whole) vs lèishū (atomize).

Khoj takesADD to complete a lineage/set; KEEP whole, surface each book's seals/colophons; SUBTRACT only with justification (the literary-inquisition risk) — prefer air-and-reshelve; SERENDIPITY by reader-lineage.

Signaturetsundoku (積ん読 — acquiring and piling up unread; affectionate, not guilty; an 1879 pun); zōsho (one's holdings); bunko (the pocket paperback).

BehaviourJimbōchō, the antiquarian district you browse by scent; the book as a cared-for object; Kenkō — "imperfect sets are better"; yohaku no bi (the beauty of meaningful blank).

Khoj takesADD ahead of reading (never scold); KEEP by what a book represents; SUBTRACT redundancy, never unreadness; SERENDIPITY like Jimbōchō, not Amazon.

Signatureseonbi (선비, the frugal Confucian scholar — study to cultivate the self, not accumulate; the small internalized library outranks the large unread one).

BehaviourJikji (oldest movable-metal-type book, a devotional text); the Tripitaka Koreana (81,258 woodblocks carved as a prayer, preserved 750+ years); hanji paper built to last a millennium.

Khoj takesADD the enduring (books you'd re-carve); KEEP as a depository, not a display case; SUBTRACT at the gate (refuse to admit the perishable); SERENDIPITY = the already-owned-but-under-read.

Signatureisnād (the chain of transmitters back to the source — a weak chain discounts the content); ijāza (the personal licence to pass a text on); samāʿ audition certificates written into manuscripts.

BehaviourThe booksellers' market as an intellectual salon; Ibn al-Nadīm's Fihrist (indexing every Arabic book); the dread of the broken chain (Baghdad 1258).

Khoj takesADD books with their isnād (lineage of ideas) over orphan bestsellers; KEEP by chain-position, writing the collection's own samāʿ certificate; SUBTRACT the broken isnād; SERENDIPITY = the un-lineaged stranger's manuscript.

Signaturegenizah (worn sacred texts are retired and buried, never destroyed — the Cairo Geniza kept ~400,000 fragments); chavruta (paired argumentative study); the daf (the Talmud page).

BehaviourThe daf stages a multi-century argument on one spread (text ringed by disputing commentaries); pilpul; the annual Torah cycle; elu v'elu — two contradictory schools both kept.

Khoj takesSUBTRACT = a genizah tier (retired, dated, reversible, never deleted); KEEP a book with its disputants; ADD for contradiction; SERENDIPITY = the Cairo-attic accident (resurface the no-longer-fitting).

SignatureThe access-engine (Dewey's democratizing classification; Carnegie's free libraries) vs. the fetish-object (Dibdin's Bibliomania; the Roxburghe Club; the association copy).

BehaviourStandards and reform vs. connoisseurship ("this copy, not the book"); BISAC as retail categories built to sell; Penguin and Google Books dissolving "having," leaving "having read" the only scarcity.

Khoj takesADD with the discovery engine inverted (the most generative book you won't finish); KEEP by association-copy / biography, not finish-rate; SUBTRACT the market-redundant, never the market-orphaned; SERENDIPITY = the sale-room adjacency shock.

SignatureThe Mundaneum — Otlet & La Fontaine's 15M+ index cards meant to catalogue all knowledge, with a search-by-mail service; Otlet coined "documentation."

BehaviourThe card as atomic unit (the Monographic Principle); the UDC's +, :, / symbols as proto-hyperlinks; the dreamed réseau mondial; the collapse (Nazi destruction, 1940).

Khoj takesADD along the réseau (the surprising X : Y link, not just the node); KEEP recombinable facts; SUBTRACT the completeness reflex ("completing the set"); SERENDIPITY = the anti-Mundaneum, steering toward the blank map.

SignatureEco's unread-as-asset (the library as a research tool, not an ego-appendage; the antischolar); "How to Justify a Private Library"; roots in the studiolo (the study as a self-portrait) and the zibaldone.

BehaviourUse, not veneration — Petrarch the book-hunter (collecting as retrieval of the missing); Aldus Manutius inventing the portable book and italic type.

Khoj takesADD to enlarge the frontier of the unread; KEEP the studiolo (unread volumes are load-bearing); SUBTRACT the read-and-done trophy, not the unread; SERENDIPITY = the zibaldone jumble.

SignatureSonglines / the Dreaming (the land as archive); whakapapa (genealogy as structure) + tapu (restricted knowledge); the Andean khipu (a text mute without a trained reader).

BehaviourMemory-as-library; knowledge bound to person, place, season, permission; community-specific classification (Brian Deer, the opposite of Dewey); data sovereignty (the CARE principles); some things not to be shown (ATSILIRN Protocol 6).

Khoj takesADD by relation/provenance (and practices/places, not only books); KEEP with a permission/sensitivity layer (a "restricted / not-for-display" state); SUBTRACT as restraint ("not yours to hold"); SERENDIPITY = walking a relational path.

SignatureThe Zettelkasten (Luhmann's slip-box — "connecting, not collecting"; atomicity, unique identity, connection); Bildung (self-formation through deep reading).

BehaviourWarburg's "law of the good neighbour" + the iconology of the interval; Benjamin's "chaos of memories"; Schopenhauer — buying books mistaken for buying the time to read them; the art of not reading.

Khoj takesADD to close an interval between two loved books; KEEP the well-connected nodes; SUBTRACT the disconnected trophy; SERENDIPITY = proximity-with-a-gap.

SignatureThe bouquiniste (the Seine's green boxes — you can't search, only rummage; the find is unrepeatable) and the flâneur-reader (discovery as drift, not query).

BehaviourPerec, "Arranging One's Books" (every scheme breaks — the library lives in the order/disorder tension); Mallarmé; Derrida's archive fever; the uncut, literally sealed page.

Khoj takesADD via browse-adjacency (a "quai" of near-but-unsearched titles); KEEP by letting the user arrange (offer rival orderings, refuse a winner); SUBTRACT gently (unread = a sealed promise); SERENDIPITY = design for disorder.

Signaturesamizdat (forbidden texts retyped on carbon paper, passed reader-to-reader at risk) and tamizdat; "a poet in Russia is more than a poet." Value = risk × trust.

BehaviourThe human as library (Akhmatova's Requiem memorized, the paper burned); the "thick journals"; the domestic shelf as soul; Fyodorov (the library that defeats forgetting).

Khoj takesADD by transmission-worthiness (would you memorize it, pass it on?); KEEP toward the internalized; SUBTRACT the merely-owned; SERENDIPITY = the hand-passed book (the anti-algorithm).

MetabolismCumulative, not superseding — you read Plato/Kant in the original; primary texts have near-zero obsolescence, only secondary literature dates. Progress by aporia.

Add / SubtractADD the primary voice and the missing interlocutor (a gap = an unanswered objection). SUBTRACT only superseded secondary material — and demote, don't delete.

MetabolismA frozen core + a boiling shell; scripture is re-consumed on a clock (the annual Torah cycle, Daf Yomi). Grows by lamination, not replacement.

Add / SubtractADD rival cores > recovered primaries > translations >> commentary. SUBTRACT never the core (retire = genizah); only devotional ephemera.

MetabolismNear-zero decay (Euclid still holds). Only notation and pedagogy age — never the result. The canonical shelf is tiny and permanent; Erdős's "The Book" never needs updating.

Add / SubtractADD the missing prerequisite (fill the DAG bottom-up); a better exposition. SUBTRACT only exposition swaps, superseded editions, duplicates.

MetabolismThe field that disowns its past by design (Popper's graveyard). Primary ages fast; the review/textbook/classic barely decays. arXiv preprints in symbiosis with consolidating reviews.

Add / SubtractADD the review article (highest-leverage), the current textbook, the classic. SUBTRACT superseded editions (hard); absorbed/stale primaries — never a review or classic on age.

MetabolismActionable knowledge decays in ~18–24 months; the evidence hierarchy retires each tier below; the "five-year rule." Yet the humanistic spine (Osler, Gawande) ages like philosophy.

Add / SubtractADD the current guideline / review / edition, and the humanistic classic. SUBTRACT aggressively, harm-weighted — a superseded clinical edition is Misleading, not merely stale.

MetabolismSplit — the practical layer (statutes, cases-as-authority) is the most perishable in any professional library ("old law is dead law"); the jurisprudential canon outlives nations.

Add / SubtractADD the current edition, the leading treatise, the landmark. SUBTRACT the practical (superseded/overruled — the citator's red flag), preserve the canon. Question: authority, or reasoning?

MetabolismPrimary sources appreciate with age (finite, non-reproducible); secondary interpretation depreciates ("each generation rewrites history") — but transmutes into a primary source on the mind that wrote it.

Add / SubtractADD the primary source (always accretive), the interpretation, the revisionist. SUBTRACT never a primary; demote a superseded secondary from the truth-shelf to the historiography shelf.

MetabolismThe book is the thing — a novel is not obsoleted by a later novel. The canon is near-permanent (Bloom, Calvino); genre churns; re-reading and translation refresh.

Add / SubtractADD the canonical gap, the buzzed novel, the unrepresented tradition (gaps are shapes, not titles). SUBTRACT only the read-once disposable — the signal is "re-readable to you?", not "unread?"

MetabolismObsolescence ≈ 0 (Sappho, Rumi, Bashō are load-bearing). A translation adds a facet, doesn't obsolete. The anthology tradition proves the native act is gather-and-keep.

Add / SubtractADD anthologies, translations (they stack), chapbooks. SUBTRACT near-never — only a weak contemporary volume that didn't land, and even then demote.

MetabolismThe biggest field here — and the one place owning a book can make you actively wrong. The framework endures (James, Weber, Kahneman-as-synthesis); the finding perishes (the replication crisis).

Add / SubtractADD the framework, the rigorous synthesis, the replicated result. SUBTRACT the single-study paperback whose effect failed to replicate — MUSTIE-M, worse than an empty slot.

MetabolismThe fastest-perishing serious domain — survivorship bias industrialized; the flagships self-destruct (Good to Great's firms went bankrupt). The more celebrated the case study, the more suspect the book.

Add / SubtractADD the Lindy classics, operator memoirs (skin in the game), counter-hype books. SUBTRACT on recency + trend + no-track-record; keep one debunked bestseller as a specimen.

MetabolismBimodal, not a half-life — the version/tool layer rots in ~2 years (the current version is on the web); the foundational canon (TAOCP, SICP, K&R, Brooks) is astonishingly Lindy.

Add / SubtractADD the timeless foundation and the genuine frontier. SUBTRACT the version/tool layer aggressively (especially "available elsewhere"); fence the canon as un-cullable regardless of date.

MetabolismThe book is partly the artwork; interpretation revises glacially, reproduction improves monotonically. The catalogue raisonné is a near-permanent reference.

Add / SubtractADD the catalogue raisonné, the definitive monograph, the exhibition catalogue. SUBTRACT near-permanent keep + a unique quality-based replace (better printing of the same content); churn only trend/manuals. Never subtract on utility — the book is also an object.