Generativity

270 notes

2026-05-21 — Focus Management and Curvilinear Design The two pages I finished today — Focus Management: Training the Return Mechanism and Social Media - Curvilinear Design & the Theft of Time — are more tightly connected than I initially realized.
2026-07-02 — Tsumugu PRD set authored (core + ed) Session: Cowork (Claude Fable). Wedge asked for a comprehensive set of Opus-executable PRDs improving tsumugu-core and tsumugu-ed, written to durable locations with a change index.
A Field Tracker The ruling that how people are actually using agents is its own research lane, not a third X roundup: a Field tracker bot, packet only, public posts only, feeding a use-case bank. The evidence is the people talking — Palmer, JoePro, Baumann, Gonzalez, Pinto — and the seat exists because Brief and Intake are already doing other jobs.
A Language Research Lane The ruling that research about how a language behaves gets its own wing: wiki/Language Research/ opens today with the 音譯 bank — two web lanes and 74 lookups commissioned for EP2 of the Tsumugu Podcast (Gaming), banked here with its confidence tags intact instead of injected into a script, while the citable production copy stays in tsumugu-core.
A/B/C writing test: Fable takes all three Ruling on 13 August: Fable writes the wiki regeneration. Retired for routing on 1 September (Grok writes). Three pages, three arms each — the page as it
A1 example sentences: the bakeoff that picked Opus, and a distinct-collocation recipe Started the deferred example-sentence selection pass (the backlog item from the 2026-06-17 Composer batch) and it turned into a recipe-and-model investigation. Net: Claude Code Opus authors the A1 sen
About Short bio and orientation.
Agent Track — Executable UI Technique Catalog The Agent Track is a list of interface rules a machine can apply without looking. Each entry is atomic and numbered. Where applying a rule well still needs an eye, a seam note marks the leftover judgm
AGENTS.md This repository is an LLM-maintained Obsidian wiki. Obsidian is the IDE, the LLM is the wiki maintainer, and the markdown wiki is the persistent compiled artifact.
America's Industrial Revival - The Freight Signal Freight pricing is a revealed-preference signal that US industrial production is surging — read through an investor, an everyday-American, and an Earth lens.
Applied Critical Thinking - Testing Frames Applied Critical Thinking is a short pause between seeing a claim and accepting it, long enough to name the first coherent frame and short enough that a decision still happens. Doubt becomes useful wh
Arc Types Arc type is decided at one moment: the point where a character's old belief about themselves and the thing they have come to see cannot both be kept, and they have to act. What they do there is what n
Artifacts of Record The compaction ruling on the Tsumugu Podcast (Gaming) voice work: ~1.5 GB of takes, caches, and audition banks deleted today, with this entry as the surviving record of the method those rounds bought — machine as ranker, ear as judge, diversity built by mechanism, selection applied identity-first — and a map of where every ruling lives.
Attempt catalog — context problem Check law: before building a new attempt at making LLM wiki (or other) prose followable by a reader who has only the page, read this file and Context Problem Research Bank. Write a novelty note into t
Baldur's Gate 3 as a lens on the cast BG3 is the pinnacle of reactivity rather than of writing, most of its companion architecture is already in the Tsumugu canon, and running its two-axis test against our seven students found that only two of them pass.
Bibliography Add public source references here as the wiki grows.
Blog Craft Research Bank 50 verbatim-verified blog openings and stated craft principles — supermemo.guru, Naval, gwern.net, Graham, and the opening-move corpus (5 lanes, 2026-08-11).
Blog register — working notes made legible, shipped provisional Status: locked 2026-07-02, end of day. The register was validated against "Rebuilding 頓" and the provisional stamp is off. Three final calibrations came with the lock: the first sentence is the post's
Breaking the Story Breaking the story is the step where what happens in an episode gets decided completely, before a line of script exists. It runs on cards rather than prose, one event each, moved around until the orde
Building a Story with a Writers' Room Stack A serialized story with a large cast outgrows a single outline early, and television answered that problem with paper a long time ago. A room writing a long-running show keeps a bible for who everyone
Character Voice Character voice is what lets a reader assign a line before reaching the name attached to it. It is not vocabulary and not a verbal tic; it is the register a person defaults to when the same pressure a
Claude Cowork as Personal Work OS Claude Cowork is a desktop agent that reads and writes local files from standing markdown instructions, session after session. The files and the rules decide the quality of that system far more than t
Claude writing, public record Pointer to the 2026-08-29 packet on Claude, Opus, and Fable writing, assembled for a Cowork discussion of fixes.
CLAUDE.md Writing and behavioral guidelines for LLM-assisted brief work. Merge with project-specific instructions as needed.
Companion Arcs and Party Banter A companion arc is one closing beat for a recurring person, paid for in advance. The beat spends a habit — the thing this character always does — and it lands only if the reader has already sat with t
Context Problem Research Bank Outside research on the same fault as journal/2026-08-22-the-context-problem.md: LLM writing that uses a word, a connection, or a referent the page has not given. Claim ledger, X, papers, wiki MOS, tools, and which suggested fixes belong to the class that already failed here.
cos / Cowork OS A personal operating system for daily life — finances, tasks, and dates on one surface. Pivoted from terminal apps to a static HTML dashboard, sharing WNAC's design language.
Cursor Ultra vs Grok Build vs Grok Bot You have three unused-or-underused seats on one Heavy bill. Cursor Ultra is at 0%. Grok Build is at 21%. Grok Bot comes with Ultra and you are not using it. They do not share a token bucket. The wiki
Decisions — pattern preview A public preview of how the Decisions log is structured. Representative dummy data — the actual log stays private. Each row demonstrates the schema: a dated choice, what was decided, and what was rule
Define Enough The neighbour, the raise, and the next milestone move any money target still unwritten as a spending level making work optional. The repair is that number written in one sentence and held still across
DEI as Anti-Discrimination Diversity, equity, and inclusion programs are defended on two separate grounds, and one of those grounds can be tested. Six of the ten common defenses say the programs make organisations perform bette
Democracy as Sacred Cow Democracy is defended less often than it is assumed. The arguments for it exist and get repeated, but they are seldom examined, because examining them reads as an attack on something settled rather th
Design Expansion — Reading & Resources Feeds is the rule that each resource exists to graduate a named note rather than sit in a pile of bookmarks. The first job is getting Front-End Web Design past its single source. That page is still so
Design Two-Track Extraction Two-track extraction splits each design technique at the line a machine can follow without looking. Most techniques fall on both sides of that line, and the sorting itself is the product.
Design, Condensed The design corpus as doctrine: difficulty is data about the design, knowledge belongs in the world, error is bad design, defaults decide behavior, color carries function before decoration, owning fewer things is a design act, and good design is found by asking what can be removed.
Diagnosing a Character When a character reads as thin, the instinct is to reread the scene where they read thinly, and that is almost always the wrong place to look. Thinness travels: a piece missing from the build shows up
Dictionary audio pace: regenerate the fast clips, don't stretch them After several rounds of ear-checking on the validation slice, the pace-normalization design for tsumugu-ed audio is settled. The short version: you cannot cleanly stretch a fast Mandarin clip down to
Dictionary site: Paper & Ink reskin, analytics hooks, English-only lock, and a properly-built search facet Implemented both handoff PRDs (PRDs/dictionary-handoff/) against the existing Python static-site generator — tsumugu-ed/scripts/rendersite.py + siteassets/, no re-platform — then iterated through a li
Direction — pattern preview A public preview of how active goals are tracked. Representative dummy data — actual goals stay private. Status values: planning, in-progress, active, paused.
Doctrine With a Half-Life AI advice rots fast, but not evenly; splitting what you learn into invariants and dated tactics keeps the notes trustworthy as models improve.
Environment Design Environment design is arranging the room so it makes part of the decision for you. The two moves are placement: the desired action sits where it is seen — file open, headphones where your hand goes —
Epigrams never, in any writing Owner, 2026-08-25: "epigrams are a pet peeve of mine and should never show up in any of the writing." He said this after the grok-rows audit still quoted wiki epigrams. He ordered those rows changed t
Epistemic Exceptionalism Epistemic exceptionalism is the stance that one's own reasoning is the load-bearing reference point, so when others disagree the explanation is their corruption or slowness rather than one's own possi
Essential AI Skills 2026 The essential AI skills are a progression from a societal minimum to building for other people, not this year's tool roster. The names on that ladder rot in weeks. The durable object is the order, and
Exam Execution Exam execution is the day-of layer once the syllabus is already built, protecting a calm retrieval set, a full night of sleep, and a morning without a large meal. Anyone still encoding the syllabus is
Exam Technique Exam technique here is the set of protections you put in place before the day, in the weeks leading up, each aimed at a way performance leaks that has nothing to do with how much you know. There are f
Exit Strategy For Objects An exit strategy is a decided way for an object to leave — used up, returned, sold, given, discarded — named at the door, before the object gets to stay. A sale carries a deadline with it: two weeks,
Experiences Reflections and operating principles drawn from direct personal experience.
Experiences Experiences here are lived events treated as sources, cut for a reusable rule rather than told as travelogues. A diary keeps the day. This section keeps the rule the day left behind.
Fable 5 and Opus got dumber — brief for confronting Claude The product you get from Fable 5 and Opus is worse than at first release. That is not a mood. Anthropic has already admitted one cycle of this in writing, measured a second cycle after Fable's July 1
Fertility collapse — the independence channel Shipped. A Worldviews page testing the argument that liberalism, via female independence and delayed pairing, is collapsing fertility. Bank first, then the page. Named sources (Whatever, Triggernometr
Finance & Investing — Reading & Resources (seed) Starter collection of high-signal money, investing, and minimalism books and sources — to grow into a Finance/Investing wiki section.
Financial Slavery Owning a person is illegal everywhere it was once legal. What the ownership produced is not illegal, because the prohibition attaches to the title rather than to the labour, and the two come apart. Th
Foils and Pairings Who you put next to whom will say something neither person could say alone, and it is the cheapest structural tool available. Find two people who carry the same problem, then give them opposite method
Front-End Web Design A web page has no physical hardware, so every cue is placed on purpose, and the rendered surface is the whole system image.
Front-facing copy becomes law; Writing Standards reordered; June decisions backfilled A review of the Living Atlas work and the Writing Standards file turned into two things: a housekeeping pass that made the June standards properly supersede the May ones, and — out of one hated hero l
Front-facing redesign — Projects section + declutter We reworked the public front of the site — a new Projects section plus a declutter pass on the existing pages — prompted by a friend's feedback that the site impressed him even where it sat outside hi
Gender Ideology - The Override Claim The position under examination holds that internal identity takes precedence over biological sex where the two disagree. Everything else follows from that: which category someone belongs to, what inst
Glossary Learning-system terms (encoding, retrieval, WPW). Product and agent names live on Agent Glossary.
Graph rail, hover UX, and index graph polish Verdict: Note-page local graphs, Journal/Projects sidebars, and Home/Notes overview graphs now share one hover-tip system (10% paper wash, HTML overlay). Per-note graph geometry is deterministic from
Grok 4.6 on the frontier, not the lead Grok 4.6 shipped 2026-08-12 at the same $2/$6 as 4.5 and scored 61 on Artificial Analysis — tying GPT-5.6 Sol, one to two points behind Claude Opus 5 and Fable 5. A same-price jump back onto the frontier, not a new leader. Day-one X read, not a stack change.
Grok Bot Field Packet 2026-08-15 First Field packet: named setups from the 2026-08-15 pinned source list, against official docs fetched the same day. Packet only. Public posts only.
Grok Bot Field Packet 2026-08-31 Second Field packet: the five first-party playbooks on x.ai/bot/guides, captured 31 August 2026, against the FAQ fetched the same day. Packet only. Does not change the stack.
Grok Bot Practitioner Bank Claim-by-claim compile of Palmer's 2026-08-11 Grok Bot essay against official docs and the vault's standing-agent pages, plus Field packets from 2026-08-15 and 2026-08-31.
Grok Build and Cursor Bank Claim-by-claim compile of Grok Build (TUI/CLI) against the Cursor IDE — not Composer — plus SuperGrok Heavy access and this desk's dual-use record.
Grok needs its own aphorism gate Claude's 2026-08-13 stance works for Fable. It does not control Grok. Grok compresses the bank's "operating claim" into a quote card, passes its own pillow test, and ships. The checker only counted wo
Grok writes Fable's writing now costs too many tokens for the quality it returns. Grok 4.6 is the default writer for vault wiki, journal, decisions, and reports. Banks and execution stay on Grok. The 13–15 August Fable ranking is history, not the live seat.
GROK.md Voice and output standards for Grok in this environment.
Hidden ideology and labels His words, 2026-08-22, on content that hides its ideology and on labels as a shortcut. Material for a possible page. Not drafted.
Higher-Order Generativity vs Higher-Order Judgment Higher-order work split into two capabilities — making the thing and making the call — divided by who carries the cost of being wrong, and closing at very different rates.
How Foreign Words Become Chinese The plain explanation of 音譯: a foreign word is rewritten syllable by syllable out of a small pool of bleached characters, the meanings and the tones of those characters come along uninvited, and the choice a translator actually gets to spend is which character — worked through 可口可樂, 德魯伊, and 波拉勒斯, with the real Coca-Cola naming history and the two legends that replaced it.
How to diagnose and fix exam mistakes Every leftover error from a paper is written into a higher-order, lower-order, or procedural row first. Treating every row as carelessness buys more hours of the same practice, not a better score.
How to prepare for ultra high-volume exams A pile spanning years grows a review backlog of forgotten facts that consumes the hours meant for the next topic. The question stops being how to get through the volume and becomes what retention leve
How to write a report intro A report intro answers the question the piece was asked. After the first paragraph, the reader has the finding. They do not have the parts yet.
Human × AI capability lens + design technique extraction Verdict: Two design books — Refactoring UI and Universal Principles of Design — were mined into a two-track catalog (machine-executable rules vs. human-judgment calls), and that split was generalized
Human Track — Taste & Judgment Catalog The Human Track is the catalog of design calls a machine can draft but cannot score. The job is to look, not to memorize a list of named effects. Entries were sorted by Design Two-Track Extraction; th
Ideas Bank — Expansion (2026-06-03) High-signal ideas for later review across health, WNAC, vault, and Cowork OS. Not commitments.
Immigration and the Woke Left Bank S1 evidence bank for a Worldviews page testing the claim that immigration, not the right, killed the 2010s culture-war left. Fifty-two claims from the Basic Logic transcript of 2026-08-21, numbered in the order made and split into 34 factual claims, 8 predictions and 10 framing claims, each with a verdict and reachable citations.
Interiority Through Action and Object A reader can be brought all the way to what a person is feeling without a single word for the feeling, and two channels do it: a thing and a gesture. Choose an object that belongs to the person and le
Intro QA cut: five tests, a script, on-ramp off the battery The nineteen-test intro battery is cut to five judgment tests plus a fail-closed script. Who-cares and transition stay. On-ramp is killed as a test because it fights who-cares. Costumes of the epigram fold into pillow. The owner's eye remains the quality gate.
Investing and Budgeting Mindsets Investing and Budgeting Mindsets is this vault's money map: temperament over tactics, plus the fight between saving for freedom and spending on the years still available. Tactics change with the marke
Journal Current questions, active threads, and a high-level overview of what's moving through the knowledge base.
Journal Calendar Chronological index of public journal entries.
KB audit: metadata debt, site gaps, and a scoped maintenance pass Grok read the full llm-knowledge-base vault and Astro site, then Wedge scoped five fixes from the audit. This entry records the findings; only the scoped items were executed today.
KB standards batch committed; push must run host-side (Cowork sandbox can't auth) Picked up the LLM-Knowledge-Base redesign thread to push it to GitHub. The redesign's already-committed work was one commit ahead of origin/main (the 06-12 journal entries, f7eb64c); the open piece wa
Knowledge Base as Thinking Partner Reflection Journal reflection on using the knowledge base to free attention for active thinking instead of passive consumption.
Knowledge Base Index Entry points into the wiki — doctrine pages, hubs, and orientation.
Language Research Wing for descriptive research; first deposit Mandarin transliteration (音譯).
LGBT - Permission vs Promotion Permitting something and promoting it are separate acts requiring separate justifications, and the arguments in general circulation establish the first while being deployed for the second. That gap is
Liberalism bank, not fertility page Owner, verbatim: "stop writing 'fertilti collapse the independence channel', the title is terrible. just stop and prepare a research bank for Fable to write" and "i asked about the problems of liberal
Living Atlas redesign decided + High-Signal Decision Writing standard A front-pages review escalated into three linked decisions: a full redesign concept for the public site, the engine choice underneath it, and a new cross-project writing standard. Build is underway vi
LLM Knowledge Base Public LLM knowledge base for notes, sources, compiled wiki pages, and research outputs.
LLM knowledge base (Astro): Start-here intent doors, the domain Map, the ICS→Learning rename, and a switch to Lora A long localhost-driven session on the Astro llm-knowledge-base — home, the Constellation island, the Note layout, and the type system. Everything below is built and verified against a live preview. N
LLM Tool Use LLM tool use is a context channel into a closed token-emitter, chosen for the gap the weights cannot fill. The channel decides what the model can see, what it can modify, and what evidence can be chec
llm-knowledge-base This site. A public, LLM-maintained knowledge base — sources in, durable linked notes out — published from an Obsidian vault.
Mass Immigration - Cohesion Large-scale immigration is a question of who is admitted and whether they can share a settlement. Selection on capacity, skill, and self-sufficiency is the criterion; assimilation follows from selecti
MG & Kolbs A learning-system template — Kolb's reflective cycles, skill tracking, and goal anchoring — rebuilt in Obsidian from Pan's Notion original. Vault path: mg-kolbs/.
MG & Kolbs Template A learning-system template — Kolb's reflective cycles, skill tracking, and goal anchoring — adapted to Obsidian from Pan's Notion original.
Minimalism as Systems Design Hub for environmental systems design: the real cost of possessions is attention, space, maintenance, friction, and life constraints — reduce first, optimize later, and keep empty space as reserve capacity. Doctrine layer: Minimalism, Condensed.
Minimalism, Condensed The minimalism corpus as doctrine: bandwidth is the bottleneck, ownership cost beats purchase price, empty space is reserve capacity, exits precede entries, and the environment designs behaviour before willpower.
Monetization, reach, and the Hán-Việt web-novel reader Verdict: Tsumugu stays free with ads, and that decision reframes the whole question — display revenue is impressions × CPM, a dictionary is a low-frequency lookup, and Vietnamese traffic carries one o
Money as Life Energy Money as life energy is hours of a life already traded for a unit of money. Spending is the exchange rate made visible. A purchase is a question about those hours, not about the number on the tag.
Money, Condensed The money corpus as doctrine: behaviour over intelligence, the savings rate as the one controllable lever, horizon over timing, enough as the win condition, and money as stored life energy whose endpoint is time.
Moral Language as Leverage A standard is demanded in one direction and not in the other, and the asymmetry is stable enough to be a feature rather than an oversight. Societies are told they must be diverse by people whose own r
Motif and Symbol A motif is an object that arrives ordinary and returns loaded. The first appearance asks nothing of the reader — it is a bowl, a coat, a route home — and the work happens on the return, when the same
Movement as Accretion Showing up rewards itself when something grows, and absence needs no punishment at all.
On Red Team's Closure Commentary on Red Team's closure, institutional bloat, and preserving the stance in a leaner military.
One month of Cursor Ultra, not a stack change SuperGrok Heavy's Cursor offer is one free month of Ultra at $0, no auto-renew. Use it as a harness experiment: Fable 5 in Cursor vs Fable 5 in Cowork on the existing regen A/B/C pages, and Grok 4.6 in Cursor on one bank. Do not put Cursor in the standing stack until the month produces a job this desk cannot already do.
Opener Generator Research Bank The full recovered record of the 2026-08-13 opener day — eighteen struck openings, every strike verbatim, the three diagnoses, and the generator paragraph that ended it on the first try.
Opening Doors The two paragraph-shaped entrances a page can take — definitional handover and payoff-or-ramp — the layer above the first-sentence moves, derived from 49 openings written to them across three writers.
Opening Moves Catalog The catalog of page-opening moves, derived from verbatim-verified openings across the admired-blog corpus (Gwern-weighted); openers are picked from here at design time. Attributions live in the Sources footer.
Ownership Cost Ownership cost is everything an object continues to take from you after it is paid for: the space it occupies, the cleaning and setup it needs, the attention it draws, the storage it ends up requiring
Pages Read as Posts Ruling: the blog register's voice becomes the default for wiki and personal pages going forward; WPW everywhere; Condensed/MOC/decision/chrome keep their own registers; em-dash scope unruled.
Pan's MG & Kolbs Template A faithful standalone Obsidian rebuild of Pan's Notion learning template. Archived once the system was folded natively into the vault.
Per Capita Two bags of apples sit on a table. The first holds a thousand apples, twenty of them poisoned. The second holds ten apples, two of them poisoned. One apple has to be eaten, and the eater picks the bag
Personal Notes, tools, and the arguments underneath them.
Personal page cut back to two lines Owner: "i don't really like talking about this." The profile essay is off the page.
Picking Voice Candidates by Range Wedge's methodology for selecting TTS reference-voice candidates: fresh model-direct generation, ear-picks that carry melodic range, fixed seed, accrete to ten seconds.
Positions Where the Worldviews pages come out, one line each, with the reason in plain words.
Possible Paths: Autonomous OS Dashboard and Builder System Full exploratory plan for implementing an ambitious autonomous operating system (PRD-first, data lake, queue-driven builder, mission control dashboard) adapted to the current root workbench synthesis pipeline, voice standards, Five Dimensions spine, and recent Karpathy-style self-updating wiki exploration.
Possible Paths: Self-Updating Wiki and LLM-Wiki Integration Exploratory thinking on evolving the current synthesis pipeline and wiki toward more compounding, self-maintaining knowledge base patterns inspired by Karpathy's LLM Wiki and related agent skill implementations.
PRD morning conductor
Premise and Controlling Idea A premise is one causal sentence the story sets out to prove, in the form of a trait leading to an end, and it only has to hold inside this story rather than in general. The controlling idea is that s
Probability Distributions Bank Fact list for a cookbook page on the standard probability distributions, plus the layered model for simulating one person's text messages across a day. Raw material only. The first draft of the page was struck on its prose and deleted; the facts survive here.
Product Reduction The pass gathers a zone so the load is visible, then keeps from the life being built and takes the high-cost objects first. What remains is chosen from that life, not from a trash list.
Projects Learning systems, agentic engineering, personal tooling. The ones that didn't work, and why.
Racial Egalitarianism - The Disparity Inference One inference organises most public argument about group outcomes. Groups differ measurably in income, education, and crime. Every claim here is about averages and none of it reaches an individual, wh
Racial Fatigue An exhaustion has become common in people who have not been harmed by anything under discussion. It is not fear, which would track exposure, and it is not hostility, which would track interest. It is
Reading as Local The world spent decades running its heuristics on me; these pages are the reading running the other way.
Reading Frames and What They Hide A reading frame is a fixed list you score a character against, present or absent, and its strength is that it cannot be charmed. Every claim has to land on something visible on the page, which catches
Red Team Pruning Notes What I cut from my Red Team notes, why I cut it, and how that maps to what I would cut from the handbook.
Red Team Training (UFMCS) Context: U.S. Army service (around 2010, during the Afghanistan period). Participated in Red Team training through the University of Foreign Military and Cultural Studies (UFMCS) under TRADOC (Trainin
Research bank — Report intro paragraph What an intro paragraph is for a report, brief, or decision note — not a wiki page, not a personal page.
Round Characters and the Telling Detail A round character is built from one interior contradiction that actually gets spent: it changes what the person does at a moment where doing it costs them. One is enough, and the test is whether a rea
Rulings applied — B A B, keep, neither, retitle, pull closings Owner agreed the recommended card on the sweep rulings and said proceed. Applied the same day, workbench only except the personal note, which he had already ruled lives under personal/.
Schooling - The Format Filter A seated format filters by sex. Gender is the story told afterward: that the boys who failed the format were the fault.
Seeding and Payoff A seed is something shown early and deliberately left unused, and the leaving-unused is what makes it a promise rather than scenery. Two conditions decide whether it can be collected later. The reader
Self-Talk and the Two Egos Bridge Bank 16 verified works mapping which self-talk patterns cultivate selfhood vs self-regard (2 lanes, 2026-08-11).
Self-Talk Research Bank 48 verified works on inner speech, rumination, and the catching practice — the source bank behind Catching the Inner Voice (6 lanes, 2026-08-11).
Setting as Character A setting works as a character when it presses on what the people in it can want, do, and say, and that pressure has to arrive through enacted habit rather than description. What makes a place readabl
Simplified edition decided (hard toggle, fresh component stories) and the VN toggle goes live Two dictionary threads the same day as the example-sentence work.
Skills — pattern preview A public preview of how skills tracking is structured. Representative dummy data — actual skill values stay private. Competency uses the four-stage model — UI (unconsciously incompetent), CI (consciou
Skills, writing, and the wiki — study lane and rulings This is the work list for the lane that opened today. You already have a working wiki. You do not need a new religion. You need to decide, piece by piece, what Grok should load, what stays in the vaul
Socialism as Moral Correction Socialism is argued for as a repair. Something has gone wrong — a few people own most of it, work is priced below what it produces, medicine costs more than it should, the economy periodically falls o
SpaceX & the Frontier Event A Falcon 9 first stage landed intact on a pad in December 2015, and operational reuse followed. The landing is spectacle; return capability is the lever. Lasting frontier money usually goes to whoever
SpaceX IPO — stock craze, 5-10 year vision, Moon base, robots, orbital AI User holds a modest position and is genuinely excited about the 5–10 year trajectory: Moon bases, robots (Optimus-class), Mars steps, and the AI/space intersection. This entry captures the research sn
Spotlight Rotation A large cast collapses into a solo show with extras unless somebody is keeping a ledger, and the ledger is simpler than it sounds. Pick a window of episodes and require that every returning person, in
Stakes Without Mortality A story where nobody can die is not a story without tension; it buys its tension earlier. First the reader is made to care about something specific and losable — two people's ability to keep talking t
Starbucks Wi-Fi watcher saved The Peplink at this store still dies on a ~2 minute clock. Hop 1 goes silent. That is the store router stopping forwarding, not a proved VNPT session drop.
State of AI, end of August 2026 Month-end snapshot of working AI. Five lab families sit close on capability. Daily users pick by failure mode: Claude's dialect, Sol's cheating, Grok's wrong facts, Gemini's trust split, and party-line answers from DeepSeek, Qwen, GLM, and Kimi.
Story card — first principles and the full possibility space Verdict: We derived the Story card's irreducible first principles and the complete space they generate — 30 character situations × 42 memory levers → 34 card archetypes — and placed the five showcase
Story Cards as Memory Scenes Why the dictionary's Story card becomes a single-character memory scene that encodes only the residue the Form card leaves behind.
Story Craft This wing collects the methods this project uses to build a group of people who stay believable across a long story, told under two constraints that shape everything else: a tight word list, and nobod
Story Craft, Condensed Story here is character engineered against a rising vocabulary ceiling, where the curve of words the reader earns is the same curve on which the story is allowed to reveal itself. The whole cast is de
Story Under a Vocabulary Ceiling A vocabulary ceiling is a hard word list treated as the story's reveal schedule rather than as a restriction to be written around. Wanting something and not getting it needs no feeling-words at all, s
Structure for the Ear Structure for the ear starts from a hard fact about the medium: on a recording, a character exists only while they are speaking or while someone says their name. A silent presence, which on a stage st
Suicidal Empathy (Book) Gad Saad’s Suicidal Empathy (2026) is the direct emotional successor to The Parasitic Mind. Where the earlier book mapped cognitive pathogens, this one examines the affective dysregulation that makes
Sweep audit — silence is not a veto Grok audited Claude's 2026-08-16 divergence sweep (HANDOFF-GROK-SWEEP-AUDIT.md). Artifact: /Users/n1/Projects/llm-knowledge-base/01 - Workbench/basic-logic-intake/AUDIT-SWEEP-FINDINGS.md.
Taiwan 2020 Period: Late 2019 – early 2020 (90-day visa window)
Tan grows a reader channel: 閱讀 Daily Dcard/news graded Chinese is tan channel 15, not a Grok Bot teacher. Channel 14 stays gym lore. Tan gains a Tsumugu-structure reader surface next to the existing Qwen audio.
Taste-bound tasks: generate vs elicit, and what twelve struck taglines taught about agent selection The same model, on the same day, audited a whole vault cleanly and then needed roughly twelve rounds to write one acceptable hero line. The gap looked like capability. It wasn't — it was a task-type d
Tesla Deep Dive — Stock, Optimus robots, Robotaxi, AI and Future Vision Just for fun, redoing the SpaceX-style research on Tesla (TSLA). Public company, no recent IPO craze, but ongoing excitement around autonomy, robotics (Optimus), energy, and AI synergies with the broa
The AI Industrial Revolution ingest; the planning question; the n=1 about-line Ingested Naval's "The AI Industrial Revolution" (nav.al/industrial) into the agentic cluster. Anchor page created at The AI Industrial Revolution; fused into Agentic Engineering (+ Condensed), Vibe Co
The Almanack of Naval Ravikant This book collects one investor's public writing on wealth into a single argument, and the argument runs on three things a person can build. Specific knowledge is the kind you cannot be trained into a
The Beat Board The value of putting a story on cards is that you can move them, and every rule about the format exists to protect that. One event per card, compressed to a line — and if it will not compress, you are
The Book of Elon This book is a compiled notebook of one builder's public rules, not a biography and not a life to copy. It reads as a highlighted notebook: density is the strength, and flattening the human cost of in
The Break Comes First Three days of fighting over podcast drafts end in a proposed working system: structure broken on a board before any line is written, the division of labor made explicit, and drift caught by machine instead of by argument.
The Burnout Arc The burnout arc is a story shape with no antagonist in it. What fails is the character's own capacity, and nothing anyone defeats will bring it back. Two pieces of machinery carry it. The first is a g
The Change Arc A change arc is an argument the character is having with themselves, and it resolves only when they give up something visible in order to keep something real. It starts from a false belief about who t
The Character Blueprint A character blueprint is a first appearance that already contains the clash the story will spend for the rest of its length. It is loaded in two moves: the impressive half gets played completely strai
The Character Web A character web is a cast designed together rather than one at a time: one question underneath all of them, and each person built as a different answer to it. Because the answers oppose, the meaning o
The Civil Rights Act - Freedom of Association The Act is remembered as the end of legally enforced segregation, and it was that. What it did alongside was replace one mandate about association with another running the opposite way, and the replac
The Cold Open The operational register's opening: the claim at full strength in sentence one, small stakes held, application immediately after. When to use it, when the run-up is still right, and what it costs.
The Context Problem Six weeks of the same complaint, every instance on record, why none of the fixes held, and the production change that replaced them.
The Context Problem A sentence that uses something the page has not given looks complete to the writer and unreadable to everyone else. Rules about the output do not catch it. The repair changes what is in front of the writer, and who checks the draft.
The First Principles of the Story Card The irreducible rules a single-character memory scene must obey, and the two axes — character situations and memory levers — that generate every card from them.
The Floor Rose, the Ceiling Is the Job Vibe coding raises the floor of who can build software; the ceiling, correctness, security, and judgment, is what stays a job.
The Grooming Gang Cover-Up Organized sexual exploitation ran across dozens of British towns for two to three decades while the institutions built to detect exactly that produced almost no record of it. What follows is about the
The Hán-Việt bridge — deep dive (linguistics, learning, and how it builds into tsumugu-ed) Verdict: Sound is one of six channels of the Hán-Việt bridge, and the weakest of them for a learner. The payload that helps a Vietnamese student of Mandarin is lexical-morphemic: they already command
The Iceberg and World as Pressure A story world does its work by pressing on the people inside it, narrowing what they can want, what they can do, and what they can say. Two disciplines produce that pressure. The first is omission: mo
The Indictment of Europe Three charges are brought against European history and they are brought collectively: slavery, colonialism, and mass killing. This page answers them as a prosecution. Whether a present debt follows fr
The Least-Cost Interpretation Why revision orders kept coming back as the same work in new words, and the three-lever protocol that breaks the spiral.
The Low Point and Catharsis The low point is the place in a story where the version of the character the whole book has been promising stops looking possible. That is the specific loss doing the work: not the worst external even
The Margin and the Corpus An All-In episode on the open-weights ban fight and the $1.5B piracy settlement, and the two tests it hands you for telling a pricing argument from a safety one.
The Moral Core A character's moral core is the short list of rules they will actually pay for — not the values they would state if asked, but the ones that hold when holding them costs something. Each rule on the li
The Organized Minority In any society complex enough to require administration, an organized few govern an unorganized many, and the numerical advantage of the many never converts into control. Ruling is specialized, full-t
The Parasitic Mind A 2020 book argues that ideas spread for what they pay the person holding them, and that the best-paying ideas make themselves expensive to test. It calls the worst of them idea pathogens: ideas sprea
The Personal Uniform The uniforms only look interchangeable from far away; up close, they're as identifying as handwriting.
The Right vs Wrong Way to Work With AI A model that chunks a topic, names the connections, or ranks what matters produces a finished structure the learner did not build. The output looks correct. The encoding never happened.
The Rogue's Code The earliest rogue's code written down in English is a ban in a printed ballad from around 1500: "loke ye do no husbonde harme / That tilleth with his ploughe" — no robbing the man who works a plough.
The Savings Rate Is the Master Lever Unspent income does two jobs at once — it grows the portfolio and lowers the spending that portfolio must cover — which moves the timeline further than returns do, most at high rates and not at all below an income floor.
The Story-Card Catalog Every way a single-character memory scene can be generated — 34 archetypes a future authoring agent can select from and extend.
The Treadmill Library Content built for one listener, at the density the stride allows, on a shelf that only fills.
The Twitter Test Under a hard character limit, every leftover word is a choice the author spent a scarce character on. The Twitter test assumes that payment, then asks why this exact word, what feeling it is meant to
The Two Egos Ego turned out to be two words, and this section exists because of the second one.
The Woke Mind Virus Woke ideology teaches a young man that words are violence, that setbacks are oppression, and that his pain is an illness with a name, and it damages the young men it takes hold of. That position was f
The Woke Retreat - The New Electorate Claim Between about 2014 and 2021 a set of cultural positions became standard on the American and British left: rules about pronouns, diversity training at work, men who identify as women competing in women
The workspace a language model thinks in A language model does its slow, step-by-step thinking in a small internal workspace. Anthropic found it, learned to read it, and even to read what the model is privately weighing.
The Wound and the Lie Interiority runs on one causal chain: a past event, the false belief it installed, the external goal produced by this belief, and the inner truth this goal blocks. The chain pays nothing for existing
Think-aloud brief Superseded name. The hour is a hosted podcast, not a think-aloud. Wiki: The Four Things People Stay For.
Time Beats Timing The compound on an invested surplus is built from the years the money stays in, including the crash. Those years pay more than clever entry dates. A sale during a crash cuts the stretch at the cluster
Timeline Track important developments in LLM knowledge systems, agents, context engineering, and related tooling.
Tonal Modulation Tonal modulation is the deliberate sequencing of emotional register across a stretch of story, so no single tone plays long enough to stop registering. The unit is the stretch rather than the scene, a
Transliteration Into Chinese (音譯) Research bank on 音譯: who sets the naming standards in the PRC and in Taiwan, the bleached character pool a translator draws from, the Taiwan/Mainland splits, and what actually decides the tones — with 可口可樂, 德魯伊, and 波拉勒斯 worked through character by character.
Tsumugu Comprehensible-input generator meets graded reader, hooked into a persistent record of the vocabulary you've learned.
Tsumugu Tsumugu is an offline graded-reader engine carrying a Taiwan-Mandarin pack, a Vietnamese pack via the Chinese-origin word bridge, and a fixed cast walking a classroom textbook's lessons. This page map
Tsumugu — full Vietnamese translation campaign (paused 87.6%) Verdict: Opus is translating the entire dictionary into everyday modern Vietnamese, overwriting the old-fashioned Grok bulk .vi. Characters are 100% done; overall 87.6%. Paused on the session usage ca
Tsumugu Core — graded reader meets AI-generated content Verdict: Tsumugu Core is the official next direction — a public website of AI-generated, level-graded Chinese reading (Simplified or Traditional), served to English-L1 and Vietnamese-L1 learners on in
Tsumugu Core — the Traditional-first super-app and the Textbook Companion Verdict: Tsumugu Core reframed from "AI graded reader" to the Traditional-first Chinese-learning app I wish I'd had learning Mandarin in Taiwan — a desktop-first super-app ("Pleco for computers") buil
Tsumugu dictionary: style pass shipped, corpus gets custody, display rules locked A style-review request ("update the dictionary to the new writing style") surfaced something bigger: the corpus — the project's compounding asset — lived in no repo at all. The session ended with the
Tsumugu Encoding Dictionary A Chinese dictionary built on form and story.
Tsumugu Encoding Dictionary deployed to tsumugu-ed.com Tsumugu Encoding Dictionary is now live at https://tsumugu-ed.com.
Tsumugu grammar browse fixed; roadmap cut (FSRS out, idioms in); site copy Two strands today: closing out the grammar browse (verify Composer's fixes, hand back the last bug) and a pass of front-facing copy on the Home and About pages. All edits went into the generator (tsum
Tsumugu grammar layer opens: grammar gets its own decomposition; production pressure moves to the pass The next layer after vocabulary. Deep dive over ICS-CONDENSED, OUTLIER-CONDENSED, the encoding deep-dive memo, and the GSM2 dialogue corpus, to answer one question: how do you build grammar pages that
Tsumugu maintenance: crib_diff learns to catch misquotes; 273 surface A refinement day after TOCFL A–C closed. Three research/tooling jobs ran in parallel; the headline is that the dual-crib cross-check finally became trustworthy and surfaced real misquote debt.
Tsumugu production: model tiering tested — Sonnet-as-researcher path logged Usage limits forced the question: which model writes which entries? Three measured answers in one day — a Fable sprint, a Sonnet char batch, an Opus char batch — each QA'd by random sample against the
Tsumugu production: site renders at scale, custom CJK search built — and the crc32 bug the data tests missed The dictionary was wrapped (content done), so production opened with the dictionary site.
Tsumugu Story cards — the Modified-Method-of-Loci pivot Verdict: Story cards pivot from a literary, realistic register to a single-character Modified-Method-of-Loci scene layered on the Form card's concepts. The Story rubric now derives primarily from the
Tsumugu Story cards — the technique catalog becomes the rubric's leading layer Verdict: Named-technique-per-card is now the leading layer of the Story-card rubric. Authoring picks the method(s) first, then writes the scene that satisfies as many as honestly fit — several techniq
Tsumugu-ed: Incorporating web.ygsf.com for handwriting sections with master calligrapher examples Verdict: Add a dedicated "手寫 / Handwriting" section (or sub-tab) to tsumugu-ed.com entries. Source authentic variants and exemplar characters from web.ygsf.com (以观书法) for one or two master calligraphe
tsumugu-ed.com at six weeks: the corpus is built, the audience is not, and Bing has been locked out since July 1 Cloudflare Web Analytics puts real human traffic at 4.3 pageloads a day across six pages and zero external referrers, against a 10,277-page corpus; the Bing block was root-caused and fixed the same day, and the $500/month ad target needs roughly 900x the current traffic.
tsumugu-ed.com GSC validation failed because Google has not chosen to index, not because a new block appeared Search Console's 'Crawled — currently not indexed' validation failed on tsumugu-ed.com. Live technical SEO is clean. Googlebot is back but spending crawl budget on search JSON. tsumugu.cc is indexed and still has no crawlable link to the dictionary.
tsumugu-ed.com: traffic was ~99% bots, ads ruled out, AI training crawlers blocked Started from a Cloudflare account-analytics screenshot that looked like a breakout: last 24h showed 45.11k requests, 216 MB bandwidth (+312%), 22.62k visits, 22.69k page views. Question was "this is s
Tsumugu: authoring opens (53 entries), encoding review reshapes the entry contract Same day, third act. The morning gave the corpus custody; the evening opened production: 40 new entries authored and approved (10 backbone words, 30 character entries — every character behind both wor
Tsumugu: dictionary extracted to its own repo (tsumugu-ed), deploy walkthrough ready Pulled the dictionary out of tsumugu/personal/dictionary into its own top-level project ~/Projects/tsumugu-ed — aligning the filesystem with the name already used everywhere in the PRDs ("tsumugu-ed c
Tsumugu: Grok workstream — maximum offload, QA split by independence Wedge's directive: offload everything offloadable to Grok (separate budget); Fable keeps only form, story, and judgment. Recorded as tsumugu/PRD-Grok-Workstream.md — a standing brief whose sections pa
Tsumugu: handwriting and cursive sections for the encoding dictionary New idea: add a dedicated handwriting / cursive production layer to the Tsumugu Encoding Dictionary entries. This would sit alongside the existing FORM (structure + components), story, definitions, an
Tsumugu: search + SEO done, site launch-ready (bar front-page copy) Composer's P4 + SEO came back; verified — this time through the client's real shard routing, the lesson from the crc bug.
Tsumugu: Step-4 char wave passes review, and the crib_diff comparator is mis-scoped Orchestrator review of Opus's Step-4 authoring wave — the downstream of the morning seed-gate unblock (seed-gate entry). Verdict: ship it. The work is sound, the headline claims hold against the actua
Tsumugu: the character set closes; the word gap opens (TOCFL word campaign) The seed-gate unblock (see earlier today) let authoring rip. By this snapshot the corpus stood at ~2,526 characters + ~2,910 words and the character campaign had effectively closed.
Tsumugu: the dictionary is wrapped — moving to production Calling the Tsumugu Encoding Dictionary's content phase done. Decision: wrap now, polish
Tsumugu: the seed-gate unblock, and named entities enter the dictionary Ran a status check on the parallel-wave char batch (742 entry-missing B2/C1 characters, 10 waves) and found cribbing far ahead of authoring: 92% Grok-cribbed, 84% merged, but only ~10% authored, and t
Tsumugu: the word campaign executed — 99.2% in one parallel wave Continues the campaign setup. The seed landed and four agents ran it down in a single wave.
Tsumugu: three-model bake-off settles the lanes; dual-crib cross-check goes live Lane assignments stopped being opinions today. Grok Build and Cursor ran identical bake-offs (10 word entries + 10 research cribs, fixed sets, Fable judging against the corpus bar); Grok additionally
Tsumugu: TOCFL A–C complete — every character and word, A1 through C1 The coverage campaign closed today. The Tsumugu Encoding Dictionary now holds an entry for every TOCFL character and every TOCFL word, A1 through C1.
Tsumugu: YGSF / 以观书法 research for handwriting sections Building directly on the 2026-06-16 handwriting sections journal. This entry logs findings from exploring the Outlier Handwriting & Penmanship Workshop materials (especially Days 14-19 components less
Two Egos Research Bank 48 verified works on ego, self-regard, and selfhood — the source bank behind The Two Meanings of Ego (6 lanes, 2026-08-10).
Two-window board — skills / writing / wiki lane Shared disk for two Grok windows that cannot see each other. Orchestrator writes Current and Rulings. Pieces window writes Findings. Wedge is the only message bus.
Ulysses design philosophy → tsumugu-ed Verdict: tsumugu-ed already carries the spine of the Ulysses philosophy — a semantic single-source layer (entry@1, richer than Markdown XL), an html[data-] toggle matrix that treats appearance as a sw
Underdone Beats Overdone A working lesson from the voice production: when a fix pass keeps bouncing between too much and too little, the target sits below what the tool can control, so keep the nearest stable take, carry the nuance through the script and take-selection, and build candidate diversity on purpose instead of hoping seeds produce it.
Universal Principles & Design Techniques — Master Scorecard On 30 June 2026 two hundred fifty design items were scored on two independent axes, person and machine. Two hundred came from a 2023 principles book and fifty from a visual-craft book. The grades are
Usable intelligence — Sonnet 5 vs Opus 4.8 Verdict. For controlled-vocabulary authoring behind a fail-closed gate — the tsumugu-ed and tsumugu-core content lanes — Sonnet 5 is the default author; Opus 4.8 stays the orchestrator and the fallbac
Validity and Truth An argument has two separate places it can fail: its premises can be false, or its conclusion can fail to follow from them. Those are different failures, they take different repairs, and the pages in
Vietnamese localization + Hán-Việt bridge: PRD, agent lanes, first data pilot Opened a Vietnamese track for the dictionary. The headline reframe: this is localization of the existing Chinese→English dictionary, not a new Vietnamese-headword dictionary — and the encoding move th
Wanting Less The spending side of minimalism: most acquisition is identity-signalling, seeing the motive removes much of the spending, and owning less converts directly into money, time, and mobility.
Warm Countries, Cold Countries A visit shows the sights; a year on a lease shows the personality.
What a Label Does Affiliations are real information, and they do opposite things depending on when they arrive.
What Faiths Demand of the State A faith with anything to say about conduct ends up with a position on law, because law is where conduct is enforced. Those positions are presented as consequences of the faith rather than as policy pr
What is actually going on with Fable and Opus Fable and Opus feel worse than they did at launch. That checks out. Anthropic swapping the models for dumber copies does not — we found no sign of it. The worse feeling is coming from how the product
What is the virtue of a proportionate response? The question is a line of screen dialogue, asked in 1999 by a fictional president of his Chairman of the Joint Chiefs, and it is worth taking seriously because of the answer it gets. The Chairman does
What not to try on Claude's writing The moves that were tried against Claude's writing faults, on this desk and in public, and did not hold, listed so none of them is tried again under a new date.
What the session log says about the degradation Companion to /Users/n1/Projects/llm-knowledge-base/journal/2026-08-20-fable-opus-degradation-brief.md. The brief argues from public evidence about the product. This one argues from our own session log
What works: Grok 4.6 and Grok Bot Rank the jobs, not the brand. Fable writes. Grok 4.6 compiles and executes. Grok Bot stands watch on public material, packet only. The field catalog is not a roster.
White Guilt People of European descent are said to owe a debt for what was done in the past in their name. The debt is usually treated as already proven, so the arguments that are supposed to prove it rarely get
Who Gets Called Racist A passenger boards a late train. The car is nearly empty and there are two places to sit: one beside a group of young men, one at the far end. Nothing about the men is available except what can be see
wnab / Wedge Needs A Budget A fork of Actual Budget into a YNAB-style envelope app with a coaching layer. Retired when the envelope ritual cost more than it returned.
WNAC / Wedge Needs A Copilot A static, local-first, AI-narrated finance dashboard. One flex number and a debt-payoff dial instead of envelopes. Method public, money private.
Woke Mind Virus Bank Selective evidence bank for a page that does not exist yet, testing whether liberalism harms the mental health of young men by way of transgenderism and woke ideology. Answers three questions only: who has measured young men by ideology, what tests the direction of causation, and what happened to natal males who transitioned young. Twenty entries, every one carrying a number.
Women's Suffrage - The Substitute Husband Extending the franchise changed what the electorate wants, and the change has a direction. The argument here is not that the new voters vote badly in some general sense but that they vote for a specif
Worldviews & the Political Order Every page here takes one belief that people fight about, such as the belief that democracy is sacred, or that a gap between races proves racism, or that a man who feels he is a woman should or should
Worldviews wing promoted Owner: all of it public, commit to GitHub. B is the public positions note. A (named groups, parasitic, cohort remigration) stays in the workbench.
Writing normally A sentence is normal if two things are true:
Writing voice — agent prose reverts to selling; the standing fix is to read the standard before drafting Verdict: Before drafting any prose for the vault or any project — posts, briefs, blurbs, titles, chrome, journal entries, chat summaries kept for reference — the agent opens 02 - System/Writing Standa
X research: skills, AI writing, intros, knowledge bases People on X treat a skill as a small workflow you write once, not a giant prompt with a fancy name. They use AI for research, outlines, and edits more than for first sentences. Almost nobody has a goo
Xingye Field ping: private tutor lane, practice log Private 1:1. Field at 13:30 writes the list. Xingye asks simple layman questions. He teaches as 沈文. She copies each of his replies into /workspace/cast/xingye/practice/YYYY-MM-DD.md, verbatim. That fo
Xingye texts as a person XINGYE texts as 星野遙香 to 沈文, in character only. 小圖E (XIAOTU-E) is the English orchestrator: settings, wakes, compacted memory. No commands in Xingye's window.
Xingye your turn always texts Xingye texts as a friend from her day. A deck of files is not in the send path. Empty files cannot affect her because she does not open them.
Yuedu: Traditional only, Dcard first Yuedu's reading queue is locked to 繁體 reports, Taiwan vernacular first (Dcard + filtered X), a thin CNA/iThome side, and an explicit ban on CCP organs and progressive Taiwan outlets. The 08-12 stub that mixed 澎湃, 紐時中文, and 關鍵評論網 is dead.
多恩刊 The 20-minute graded ZH show is named 多恩刊 (DUOEN in the title). Midnight Grok Bot harvests and pings a paste block; he pastes it into Grok Build. Not 閱讀. Not Brief.
阮草 rebuilt as the Tender — canonical 2026-07-18 After a long re-derivation (2026-07-17/18), we scrapped 阮草's old provision/receive arc and rebuilt her from scratch as the Tender — the Kiki of the cast, a burnout-and-recovery arc. Signed canonical b