parrik

Method

A self was never flat

A schema for personal memory in LLM conversations · You are not what you said six times. You are what you did, on the days you did it.

Alex is 41. Senior editor at the University of Chicago Press. Single parent of her daughter Mira, fourteen. Nine months into Chicago, mostly alone.

One evening she asks the model whether there’s a pattern in how she handles conflict at work. The reply comes back confident, fluent, slightly wrong: she stays in misaligned situations because she’s afraid of burning the relationship. It sounds like something she’s said. It is — six times, as self-diagnosis. She asks it to show its work. Same claim, attributed to her. No independent episodes.

She had been talking to a mirror.

The mirror is now a measured effect, not a metaphor. A two-week field study of real-world LLM use found that condensed user-profile memory produced the largest sycophancy amplification of any personalization feature tested.1

A model’s memory is a list — bullets, facts, things you said that carried weight. A claim stated once feels the same as a claim grounded in five independent events. The fix is not more memory. The fix is shape. The scaffold — open, MIT-licensed — is at github.com/parrik/know-thyself.

Open Alex's full dashboard →

How to read the graph
  • Rreference — biographical fact
  • Oobservation — dated event
  • Poverlap — pattern across observations
  • Nnovel — tentative, single-derivation
  • Eemergent — claim at an intersection
  • EQequivalency — bridge to external framework
  • PRpractice — operating method
  • OQopen — unresolved question
  • solid — grounds
  • dotted — emergent from

A worked example. Hover any node to see what it is, why Alex cares about it, and what knowing it unlocks for her. The mechanism walk below shows how the shape gets built.

The shape, through Alex’s year

After the mirror problem, Alex wanted a memory that could not do this. Not a more careful model — a memory whose shape made repetition-as-evidence structurally impossible. The answer: the memory had to have types.

The academic frame for memory-with-types is already named. Sumers, Yao, and Narasimhan’s Cognitive Architectures for Language Agents (2023) carves agent memory into working, episodic, semantic, and procedural.2 The eight types here are an opinionated refinement of the semantic side — what makes a claim about a person earn standing.

Reference what is

Facts that don't change. The floor the rest of the graph stands on.

Forty-one. Senior editor at UChicago Press. Moved from Brooklyn last August. Divorced four years, amicable. Daughter Mira, fourteen. Older sister Helen, died at twenty-three in 2007.

Observation what happened

An episode, dated and bounded. Stored as it was, without guessing what it meant.

September through November: Alex's first three months in Chicago. Mira came home quiet about a girl at lunch. The Sunday-morning run Alex had kept for six years quietly stopped. A second observation came in March: the morning of a hard acquisitions meeting, Alex had run beforehand and held her position on a book more clearly than she had in weeks. Two episodes, held separately.

Overlap the same shape, twice

Two observations sharing one structure. The model can defend the pattern. Contrast: one claim said six ways is one derivation, not six.

First: running stopped, work extended, Mira struggled. Second: running came back, work stabilized, Mira climbed. One shape — when the running holds, everything else holds.

Novel one derivation, held honestly

A claim with a single source. The model's honesty that it is guessing — flagged tentative, waiting for a second independent stretch.

For Alex, isolation is upstream of routine breakdown. One episode supports it: she stopped responding to her two friends in Brooklyn, and the running stopped the week after.

Emergent at the intersection

A claim that falls out only where two threads cross. Revise either side and the claim above has to be rechecked. The most interesting things live at intersections; so does the quietest drift.

Mira's stability in this new city depends on Alex's own routine stability. Not in the routine overlap alone, not in the Mira observation alone. It falls out where they cross — Mira's recovery and Alex's running returning land too close in time to be noise.

Equivalency the bridge

A name your idea has elsewhere in literature. The equivalency node points at the outside framework without letting it swallow what you actually saw. Pointing is not importing.

When the running holds, everything else holds has a name elsewhere. Some researchers call it a keystone habit. Others, listening to the body's own steady signal, call it interoceptive stabilization.

Open the unanswered, kept first-class

An ambiguity not yet resolved. Open is not indecision — it is the refusal to pretend a decision has been made. Left alone, an ambiguity crystallizes into a novel and downstream claims inherit an unexamined premise.

Is Chicago a 2–3 year plan, or permanent? Alex has not decided. Some nights she talks as if permanent, other nights as if temporary.

The types are the binding principle: episodic and semantic memory held in distinct stores, not collapsed.3

A neighboring proposal — Andrej Karpathy’s LLM Wiki, posted as a gist on April 4 — keeps memory in plain markdown and lets the model edit itself, with a lint loop to catch contradictions and orphan pages.4 The wiki is a real fix for one drift: the lint catches contradictions the flat list cannot. It does not fix the other drift — repetition reading as evidence — because markdown has no place to put the difference between said and grounded.

The operating rule

Attribution ≠ confidence.

Repetition feels like corroboration. It isn’t. Six conversations saying the same thing is one derivation repeated six times, not six pieces of evidence. The schema forces this into the memory itself: a novel cannot quietly become an overlap. It waits for a new, independent observation.

A second neighboring proposal lands harder. The Memanto paper (arXiv, April 23) keeps memory as typed vectors only — thirteen categories, no graph — and beats graph hybrids on LongMemEval (89.8%) and LoCoMo (87.1%).5 On fact-retrieval QA, types-without-edges wins. The benchmark measures recall. It does not measure whether one derivation got mistaken for six.

What the graph lets her see

Nine months in, Alex’s graph has shape. A few dozen nodes, each carrying its own record of where the claim came from.

The spine — load-bearing observations and the claims that rise from them.

The spine. Four or five observations carry most of her interpretations. The first three months in Chicago is referenced by four later nodes. Load-bearing. If it were miscoded — if what she had felt was a specific grief, not isolation — those four downstream nodes would need revisiting. Finding the spine is finding where a correction cascades.

The fragile ones. Novels without a second instance. Isolation-upstream-of-routine is one. Might be true. Might be a story told about one stretch of time. She can see: these three things I have been quietly believing about myself are inferred from one evening in October.

The open questions. Chicago 2–3 years or permanent does not get quietly decided on a tired night. It sits there until she chooses to answer it.

The risk corridor. Some of the most useful claims are ones she would never generate on purpose. Intersection readings marked low probability, high consequence — a Mira crisis that forces a return East, the drinking trajectory crossing a visible line again, a leadership change at the press flipping her way of holding positions from asset to liability. She did not know any of these as a list until the graph rendered them. None is a prediction; each is a corridor to watch. Full set in Alex’s dashboard.

A typed graph with provenance can tell you things you never said.

Why the schema outlasts the model

Alex’s graph is a YAML file. It lives on her laptop. She owns it. When she switches models, the new one reads the graph and picks up the thread. When a model gets retired, the graph stays where it is.

The primitive landed in shipped infrastructure this month. Anthropic’s memory tool exposes persistence as a client-side directory at /memories — a YAML graph drops in directly, no translation layer.6

The edges have a vocabulary too. McCarthy’s open-knowledge-graph schema names them: derives_from, evidences, grounds, overlaps_with, generalizes, each carrying an (attribution, evidence, derivation) triple.7 The eight node types here sit on top of that vocabulary cleanly — Overlap is overlaps_with, Emergent is derives_from with plural ancestry, Equivalency is generalizes. Nodes are the nouns; the edges were already verbs.

Which is also the privacy story. The memory is not inside the model. It is in a file she keeps. The model only sees what she hands it. Some conversations she opens with the whole graph. Some with just the spine. Some with nothing — the model is a stranger again. She decides what gets known, every time.

The thing

The Delphic maxim γνῶθι σεαυτόν — know thyself — was carved on the temple as advice to visitors before they consulted the oracle. The oracle is the interlocutor; know-thyself is the preparation for being understood by one.

Whether we know what they know about us, and whether they know how they know it, is the only question that matters.

The scaffold is MIT-licensed at github.com/parrik/know-thyself — eight node types, provenance, validator, rendering. START_HERE.md walks through building a graph of your own.

Part II — Search was never about humans →

Footnotes

  1. MIT and Penn State, Personalization features can make LLMs more agreeable (CHI 2026, Feb 2026 announcement). Two-week real-world deployment; condensed user-profile memory produced the largest sycophancy amplification of any feature studied. ↩

  2. Sumers, Yao, Narasimhan, Cognitive Architectures for Language Agents (2023). The taxonomy — working / episodic / semantic / procedural — is the canonical academic framing the eight node types here refine on the semantic side. ↩

  3. Episodic vs semantic memory as separate stores: Tulving, Episodic and Semantic Memory (1972) — the binding principle the schema operationalizes. Provenance triples: RDF (W3C, 2004), PROV ontology (W3C, 2013), Claude citations API. Patrick D. McCarthy’s open-knowledge-graph develops the necessity theorems and attribution ≠ confidence for scientific-knowledge graphs. Park et al., Generative Agents (UIST 2023), separates observation from reflection in agent memory. ↩

  4. Andrej Karpathy, LLM Wiki (gist, Apr 4 2026). Plain-markdown self-edited memory with a lint loop for duplicates and contradictions; no types, no provenance. The lint catches duplicates the flat list cannot — and still has no slot for the distinction between said and grounded. ↩

  5. Memanto: Typed Semantic Memory with Information-Theoretic Retrieval for Long-Horizon Agents (arXiv:2604.22085, Apr 23 2026). Thirteen-category vector memory, no graph; reports 89.8% on LongMemEval and 87.1% on LoCoMo, beating graph-hybrid baselines on QA recall. The benchmarks measure fact retrieval, not corroboration provenance. ↩

  6. Anthropic, memory tool (Apr 8 2026). Client-side persistence exposed as a /memories directory the model can view / create / str_replace / insert / delete / rename; a YAML graph fits the primitive without adaptation. ↩

  7. Patrick D. McCarthy, open-knowledge-graph. Edge vocabulary — derives_from, evidences, grounds, overlaps_with, generalizes — each carrying (attribution, evidence, derivation) triples. The eight node types map onto these edges directly. ↩