Journal Calendar

A lightweight chronological index for public journal entries and thinking-log updates. This is manually maintained so it works the same way in Obsidian and on the website.

2026-05-17

Expanded the agentic engineering layer and made the local-agent workflow more explicit.

The important shift:

one model as assistant
-> Hermes as local agent
-> Grok as high-level reasoning layer
-> AGENTS as operating contract
-> raw/sessions as reviewable agent memory

Added Current Agentic LLM Stack to document the actual working stack: Grok for higher-level synthesis and system design, Hermes 3 through Ollama/Aider for local file work, and AGENTS.md as the main operating contract. Refined Hybrid Model Workflows, Grok + Hermes so the workflow is less about comparing models and more about assigning them to the right layer of work.

Updated AGENTS with model-specific guidance:

Grok handles conceptual framing, synthesis, and system-level coherence.
Hermes handles local implementation, audit work, link hygiene, and small reviewable edits.
Hermes should use explicit instructions, narrow write targets, and session summaries.
raw/sessions/ becomes the place for raw agent activity, proposals, and post-work memory before promotion into the public wiki.

This matters because the knowledge base is no longer only a collection of pages. It is becoming a human-plus-agent operating environment. The safer version of that environment needs boundaries: private source access stays human-gated, public wiki writes stay reviewable, and raw agent work gets captured before being promoted.

Added 30-Day Challenges as a practical layer over the Five Dimensions.

The section now has one challenge page for each dimension:

The useful frame: each challenge turns a dimension from a diagnostic category into a four-week practice arc. Mindset tracks evidence of improvement, Self-Management audits systems and lowers friction, Self-Regulation builds real-time monitoring, Deep Processing pushes past recognition into relationships and explanation, and Retrieval uses recall as both memory strengthening and diagnosis.

Continued refining How Top Performers Learn.

The page now makes the learning-asset model more explicit:

understand
-> remember
-> apply
-> perform consistently under pressure

This connects the Five Dimensions to the practical question of whether a technique has actually become part of the system. A technique is not locked in because it sounds correct. It is locked in when it can be reconstructed, used in real conditions, and adapted when the situation changes.

Added fresh clippings for future processing:

Hermes Agent Masterclass
How to Build Your Second Brain
r0b0tlab filesystem-first LLM-Wiki + Obsidian + Hermes memory system

These sources point toward a possible next phase: local agent memory, filesystem-first wiki tooling, secret scanning, search/index layers, and safer promotion from raw agent work into durable public notes.

2026-05-16

Refined the knowledge base after the larger architecture migration.

The main pattern:

restructured folders
-> repaired internal links
-> strengthened dimension hubs
-> added new synthesis pages
-> made the model workflow explicit

The five learning dimensions now read more like diagnostic control surfaces. Dimensions of Learning frames top performers as learning engineers rather than technique collectors. Retrieval, Self-Regulation, Self-Management, and Mindset now include clearer common-problem patterns and stronger links to their supporting pages. This matters because the new structure is only useful if each dimension helps diagnose a real bottleneck, not just organize files.

Added How Top Performers Learn.

The useful frame: top performers build learning assets. They do not merely collect techniques. A method becomes an asset only when it is understood, remembered, applied correctly, and performed consistently under pressure. This connects directly to the Five Dimensions because each dimension is part of the system that makes a technique become reliable instead of decorative.

Added Hybrid Model Workflows, Grok + Hermes.

The useful frame: the knowledge base now has an explicit multi-model workflow. Grok is used more for high-level framing, philosophical coherence, and operating tone. Hermes through Aider is used more for local repo execution, link hygiene, taxonomy, and direct file work. This turns the agent workflow into part of the architecture instead of leaving it as an invisible habit.

Also repaired many links created by the folder migration, especially pages that still pointed to the old wiki/Techniques/ or broad wiki/Concepts/ locations. This cleanup makes the graph and backlinks better aligned with the new architecture.

Restructured the knowledge base around the working architecture that has emerged over the last week.

The important move:

flat Techniques / broad Concepts sprawl
-> Dimensions as the learning control layer
-> Systems as durable practice domains
-> Outputs as source-processing workspace
-> raw sources as local/private architecture

The study system now lives more naturally under Dimensions of Learning. Deep Processing holds encoding and schema-building work such as Bear Hunter System, Aim, Shoot, Skin, Prestudy, Schema, Syntopical Reading, and Interleaving. Retrieval holds Spaced Interleaved Retrieval, WPW, and the interleaving table. Mindset holds Marginal Gains, Fixed vs Growth Mindset, and Neuroticism. Self-Management holds Kolbs, Reverse Goal Setting, Skills Audit, and feedback. Self-Regulation holds Metacognition, Recovery, Building the Radar, Opening the Black Box, and sustainable energy.

This is cleaner because the pages are now filed by the role they play in the learning system, not by whether they happened to be called a technique or concept when first created. The old Techniques directory was useful as scaffolding, but it had become too broad. The new structure makes the five learning dimensions act like control surfaces: each page should help diagnose, improve, or practice a specific part of the system.

AI and tool-use pages moved into Agentic Engineering / Systems territory. This separates the learning system from the building-with-agents system without pretending they are the same domain. They can still cross-link, but the architecture no longer forces everything through one overloaded concept directory.

Also created GPT - How to Rest So Well You Never Feel Exhausted Again.

The useful frame: recovery is a matching problem. The activity has to reduce load in the system that is actually depleted. Low-effort stimulation can still keep the attention system active. Real recovery usually needs some combination of psychological detachment, mastery, control, and low-load environments like nature.

Moved the remaining clipping inbox files into local raw sources:

study faster;
enemies of civilization;
Ali Abdaal channel analysis;
rest and recovery;
MG and Kolbs template-tour transcripts.

The raw/source cleanup keeps Clippings/ as an actual inbox instead of a second permanent source layer.

2026-05-15

Created GPT brief for Live Clinic 61: Opening the Black Box of Learning.

The useful frame: the black box is the process layer between inputs and outputs. It contains encoding quality, retrieval quality, spacing, self-management, chunking, prioritization, emotional regulation, and decision-making. It can stay closed during normal operation, then be opened when the system gives a signal: output changes, emotions spike, recall weakens, or performance breaks down.

The practical diagnostic move:

felt symptom
-> process variable
-> decision-relevant question
-> targeted change
-> expected output
-> test result

Created GPT brief for Watch This To Force Your Brain To Study FASTER.

The useful frame: studying faster means forming usable schemas faster. The three cognitive pillars are schema construction, schema assimilation, and schema reorganization. Reorganization is the gatekeeper because it converts a growing, messy map into a structure that can be retained and used. It feels slow in the moment because input pauses, but it reduces future relearning.

The practical sequence:

collect keywords
-> build rough scaffold
-> guess relationships
-> assimilate new information in layers
-> reorganize every 10-15 minutes
-> simplify, group, remove, and clarify
-> retrieve the schema, not only the details

2026-05-14

Created Schema as a dedicated concept hub from Justin Sung’s “Watch This to Force Your Brain to Study Faster.” The three cognitive pillars — schema construction, assimilation, reorganization — map almost exactly onto Prestudy/Aim, Shoot, and Skin. The core reframe: learning speed is not how fast you consume information, it is how fast you form high-quality schemas. Reorganization (the Skin equivalent) is the step most students skip because it feels like slowing down — but it is the step that converts accumulated information into a durable structure.

Schema assimilation vs schema accommodation:

Schema assimilation and accommodation come from Piaget. They are the two mechanisms by which a schema changes.

Assimilation is what happens when new information fits into an existing schema without requiring the schema to restructure. The network absorbs the new piece — extends a branch, fills a gap, adds a sub-concept — but its fundamental shape stays intact. Most learning during a study session is assimilation: the draft schema already has categories, and new information gets placed inside them.

Accommodation is what happens when new information does not fit and the schema must change shape to accept it. A category gets split, a boundary gets redrawn, a previously unified chunk gets separated, or two previously separate schemas get merged. Accommodation is the rarer and cognitively heavier process — it requires recognizing that the existing structure is wrong, not just incomplete.

The practical distinction matters because they have different triggers and different repair strategies:

	Assimilation	Accommodation
Trigger	New information fits existing structure	New information conflicts with existing structure
Schema change	Extension or addition	Restructuring or schema replacement
Feels like	Building the map	Finding the map is wrong
Common in	Most of Shoot and Skin	Correcting misconceptions, major topic revisions, or cross-domain synthesis
Risk	Forcing a fit when the structure should change	Triggering premature restructuring before enough information exists

The Sung video uses “assimilation” in this exact Piagetian sense — adding information to the draft. His “reorganization” step is where accommodation can happen: the structure is reviewed and, if the groupings are wrong, they are changed. In BHS terms: most of Shoot is assimilation; Skin can trigger accommodation when the map reveals that the initial chunk structure was fundamentally wrong.

The learning implication: if retrieval keeps failing in the same place, and the information was covered, the problem may not be retention — it may be that the schema needs accommodation, not more assimilation. The structure is wrong, not incomplete.

Added REDO framework (Reflect, Evaluate, Define, Optimise) to Marginal Gains and expanded the three stacking requirements. Created How to diagnose and fix exam mistakes with the higher-order / lower-order / procedural error taxonomy and the Double Generation Practice Test. Folded the retrieval-before-techniques sequencing note into BHS as a named failure mode (“Technique-first”) and a pre-Aim note.

Compiled Opus briefs for Live Clinics 64, 65, and 66, then converted all three into wiki pages:

Deep Processing for Research: research quality lives in the mental schema that places each source — not in the notes, citations, or writing downstream of it. The jigsaw model: each source needs to be placed, not just filed.
How to maintain sustainable energy under pressure: the standard rest-first strategy fails once the burnout threshold is crossed. The prerequisites (sleep, exercise, connection, sunlight) act faster and more foundationally than any strategy. Alignment generates energy; misalignment drains it regardless of hours spent.
How to prepare for ultra high-volume exams: at 50% retention across multi-year content, repetition eventually consumes all study time. The strategy has to raise retention, manage hit rate (1:1 vs 1:1.2), layer confidence by mark value, and diagnose gaps by type rather than defaulting to re-reading.

Added Best-attempt Encoding as a standalone practical guide: the 1:1.2 encoding decision, what good schema structure looks like, and the five moves that actually get you from narrow to wider encoding (generate an application, find a connection, ask the mechanism question, predict a variation, check the chunk boundary).

Compiled Opus brief for How to Force Your Brain To Be Motivated and converted it into How to shift your brain to be motivated (when you don’t feel like it).

The core idea: motivation dependence makes consistency impossible because motivation is variable by design. DFUZ (Distinguish, Fake, Uptime, Zone) uses thought-action defusion — a clinical psychology technique — to separate feeling from action. The key mechanism is the bidirectional loop: actions produce thoughts that change feelings. You do not need to feel different before starting. Starting is what changes how you feel.

Expanded Writing Standards with several new rules: Section-Level Style Split (prose for framing, bullets for body), Not X But Y (no contrast-first definitions), the hard rule against opening with the page title or subject label (with the clarification that this applies only to the title — other noun-first sentences are fine), and Sources Section Policy.

2026-05-13

Added Knowledge Base as Thinking Partner Reflection.

The current question:

How do I use the knowledge base to free attention from bookkeeping while increasing the amount of thinking I actually do?

Working rule:

offload maintenance
protect struggle
retrieve before browsing
sketch before synthesis
turn links into schemas
turn schemas into action

Later expanded the entry with the current second-brain / first-brain tension: the system can surface high-order connections that would otherwise remain unseen, but those connections may not be hard-won enough to build durable schemas in the user’s own mind. The current working solution is to treat the knowledge base as a set of primers and jumping-off points for active skills and goals, while preserving first-brain struggle before asking the system to synthesize. The unresolved question is whether AI frees the mind for higher-order thinking, like calculators did, or quietly weakens the training that makes higher-order thinking possible.

Added Minimalism as Systems Design to the Current Focus list on Journal and Home.

The working cue:

Reduce ownership cost so the environment becomes easier to act inside. Empty space is reserve capacity. Every object, workflow, subscription, and commitment charges attention, maintenance, and optionality.

Updated graph coloring so Minimalism pages render as black nodes in light mode and white nodes in dark mode.

2026-05-12

Moved the longer “what this is” explanation from Home into About, so the homepage can function more like a clean graph entrance with Current Focus underneath it.

Added Minimally Viable Learning System to answer the question of whether the broader learning system can be compressed without losing the actual working core.

The current minimum:

BHS -> encode usable structure
SIR -> retrieve, space, interleave, and repair it

Added new Journal questions about:

compressing the system into a minimal viable form;
mapping that compression back into the five learning dimensions;
creating more usable frames for Priority 0 skills such as Agentic Engineering, Learning Systems, Vietnamese, 中文, Fitness, and Relationships.

Later corrected the page so the minimum viable system is not an abstract control loop. The actual current minimum is BHS plus SIR, while Agentic Engineering and Knowledge Base as Thinking Partner remain experimental extensions.

Synthesized How to Force Your Brain To Be Motivated and moved the clipping into raw sources.

Split The Age Of Nonlinear Returns into its own page and added the scaling problem, position-vs-transaction framing, and links into decision making, learning, and agentic engineering.

Added Current Focus cues to Journal and Home so the public-facing pages point toward the active practice layer of the knowledge base.

Current focus pages:

Compiled How to Unlearn Old or Bad Habits Efficiently from Live Clinic 69 and folded the older unlearning seed page into it.

Synthesized Sell the Truth from the Naval Podcast clipping and moved the source into raw sources.

2026-05-11

Built the first large outputs layer from the source corpus, including GPT, Grok, and Opus synthesis passes.

The highest-value work centered on the Live Clinics:

These outputs later became the foundation for First Principles of ICS and Are You Learning, or Just Using Techniques.

Added Priority 0 as a public focus-primer page for active identity sets: Agentic Engineering, Learning Systems, Vietnamese, 中文, Fitness, and Relationships.

Created the first version of the private Goals and Skills structure, while keeping detailed skill/goal work local unless intentionally made public.

Expanded the wiki around:

Also added output syntheses for the Triggernometry clippings, including:

critical thinking and institutional trust;
biology, politics, and ideology;
Mark Moyar’s Vietnam history discussion.

2026-05-10

Made the raw source layer local-only so the public GitHub repository keeps the architecture visible without publishing bulky source material.

This was the major privacy and architecture shift:

raw/ exists as architecture
-> source contents stay local
-> wiki and outputs carry public synthesis

Ran a system audit and cleanup pass:

removed empty placeholder folders;
normalized page type/status metadata;
renamed Deep Processing concept material into Deep Processing Practice;
cleaned broken Red Team links from the log;
codified a smaller type/status vocabulary in AGENTS;
documented the loose outputs workflow;
filed the 2026-05-09 system audit.

Regenerated and refocused Applied Critical Thinking: Testing Frames.

The emphasis shifted toward fast practical use:

media frames;
activist or narrative-driven institutions;
LLM and AI outputs;
speed/accuracy balance;
testing frames without turning every decision into a slow workshop.

Started the outputs/briefs workflow at scale:

This was the transition from only compiling wiki pages into maintaining a separate output layer for source syntheses before deciding what belongs in the durable wiki.

2026-05-09

Added Red Team Pruning Notes and drafted On Red Team’s Closure.

The working thesis:

Red Teaming still matters as a stance and identity.
The standalone tool layer had become bloated.
The things pruned from the personal wiki are also the things I would cut, compress, or demote from the Red Team Handbook.
Many tools assume a group or think-tank process that is too slow for daily decisions and too hard to defend as federally funded training.
AI has made many tool outputs cheap, shifting the bottleneck from generating alternatives to judging relevance.
The modern failure is often The Shortcut Problem, not simply “missing the black swan.”
Cultural empathy should be bounded by Suicidal Empathy and decision quality.
The reform version would preserve the lean two-week course and cut the slower, more specialized, tool-heavy versions unless they can prove clear value for specific roles.

Added Journal as the public thinking dashboard for current questions, active threads, possible essays, and a light dated log.

The first version tracks three things:

running questions about focus, active interests, and what belongs in public;
active knowledge-base threads such as ICS System, Bear Hunter System, Spaced Interleaved Retrieval, Attention Management: Preserving Flow, Agentic Engineering, and Red Teaming;
possible blog posts, especially Emacs to Obsidian and How I Built This.

The Journal is not a task manager. It is the public surface for questions, active threads, and writing candidates.

2026-05-08

Tightened the public site around the knowledge base.

The main work:

made the knowledge base the homepage and renamed the public home surface to LLM Knowledge Base;
refined the graph behavior, including wiki-only filtering, global colors, canvas sizing, and a flatter wide-screen layout;
moved the colophon and implementation details into About;
simplified the Index into the stable catalog;
added or expanded active learning and self-management pages, including Decision Making, Flow State, Focus Management: How to Enter & Recover Inside a Work Block, Procrastination: a System Problem, and Attention Management: Preserving Flow.

2026-05-07

Converted the knowledge base into a personal Quartz website and shaped the public interface.

The main work:

added the Red Team and ICS hub material, including Red Teaming, ICS System, Bear Hunter System, Fixed vs Growth Mindset, and Neuroticism;
created the personal site shell around Blog, Index, About, and the public knowledge base;
iterated on graph presentation, duplicate headings, sidebars, typography, and the pastel reading theme with neon graph colors;
restored About and the public framing of the project.

2026-05-06

Created the initial Quartz-backed repository for the public LLM-maintained knowledge base.

The first version established the basic pattern:

markdown notes as the source layer;
Quartz as the public static website;
GitHub Pages as the deployment target;
an initial landing-page template, system tags, graph filtering, and Node 24 build configuration.

2026-05-18

Ran full evolution cycle on the l3-to-l2-voice-converter skill. Generated and evaluated 4 prompt variants, selected Variant D (stronger feedback integration + Internal Check step), and applied it as v0.5.

Cleaned up the skill by removing the Python diff-capture helper to keep everything pure Hermes/Grok. Updated converter.md, SKILL.md, and created journal entry + calendar update.

Key outcome: The converter now has a mandatory internal check against style-feedback.md before rewriting, making manual L2 edits more effective at training the automation.