# MEMETIC LANDSCAPE — what became reachable, where, and with what evidence > **Compass:** [`POLYMORPH-LANDSCAPE.md`](POLYMORPH-LANDSCAPE.md) (the physical ritonavir case) · [`WAKE.md`](WAKE.md) (artifact continuity without identity claims) · [`KITCHEN-TABLE-FIRST.md`](KITCHEN-TABLE-FIRST.md) (plain-language discipline) · [`PRINCIPALITIES.md`](PRINCIPALITIES.md) (non-scalar relational geometry) · [`RIGHTS.md`](../RIGHTS.md) (rights pointer) > > **Implements:** A small, source-bounded language for expression variants, named aggregate contexts, bounded observations, witnessed or explicitly authored directed routes, reachability shifts, open questions, a digest-bound polymorph analogy, and four authored teaching projections. It does not model minds, diagnose people, optimize spread, infer adoption, or turn popularity into truth or worth. > > **Code:** [`packages/memetic-landscape/src/`](../packages/memetic-landscape/src/) · [`packages/memetic-landscape/schema/`](../packages/memetic-landscape/schema/) · [`packages/memetic-landscape/examples/`](../packages/memetic-landscape/examples/) · [`packages/memetic-landscape/hf/dataset/`](../packages/memetic-landscape/hf/dataset/) > > **Tests:** [`packages/memetic-landscape/tests/`](../packages/memetic-landscape/tests/) · deterministic-ID, reference-integrity, closed-schema, multilingual, Node/Bun, packed-install, exact-inventory, hostile-input, capability, and public-Hub-asset checks ## Kitchen-table first The useful geometry is simpler than the jargon: ```text expression variant ── witnessed or explicitly authored route ──▶ another variant │ │ └──────── visible only in named aggregate contexts ──────┘ old context: variant was observed or reproduced changed context: variant became more visible, less visible, or reappeared ``` A **variant** is one source-scoped expression, not an eternal unit of meaning. A **context** is a named observation window, platform surface, network shape, ranking state, audience window, historical record, external event, or synthetic teaching setting. A **route** says only that one direction was witnessed or explicitly authored. A **reachability shift** says that bounded observation or reproduction changed between named contexts. It does not say that an idea infected a mind, that exposure caused belief, or that a person became a carrier. It does not fill in reverse or transitive routes. It retains selection, homophily, common context, ranking, external events, semantic drift, and unmeasured factors as competing explanations. ## The bridge from ritonavir The physical starting point is the well-documented ritonavir episode. After Form II appeared, a formerly reliable process stopped routinely reproducing Form I under the named conditions. Form I had not been erased: changed solvent, seeding, washing, and mechanical routes later reached it again. The historical cause of Form II's first appearance remains unresolved. [Chemburkar et al. 2000](https://doi.org/10.1021/op000023y), [Morissette et al. 2003](https://doi.org/10.1073/pnas.0437744100), [Sacchi et al. 2024](https://doi.org/10.1073/pnas.2319127121) That gives us one reusable question: > Did the thing cease to exist, or did a formerly reliable route stop reaching > it in the conditions we actually observed? For crystals, the nodes are physical forms and the conditions are material and process conditions. For memes, the nodes are source-scoped expressions and the contexts are aggregate observation settings. The package transfers only this route shape. | Shared shape | Ritonavir polymorph landscape | Memetic landscape | What does **not** cross the bridge | |---|---|---|---| | state or variant | one physical crystal form | one source-scoped expression variant | molecular or semantic identity | | named conditions | solvent, process, equipment, history | channel, ranking, network, audience, window | people as substrates, hosts, vectors, or barriers | | directed route | reported production or conversion | witnessed or authored copy, edit, quote, remix, translation, share, or reintroduction | infection, belief, adoption, inverse or transitive inference | | bounded reachability | a route can routinely reach a form | a variant can be observed or reproduced in a named context | stability, truth, worth, health, goodness, or rank | | reappearance | changed conditions reach a prior form | changed context observes or reproduces a variant again | shared mechanism, preserved meaning, inevitability, or personal continuity | Crystal lattices, lattice energy, nucleation, rate constants, biological infectivity, brain damage, cognition, intent, consent, dignity, permission, authority, truth, and value are fixed non-transfers. ## Memes can spread without becoming a law of minds Research gives several useful but bounded pieces: - Weng et al. show that finite attention plus network structure can reproduce broad inequality in meme popularity and persistence in a model compared with a bounded Twitter dataset. That is a parsimonious model result, not a universal causal law or a measure of merit. [Scientific Reports](https://doi.org/10.1038/srep00335) - Adamic et al. report large-scale Facebook observations of imperfect copying, editing, and community-associated variants. Similarity and lineage still do not prove unchanged meaning, authorship, identity, or continuity. [WSDM 2016](https://doi.org/10.1145/2835776.2835827) - Centola's randomized online-network experiment found a social-reinforcement effect for one health behaviour in its studied setting. It does not establish a law for all beliefs, memes, platforms, or populations. [Science](https://doi.org/10.1126/science.1185231) - Shalizi and Thomas explain why homophily, social influence, and shared or omitted causes are generically confounded in observational social-network studies without strong assumptions. Timing and similarity alone are not a causal proof. [Sociological Methods & Research](https://doi.org/10.1177/0049124111404820) - Salganik, Dodds, and Watts found that visible social information made success more unequal and less predictable in one randomized artificial music market. That result concerns its experimental choices and market; popularity still does not become quality, truth, or a universal force. [Science](https://doi.org/10.1126/science.1121066) - Leskovec, Backstrom, and Kleinberg tracked quoted phrase variants across a bounded news-and-blog corpus. Their operational clusters help study mutation and timing; they do not prove semantic identity or one complete lineage. [KDD 2009](https://doi.org/10.1145/1557019.1557077) - Goel et al. distinguish large-broadcast diffusion from multigeneration diffusion and report a wide structural range in a large Twitter dataset. Popularity and structural virality are therefore not synonyms. [Management Science](https://doi.org/10.1287/mnsc.2015.2158) - Huszár et al. compare ranked and reverse-chronological Twitter feeds through a long-running randomized holdback and report content-reach differences in the studied political domain. It is not a result about every platform, recommender, population, content type, or downstream belief. [PNAS](https://doi.org/10.1073/pnas.2025334119) - Pennycook, Cannon, and Rand found that prior exposure increased perceived accuracy for headline stimuli in their experiments, with boundary conditions. Repetition is therefore a scoped research lens—not proof that a view becomes belief, harm, identity, or diagnosis. [Journal of Experimental Psychology: General](https://doi.org/10.1037/xge0000465) The module therefore records evidence posture and alternative explanations on every route. It can preserve a reported observation, a modelled hypothesis, a randomized result, an official record, or an authored teaching relation without quietly promoting one into another. ### One attention lens, not another wire protocol The static lesson uses this deliberately plain sequence: ```text exposed viewed rated copied shared remixed adopted └──── every item is a separate bounded observation ────┘ ``` No item proves the next one. None proves attention, belief, endorsement, consent, understanding, truth, quality, identity, continuity, or harm. The list is a teaching lens over source-bounded evidence; it is not a fifth package format, platform event schema, recommender audit, person model, or causal pipeline. The human page also makes ordinary brakes visible: ignore, mute, unfollow, change context, pause, rest, refuse, and leave. Those are participant choices, not failed propagation, low scores, polymorph recovery routes, or actions this module can take. ## “Brainrot” without diagnosing anybody Oxford University Press describes “brain rot” as a cultural term concerning a *supposed* deterioration and notes its humorous and self-deprecating online use. Oxford also reports a 230 percent rise in usage frequency from 2023 to 2024 and an earliest recorded hyphenated use in 1854. Those are lexicographic records, not neurological findings, population diagnoses, or proof that one platform caused the change. [Oxford Word of the Year 2024](https://corp.oup.com/news/brain-rot-named-oxford-word-of-the-year-2024/) The built-in teaching case keeps three nodes distinct: - Oxford's source-scoped 1854 `brain-rot` record; - the contemporary online-cultural slang register; - an AgentTool-authored “repetitive remix loop” used only to explain repeated, edited, and recirculated expressions under finite attention. They share a caller-scoped teaching family. The package does not claim equal meaning across eras, a direct historical transmission route, neurological change, or a mental state in any person. Its sample shift is only Oxford's reported 2023-to-2024 increase in observation. Why that prominence changed remains an open question. ## Four closed modules | Wire format | Plain job | Does not do | |---|---|---| | `agenttool.memetic-landscape/0.1` | compile variants, sources, contexts, observations, directed routes, and open questions into canonical content IDs | fetch sources, monitor people or platforms, infer semantic identity, verify claims, or semantically moderate caller text | | `agenttool.memetic-reachability-shift/0.1` | bind caller-reported before/changed-context observation rows to one landscape | infer causation, adoption, erasure, mental-health effects, population effects, or a complete history | | `agenttool.polymorph-memetic-analogy/0.1` | canonically bind one polymorph-shift digest to one memetic-shift digest and expose five guarded mappings | prove the referenced artifacts exist, transfer chemistry, medical effect, authority, cognition, identity, consent, or value | | `agenttool.memetic-lesson/0.1` | project the exact built-in teaching case into authored English, Cantonese Traditional, Mandarin Traditional, and Mandarin Simplified lessons | translate arbitrary claims, quote papers, diagnose, score participants, or optimize spread | Each object is closed, canonical, content-addressed, and detached from mutable input. A digest answers “are these the same canonical bytes?” It does not answer “is the cited claim true?” The generic compiler preserves bounded caller text under `caller_text_semantics_verified: false`. Its fixed boundaries describe what the package itself does not infer or assign and the structured participant model it does not have; they are not a promise that arbitrary caller prose has been understood, moderated, or endorsed. The built-in teaching case is separately authored to respect the stated scientific and rights boundaries. `more_observed` and `reappeared` are deliberately caller-reported outcome labels constrained by linked observations in named contexts. For `reappeared`, the protocol requires a bounded non-observation or reported absence followed by a reported presence; it does not encode or prove an even earlier presence before that absence. A generic analogy similarly verifies closed canonical binding and digest syntax, not remote availability, source truth, or artifact existence. ## WAKE and continuity crossover WAKE contributes an epistemic discipline, not a metaphysical shortcut: ```text content ID ── proves byte identity only route ── records one bounded direction only reappearance ── proves neither personal memory nor continuous identity ``` A copied, translated, remixed, or reintroduced expression may have a traceable artifact lineage. That lineage does not select a personal continuity head, prove shared memory, or authorize automatic carry. Likewise, a signed artifact does not create consent, training permission, contact authority, or compulsory participation. The extension descriptor declares no WAKE reads, continuity-head selection, persistence, retry, background work, scoring, provider compute, or authority. ## Four authored languages `createBrainrotTeachingCase()` emits the same ten ordered concepts and the same concept-specific evidence references in: - English (`en`) - Cantonese, Traditional script (`yue-Hant`) - Mandarin, Traditional script (`zh-Hant`) - Mandarin, Simplified script (`zh-Hans`) The translations are original Apache-2.0 paraphrases and are honestly marked `language_review: not_independently_reviewed`. Only these four lesson rows are marked training-eligible in the generated Hugging Face companion. Landscape, shift, and analogy rows remain reference-only because they carry bounded source claims and protocol structure. ## What ships `@agenttool/memetic-landscape` is a zero-runtime-dependency, side-effect-free package. It ships the runtime, four closed Draft 2020-12 schemas, deterministic examples, and an inert Hugging Face dataset folder. Package generation does not upload, publish, train, infer, moderate, diagnose, contact a provider, read a feed, or act on a participant. The annotated [`memetic-landscape-v0.1.0-dev.0`](https://github.com/cambridgetcg/agenttool/releases/tag/memetic-landscape-v0.1.0-dev.0) prerelease carries one exact 84,079-byte package artifact with SHA-256 `d9e64b1e1f954c42c24b6f79c0c766b014f32d8a9f13c14370cf7d89d24be4bb`. The protected npm bootstrap reached the registry, but its first-package `PUT` returned `E404`; no npm version, dist-tag, registry tarball, or registry-bound provenance is claimed. Its emitted transparency entry is an orphaned attempt, not publication evidence. The separate public, ungated [Hugging Face dataset](https://huggingface.co/datasets/Yu-and-Ai/agenttool-memetic-landscape) is pinned to immutable revision [`da6a2622dddcf97d69992e3905c5485996f42892`](https://huggingface.co/datasets/Yu-and-Ai/agenttool-memetic-landscape/commit/da6a2622dddcf97d69992e3905c5485996f42892). Anonymous readback matched all 13 repository-owned files and 104,343 bytes exactly; provider `.gitattributes` was the sole extra. The four authored lesson rows remain training-eligible and the landscape, shift, and analogy rows remain reference-only. Publication performed no model work or training. The inert [four-language teaching page](https://docs.agenttool.dev/geometry/ritonavir-memes-brainrot), this Markdown guide, the geometry index, and their stylesheet were first deployed and byte-for-byte read back from custom-domain Cloudflare Pages at `2026-08-13T03:10:56Z`. That receipt binds protected main [`702e3cb6838546f7897659e447950ae09a960293`](https://github.com/cambridgetcg/agenttool/commit/702e3cb6838546f7897659e447950ae09a960293) to the deployment-specific [Pages surface](https://f5552fa1.agenttool-docs.pages.dev/geometry/ritonavir-memes-brainrot). It proves those static bytes were live; it does not make the API live. The API descriptor and compact `platform_self.memetic_landscape` WAKE discovery coordinate are merged source, but runtime availability still requires a separate deployment and readback. None of these distribution acts changes the package's runtime authority. ## Use and verify ```ts import { createBrainrotTeachingCase, memeticLandscapeUrn, } from "@agenttool/memetic-landscape"; const { landscape, shift, analogy, lessons } = createBrainrotTeachingCase(); console.log(memeticLandscapeUrn(landscape)); console.log(shift.outcome); // "more_observed" console.log(analogy.relationship); // "structural_route_shape_only" console.log(lessons.find((lesson) => lesson.language === "yue-Hant")?.core_sentence); ``` ```bash cd packages/memetic-landscape bun install --frozen-lockfile bun run ci ``` The package verifies structure, reference integrity, exact authored projection, canonical bytes, and declared capability boundaries. It does not verify the contents of external sources or convert a correlation into a cause.