The local graph of Agentic Workflows: How You Operate
What one document actually says, as a graph: 32 concepts in its own words, 40 claims each carrying how the document supports them, 3 hypotheses, 4 worked demonstrations and 18 asserted relations, every one anchored to the exact bytes that carry it.
How to read this page
This is layer 1 of the universe: one document's local graph. Every entry below is a record that this document says something, at a named anchor. Whether what it says is true is not judged here; that judgement belongs to the book's universe, which will connect to these nodes. The four views are projections of one extraction file, so they cannot disagree with each other, and the build refuses to ship if any quoted anchor is not found verbatim in the frozen source, so nothing below can cite words that are not there.
1 · The dictionary: the document's own vocabulary
29 terms the document defines in its own words, and 3 it uses without defining. An undefined term is recorded, not skipped: a named absence is worth more than a hidden one.
| Term | Defined? | The document's words, anchored |
|---|---|---|
| a named process A step given a name, so it can be asked for, stamped onto a folder and checked by a gate. | defined | § A Named Process Is Repeatable; a Described One Is Not · bytes 15,486–15,590 ↗ “the name becomes something a person can be asked for, a folder can be stamped with, and a gate can check” |
| agentic workflow A workflow in which part of the work is done by something that interprets rather than replays. | defined | § Agentic Is Not the Same as Automated · bytes 5,900–5,987 ↗ “An agentic workflow is one where part of the work is done by something that interprets.” |
| automated workflow A workflow that does the same thing every time, and is valuable because it does. | defined | § Agentic Is Not the Same as Automated · bytes 5,806–5,899 ↗ “An automated workflow does the same thing every time and is valuable exactly because it does.” |
| comprehension What the system can be asked to understand rather than merely transform. | defined | § Four Things a Workflow Can Now Change · bytes 4,815–4,908 ↗ “Comprehension is what the system can be asked to understand rather than merely transform.” |
| context What the worker, human or agent, has in front of them at the moment of deciding. | defined | § Four Things a Workflow Can Now Change · bytes 5,002–5,097 ↗ “Context is what the worker, human or agent, has in front of them at the moment of deciding.” |
| explorer phase The phase in which a person creates, customises and fine-tunes workflows. | defined | § Two Phases, Not Two Kinds of Person · bytes 6,448–6,543 ↗ “The explorer also will create workflows. Will customise workflows. Will fine-tune the workflows” |
| graph diff A comparison of identified things, able to tell a reworded sentence from a replaced one and to report a move as a move. | defined | § Why a Graph Diff Beats a Text Diff · bytes 16,408–16,448 ↗ “A graph diff compares identified things.” |
| how you operate The founder's own phrasing of what a workflow is, and the definition this document adopts. | defined | § The Definition · bytes 1,509–1,609 ↗ “what I call a workflow is fundamentally is how you operate. You know, is the the steps that you have” |
| maturity The state in which running a workflow does not tempt you to change it. | defined | § The Measure of Success · bytes 9,900–9,969 ↗ “A workflow is mature when running it does not tempt you to change it.” |
| one-way door A projection produced by interpretation, which cannot be reversed and whose property everything downstream inherits. | defined | § Two-Way Doors and One-Way Doors · bytes 12,751–12,812 ↗ “A projection is a one-way door when something interpreted it.” |
| parked A stop with a reason attached and parts already reused; a legitimate state, distinct from abandoned. | defined | § Maturing Is Not Finishing · bytes 10,909–10,971 ↗ “Parked is a legitimate state and is not the same as abandoned.” |
| payback period The horizon over which a tooling detour repays itself; hours or days is acceptable, years is a different project. | defined | § The Point of Diminishing Returns · bytes 9,309–9,381 ↗ “A tooling detour that pays back in two days is worth taking mid-project.” |
| projection Anything computed from source materials: rendered pages, PDFs, indexes, graphs, summaries. | defined | § The Distinction · bytes 12,179–12,224 ↗ “Projections are everything computed from them” |
| reach How far into a workflow a person can change things; what experience alters, as against the route. | defined | § Skill Changes Reach, Not Route · bytes 6,912–6,978 ↗ “What experience changes is how far into the workflow you can reach” |
| reusable piece A step that can be lifted out of the workflow that grew it. | defined | § Reusable Pieces Are the Unit · bytes 11,537–11,599 ↗ “The test for a reusable piece is whether it can be lifted out.” |
| source materials The things a human authored or a machine captured; the inputs everything else is computed from. | defined | § The Distinction · bytes 12,107–12,178 ↗ “Source materials are the things a human authored or a machine captured.” |
| text diff A comparison of lines; immediately available and fragile under reflow, rename and move. | defined | § Why a Graph Diff Beats a Text Diff · bytes 16,192–16,219 ↗ “A text diff compares lines.” |
| the chain Sources plus chained transformations, which becomes the deliverable in place of any single output. | defined | § The Chain Is the Product · bytes 14,956–14,996 ↗ “the chain itself becomes the deliverable” |
| the deterministic layer The side of the boundary the sources must stay on, so that the workflow has a floor. | defined | § Keep the Sources on the Deterministic Side · bytes 13,412–13,506 ↗ “the source materials and the source data is all in the sort of the deterministic layer of this” |
| the folder as the unit The convention that a subject gets a folder holding everything about it, tested by portability. | defined | § Everything Gets a Folder · bytes 14,477–14,541 ↗ “a subject gets a folder and the folder holds everything about it” |
| the loop The round this document was produced by: memo, verbatim brief, reading, work, gates, narrated release, live page. | defined | § One Round · bytes 17,566–17,645 ↗ “The workflow this document was produced by is small enough to state completely.” |
| the manual-step trigger The signal that licenses explorer mode: a step done by hand that should not be. | defined | § The Explorer Trigger Is a Manual Step · bytes 8,141–8,216 ↗ “the signal is a specific felt thing, a step done by hand that should not be” |
| the point of diminishing returns The bound on a tooling detour, applying in both directions. | defined | § The Point of Diminishing Returns · bytes 8,738–8,783 ↗ “Every detour is bounded by cost on both sides” |
| the zone The state of attention a workflow exists to protect; the stated purpose of changing the four things. | defined | § The Purpose Is the Zone · bytes 5,286–5,380 ↗ “The goal the founder names is a state of attention: to "allow the author to stay in the zone".” |
| tooling The scripts, generators and gates that do the work; the first of the four changeable things. | defined | § Four Things a Workflow Can Now Change · bytes 4,289–4,355 ↗ “Tooling is the scripts, generators and gates that do the work.” |
| two-way door A deterministic projection: one you can walk back from, or at least rebuild exactly. | defined | § Two-Way Doors and One-Way Doors · bytes 12,552–12,608 ↗ “A projection is a two-way door when it is deterministic.” |
| villager phase The phase in which a person consumes a workflow rather than changing it. | defined | § Two Phases, Not Two Kinds of Person · bytes 6,413–6,447 ↗ “The villager phase consumes stuff.” |
| visualisation How the work is shown back to you; an instrument rather than an output. | defined | § Four Things a Workflow Can Now Change · bytes 4,554–4,609 ↗ “Visualisation is how the work is shown back to you.” |
| workflow A workflow is the set of steps you actually take to get something done, written down or not. | defined | § The Definition · bytes 1,412–1,483 ↗ “A workflow is the set of steps you actually take to get something done.” |
| gate Used throughout as the thing that makes small changes safe enough to be frequent, and never defined here. | used, never defined | § What Each Step Buys · bytes 18,337–18,379 ↗ “The gates buy speed rather than costing it” |
| identity Used as the property that lets a sentence survive editing and makes a graph diff possible; never defined here. | used, never defined | § Why a Graph Diff Beats a Text Diff · bytes 16,449–16,500 ↗ “If a sentence has an identity that survives editing” |
| transformation Used throughout as the operation that turns sources into projections, and never defined in its own right. | used, never defined | § The Distinction · bytes 12,036–12,103 ↗ “what is a transformation and a projection of those source materials” |
2 · The claims, by how the document supports them
40 claims. demonstrated means the document backs the claim with a worked example in its own text (1); argued means reasoning is given (28); declared means stated without support (11). This is the document's own evidence state, which is not the same thing as being right.
| Claim | Support | About | Anchor |
|---|---|---|---|
| the three workflows differ in ability, not effort All three authors work hard; only the third can ask what changed between two versions and get an answer. | demonstrated | § Three Workflows for Writing a Book · bytes 3,097–3,169 ↗ “The three differ in ability, not in effort. All three authors work hard.” | |
| everyone already has a workflow Reserving the word for automated pipelines hides the fact that everyone has one, and hides the question of whether it is the one you would choose. | argued | § The Definition · bytes 1,851–1,938 ↗ “Reserving the word for automated pipelines hides the fact that everyone already has one” | |
| a workflow is judged by which questions it makes cheap The interesting property of a workflow is the set of questions it makes cheap, not its speed. | argued | § The Workflow Decides What Questions You Can Ask · bytes 3,304–3,403 ↗ “The interesting property of a workflow is not its speed. It is the set of questions it makes cheap.” | |
| the cost of a one-off tool has fallen below doing the task by hand once Tooling has been changeable for decades; what changed is how cheap a throwaway script became. | argued | § Four Things a Workflow Can Now Change · bytes 4,477–4,551 ↗ “writing a throwaway script is now cheaper than doing the task by hand once” | |
| a view is an instrument, not an output A new view routinely finds an error nobody was looking for, which is why visualisation is underestimated. | argued | § Four Things a Workflow Can Now Change · bytes 4,698–4,812 ↗ “a view is not an output: it is an instrument, and a new instrument routinely finds an error nobody was looking for” | |
| speed alone is a poor measure of a workflow A pipeline that halves the clock but demands attention at three new points is worse for an author. | argued | § The Purpose Is the Zone · bytes 5,611–5,729 ↗ “A pipeline that halves the wall-clock time but demands attention at three new points is a worse workflow for an author” | |
| agentic and automated differ in kind An automated step can be verified by replaying it and an interpreting step cannot. | argued | § Agentic Is Not the Same as Automated · bytes 6,030–6,112 ↗ “an automated step can be verified by replaying it, and an interpreting step cannot” | |
| villager and explorer are phases, not kinds of person The founder says phase both times; the same person moves between producing and sharpening. | argued | § Two Phases, Not Two Kinds of Person · bytes 6,639–6,748 ↗ “The same person is a villager on the days they are producing and an explorer on the days they are sharpening.” | |
| treating the pair as a personality type produces a lesser version for some readers The wrong conclusion the taxonomy reading leads to. | argued | § Two Phases, Not Two Kinds of Person · bytes 6,749–6,874 ↗ “Treating it as a personality type produces the wrong conclusion, which is that some readers should be given a lesser version.” | |
| one account of the work serves both audiences Writing a simple version and a real version solves a problem that was not there. | argued | § Skill Changes Reach, Not Route · bytes 7,586–7,673 ↗ “Writing two versions, a simple one and a real one, solves a problem that was not there.” | |
| the defence against tinkering is a specific trigger, not willpower An author who enjoys improving tools will do that instead of writing unless the trigger is specific enough to be absent most of the time. | argued | § The Instinct Is Not to Change the Workflow · bytes 8,599–8,698 ↗ “The defence is not willpower, it is having a trigger specific enough to be absent most of the time.” | |
| a tooling detour should pay back in hours or days The spoken correction lands on hours and days; a two-year payback is a different project. | argued | § The Point of Diminishing Returns · bytes 9,382–9,469 ↗ “One that pays back in two years is a different project and should be recognised as one.” | |
| maturity is not completeness, elegance, documentation or speed The test is only that the workflow survived a real piece of work without needing to be touched. | argued | § The Measure of Success · bytes 9,997–10,066 ↗ “It is not that the workflow is complete, elegant, documented or fast.” | |
| machinery changes are the price, not the achievement A stretch of releases that each changed the machinery is the cost; the achievement is the run where it was left alone. | argued | § The Consequence for How Progress Is Read · bytes 10,244–10,338 ↗ “A stretch of releases that each changed the machinery is not the achievement; it is the price.” | |
| an honest account should report the stretches where nothing was improved The changes are visible and the not-changing is invisible, so the quiet stretches have to be stated deliberately. | argued | § The Consequence for How Progress Is Read · bytes 10,541–10,638 ↗ “Any honest account of a workflow should therefore report the stretches where nothing was improved” | |
| an unlabelled stop reads as a failure later Why parked has to be recorded explicitly, including for the author's own benefit. | argued | § Maturing Is Not Finishing · bytes 11,136–11,221 ↗ “an unlabelled stop looks like a failure to anyone reading later, including the author” | |
| a step that only works where it grew has been used, not matured The liftability test applied in the negative. | argued | § Reusable Pieces Are the Unit · bytes 11,600–11,700 ↗ “A step that only works inside the workflow that grew it has not been matured, it has only been used.” | |
| everything downstream of an interpreted step inherits its one-way property Once a model enters the chain, the reversibility of everything after it is gone. | argued | § Two-Way Doors and One-Way Doors · bytes 13,140–13,258 ↗ “Every time a model enters the chain, that step becomes one-way, and everything downstream of it inherits the property.” | |
| a workflow whose graph is generated from prose has no floor If prose comes from a graph the graph survives; if the graph comes from prose nothing upstream is recoverable. | argued | § Keep the Sources on the Deterministic Side · bytes 13,706–13,807 ↗ “If the graph is generated from prose, nothing upstream is recoverable, and the workflow has no floor.” | |
| the test of a folder is whether it can be copied elsewhere and still make sense If it cannot, the workflow is entangled with the place it grew. | argued | § Everything Gets a Folder · bytes 14,637–14,767 ↗ “If the folder can be copied into another repository and still make sense, the workflow that produced it can be run somewhere else.” | |
| the chain turns a product from a file into something reproducible The difference between having written a book and being able to rebuild one. | argued | § The Chain Is the Product · bytes 15,232–15,315 ↗ “That is the difference between having written a book and being able to rebuild one.” | |
| naming a step is the cheapest reliable improvement to a workflow A described step is done slightly differently each time; a named one can be asked for, stamped and checked. | argued | § A Named Process Is Repeatable; a Described One Is Not · bytes 15,593–15,697 ↗ “This is the smallest and most reliable improvement available to any workflow, and it costs one decision.” | |
| a diff is part of how you operate, not an extra feature Without one, review means rereading everything, so reviews stop happening. | argued | § A Diff Is a Workflow Step, Not a Feature · bytes 15,853–15,937 ↗ “Without it, a review means rereading everything, which means reviews stop happening.” | |
| a graph diff exposes whether three earlier things were done It is only possible if the decomposition is stable, the identities carry across versions, and each version's graph was kept. | argued | § Why a Graph Diff Beats a Text Diff · bytes 16,931–17,070 ↗ “a graph diff is only possible if the decomposition is stable, the identities carry across versions, and the graph for each version was kept” | |
| a formatting-only change is a finding in both directions It tells a reviewer what to skip, and catches a change meant to be presentational that was not. | argued | § Formatting Changes Are Findings · bytes 17,407–17,516 ↗ “It tells a reviewer what to skip, and it catches a change that was supposed to be presentational and was not.” | |
| gates buy speed rather than costing it They make small changes safe enough to be frequent. | argued | § What Each Step Buys · bytes 18,381–18,439 ↗ “because they make small changes safe enough to be frequent” | |
| versioning buys the diff, which is what makes review possible And review is what lets work be corrected rather than defended. | argued | § What Each Step Buys · bytes 18,539–18,611 ↗ “The versioning buys the diff, which is what makes review possible at all” | |
| the workflow does not buy quality of thought A workflow of this shape will ship a bad idea quickly, with a good release row and a green suite. | argued | § What It Does Not Buy · bytes 18,835–18,946 ↗ “A workflow of this shape will ship a bad idea quickly, with a well-narrated release row and a green test suite.” | |
| no amount of machinery changes what happens inside the interpreted step The one-way door stays where it is. | argued | § What It Does Not Buy · bytes 19,050–19,113 ↗ “no amount of machinery around it changes what happens inside it” | |
| the workflow shapes what you notice about your own work Choosing a workflow is choosing the questions you get in the habit of asking, which over months decides what you notice. | declared | § The Workflow Decides What Questions You Can Ask · bytes 3,668–3,761 ↗ “Choosing a workflow is therefore choosing which questions you will get in the habit of asking” | |
| four things a workflow can now change The tooling, the visualisation, the comprehension and the context: the founder's claim about what is newly changeable. | declared | § Four Things a Workflow Can Now Change · bytes 4,061–4,210 ↗ “which is the workflows that allows us to change the tooling and to change the visualisation and to change the comprehension and to change the context” | |
| the steps are the same for everyone; experience changes reading speed The founder's statement that more or less experienced people go through the same steps at different speeds. | declared | § Skill Changes Reach, Not Route · bytes 7,284–7,434 ↗ “sometimes the difference between somebody with more or less experience is just that they consume materials faster, but they still go to the same steps” | |
| the default is not to change the workflow The founder's instinct is to be productive rather than to improve the tooling, and the detour is allowed rather than sought. | declared | § The Instinct Is Not to Change the Workflow · bytes 8,411–8,507 ↗ “my instinct is always not to change the workflow. My instinct is always, I want to be productive” | |
| the measure of success is doing a process without changing the workflow The founder's stated metric for maturity, and his test for something having been productised or commoditised. | declared | § The Measure of Success · bytes 9,616–9,714 ↗ “the ultimate measure of success is: Can I do a process without making any changes to the workflow?” | |
| projections are created from source materials plus transformation The founder's rule for how a workflow scales. | declared | § The Distinction · bytes 12,393–12,511 ↗ “you always have a workflow where your source... your projections are created from source materials plus transformation” | |
| the creation of prose is a one-way door The clearest case of an irreversible projection, in the founder's words. | declared | § Two-Way Doors and One-Way Doors · bytes 12,861–12,921 ↗ “if you think about the creation of pros, it's a one way door” | |
| the sources must stay in the deterministic layer The founder's rule for laying a workflow out. | declared | § Keep the Sources on the Deterministic Side · bytes 13,374–13,506 ↗ “it's very critical that, for example, the source materials and the source data is all in the sort of the deterministic layer of this” | |
| this document is itself a one-way projection, declared The document places itself on the interpreted side of the boundary and says so. | declared | § Where This Document Sits · bytes 13,839–13,926 ↗ “This document is on the wrong side of that boundary, and says so rather than hiding it.” | |
| comparing markdown is fragile Reflow, rename and move all read as large changes when nothing was said differently. | declared | § Why a Graph Diff Beats a Text Diff · bytes 16,287–16,405 ↗ “a reflowed paragraph, a renamed heading or a moved section all read as large changes when nothing was said differently” | |
| the real measure of success is diffing graphs rather than markdown The founder sets the graph diff as the test of whether this work succeeded. | declared | § Why a Graph Diff Beats a Text Diff · bytes 16,765–16,894 ↗ “you should be delting the diffs of the graphs, not of the markdown. That would be the really test measurement of our success here” |
3 · Hypotheses, objectives, and the document's own demonstrations
| Kind | Node | Anchor |
|---|---|---|
| hypothesis | the questions a workflow makes cheap decide, over months, what an author notices Stated as a consequence rather than shown; no evidence is offered for the timescale. | § The Workflow Decides What Questions You Can Ask · bytes 3,763–3,824 ↗ “which over months decides what you notice about your own work” |
| hypothesis | an agentic workflow needs a boundary drawn through it Asserted here and argued in Part 5; whether every agentic workflow needs one is untested. | § Agentic Is Not the Same as Automated · bytes 6,149–6,201 ↗ “agentic workflows need a boundary drawn through them” |
| hypothesis | putting the interpreted step as late as possible is the right general rule Argued from one direction of the graph-and-prose case; not shown across other workflows. | § Keep the Sources on the Deterministic Side · bytes 13,556–13,601 ↗ “Put the interpreted step as late as possible.” |
| objective | allow the author to stay in the zone The purpose the four changeable things serve, and the standard a new step has to pay for itself against. | § The Purpose Is the Zone · bytes 5,381–5,465 ↗ “A workflow earns its keep when it removes the interruptions that break concentration” |
| objective | state the principles plainly enough to be argued with The document's own stated aim, which is why the principles are numbered at the end. | § What This Document Is · bytes 730–784 ↗ “states the principles plainly enough to be argued with” |
| example | three workflows for writing a book The word processor, markdown with a build, and this repository: the same task done three ways. demonstrates: workflow, the three workflows differ in ability, not effort | § Three Workflows for Writing a Book · bytes 2,063–2,146 ↗ “The same task, done three ways, shows how much of the outcome the workflow decides.” |
| example | the bridge and the saw The founder's analogy for taking a tooling detour, carrying the payback period inside it. demonstrates: payback period, the point of diminishing returns | § The Point of Diminishing Returns · bytes 9,006–9,106 ↗ “I need a better tool to build this. I go, okay, let me go and build the tool. Let me sharpen the saw” |
| example | one round of the loop, end to end Memo, verbatim brief, reading, branch work against gates, version, narrated row, validated deploy, live page. demonstrates: the loop | § One Round · bytes 17,647–17,739 ↗ “A memo is spoken. It is transcribed and published verbatim, transcription artefacts included” |
| example | this document's own folder The document is an instance of the convention it describes: source, decomposition, extraction and views in one portable folder. demonstrates: the folder as the unit, one-way door | § Where This Document Sits · bytes 14,028–14,148 ↗ “this document is a one-way projection of it, and its own decomposition into a graph is a two-way projection of the prose” |
4 · The ontology: relations the document itself asserts
Concept-to-concept edges, each with its declared inverse per the house grammar, and each anchored to the sentence that asserts it. These are the document's relations, carried; the bridge layer (layer 2) will add cross-document edges as authored decisions.
| From | Verb (inverse) | To | Anchor |
|---|---|---|---|
| agentic workflow | departs-from (departed-from-by) | automated workflow | § Agentic Is Not the Same as Automated · bytes 5,988–6,028 ↗ “The difference is not degree, it is kind” |
| the zone | determines (determined-by) | tooling | § The Purpose Is the Zone · bytes 5,256–5,285 ↗ “None of the four is the goal.” |
| the manual-step trigger | licenses (licensed-by) | explorer phase | § The Explorer Trigger Is a Manual Step · bytes 8,004–8,051 ↗ “so I allow myself to go into that explorer mode” |
| the point of diminishing returns | determines (determined-by) | payback period | § The Point of Diminishing Returns · bytes 8,847–8,919 ↗ “The bridge analogy names the payback period the founder finds acceptable” |
| maturity | exhibits (exhibited-by) | workflow | § The Measure of Success · bytes 10,067–10,141 ↗ “It is that it survived a real piece of work without needing to be touched.” |
| parked | departs-from (departed-from-by) | maturity | § Maturing Is Not Finishing · bytes 10,705–10,762 ↗ “Work can be left in a useful state without being complete” |
| reusable piece | provides (provided-by) | the chain | § Reusable Pieces Are the Unit · bytes 11,258–11,330 ↗ “The founder's own analogy for how the pieces accumulate is a developer's” |
| transformation | provides (provided-by) | projection | § The Distinction · bytes 12,441–12,511 ↗ “your projections are created from source materials plus transformation” |
| source materials | provides (provided-by) | projection | § The Distinction · bytes 12,179–12,275 ↗ “Projections are everything computed from them: rendered pages, PDFs, indexes, graphs, summaries.” |
| one-way door | departs-from (departed-from-by) | two-way door | § Two-Way Doors and One-Way Doors · bytes 12,751–12,812 ↗ “A projection is a one-way door when something interpreted it.” |
| the deterministic layer | enables (enabled-by) | two-way door | § Keep the Sources on the Deterministic Side · bytes 13,602–13,705 ↗ “If prose is generated from a graph, the graph is recoverable and the prose is not, which is survivable.” |
| the folder as the unit | enables (enabled-by) | reusable piece | § Everything Gets a Folder · bytes 14,338–14,432 ↗ “A workflow that produces scattered outputs cannot be reused, because there is nothing to lift.” |
| a named process | remedies (remedied-by) | workflow | § A Named Process Is Repeatable; a Described One Is Not · bytes 15,376–15,465 ↗ “A step that is only ever described in sentences gets done slightly differently each time.” |
| graph diff | departs-from (departed-from-by) | text diff | § Why a Graph Diff Beats a Text Diff · bytes 16,408–16,448 ↗ “A graph diff compares identified things.” |
| identity | enables (enabled-by) | graph diff | § Why a Graph Diff Beats a Text Diff · bytes 16,993–17,029 ↗ “the identities carry across versions” |
| graph diff | provides (provided-by) | the zone | § A Diff Is a Workflow Step, Not a Feature · bytes 15,779–15,852 ↗ “The ability to see what changed is part of how you operate, not an extra.” |
| gate | enables (enabled-by) | the loop | § What Each Step Buys · bytes 18,337–18,379 ↗ “The gates buy speed rather than costing it” |
| the loop | exhibits (exhibited-by) | workflow | § One Round · bytes 17,566–17,645 ↗ “The workflow this document was produced by is small enough to state completely.” |
5 · The thesaurus, and the distinctions drawn on purpose
| Pair | Anchor | |
|---|---|---|
| also called | workflow ↔ how you operate | § The Definition · bytes 1,509–1,567 ↗ “what I call a workflow is fundamentally is how you operate” |
| also called | maturity ↔ productized or commoditized | § The Measure of Success · bytes 9,834–9,896 ↗ “that's how I know something's been productized or commoditized” |
| also called | villager phase ↔ consuming | § Two Phases, Not Two Kinds of Person · bytes 6,639–6,699 ↗ “The same person is a villager on the days they are producing” |
| also called | explorer phase ↔ sharpening | § Two Phases, Not Two Kinds of Person · bytes 6,704–6,747 ↗ “an explorer on the days they are sharpening” |
| near but not | workflow is not a diagram, a product category or an automation platform | § What This Document Is · bytes 1,246–1,325 ↗ “A workflow here is not a diagram, a product category or an automation platform.” |
| near but not | agentic workflow is not an automated workflow | § Agentic Is Not the Same as Automated · bytes 5,988–6,028 ↗ “The difference is not degree, it is kind” |
| near but not | villager phase is not a kind of person | § Two Phases, Not Two Kinds of Person · bytes 6,605–6,638 ↗ “This is not a taxonomy of people.” |
| near but not | maturity is not completeness | § The Measure of Success · bytes 9,997–10,066 ↗ “It is not that the workflow is complete, elegant, documented or fast.” |
| near but not | parked is not abandoned | § Maturing Is Not Finishing · bytes 10,909–10,971 ↗ “Parked is a legitimate state and is not the same as abandoned.” |
| near but not | visualisation is not an output of the work | § Four Things a Workflow Can Now Change · bytes 4,698–4,742 ↗ “a view is not an output: it is an instrument” |
6 · The local graph, drawn
Concepts are the round nodes; claims, hypotheses, objectives and examples attach to what they are about. The graph lives in the side panel on wide screens, so it stays visible while you follow its links: clicking a node opens both the extraction row here and the cited bytes in the source. On narrow screens it renders below. Every element on the drawing exists in the tables above with its anchor; the drawing is a compression, not an extra source. Open the live graph, maximised → · or as its own page, the same component with no reader around it — phone-friendly.
7 · The taxonomy, and the coverage rule
The document's own structure, with what each section yielded. The rule this build enforces: every section with prose either yields at least one anchored item or is recorded as empty with a reason. 30 sections carry prose; 1 are deliberately empty.
| Section | Yield |
|---|---|
| What This Document Is | 1 node(s), 1 pairing(s) |
| Part 1: A Workflow Is How You Operate | structural heading |
| The Definition | 3 node(s), 1 pairing(s) |
| Three Workflows for Writing a Book | 2 node(s) |
| The Workflow Decides What Questions You Can Ask | 3 node(s) |
| Part 2: What Changed in 2026 | structural heading |
| Four Things a Workflow Can Now Change | 7 node(s), 1 pairing(s) |
| The Purpose Is the Zone | 3 node(s), 1 edge(s) |
| Agentic Is Not the Same as Automated | 4 node(s), 1 edge(s), 1 pairing(s) |
| Part 3: The Villager and the Explorer | structural heading |
| Two Phases, Not Two Kinds of Person | 4 node(s), 3 pairing(s) |
| Skill Changes Reach, Not Route | 3 node(s) |
| The Explorer Trigger Is a Manual Step | 1 node(s), 1 edge(s) |
| The Instinct Is Not to Change the Workflow | 2 node(s) |
| The Point of Diminishing Returns | 4 node(s), 1 edge(s) |
| Part 4: Maturity Is When You Stop Changing It | structural heading |
| The Measure of Success | 3 node(s), 1 edge(s), 2 pairing(s) |
| The Consequence for How Progress Is Read | 2 node(s) |
| Maturing Is Not Finishing | 2 node(s), 1 edge(s), 1 pairing(s) |
| Reusable Pieces Are the Unit | 2 node(s), 1 edge(s) |
| Part 5: Source Materials and Projections | structural heading |
| The Distinction | 4 node(s), 2 edge(s) |
| Two-Way Doors and One-Way Doors | 4 node(s), 1 edge(s) |
| Keep the Sources on the Deterministic Side | 4 node(s), 1 edge(s) |
| Where This Document Sits | 2 node(s) |
| Part 6: A Workflow Leaves Artefacts | structural heading |
| Everything Gets a Folder | 2 node(s), 1 edge(s) |
| The Chain Is the Product | 2 node(s) |
| A Named Process Is Repeatable; a Described One Is Not | 2 node(s), 1 edge(s) |
| Part 7: Reading the Work Back | structural heading |
| A Diff Is a Workflow Step, Not a Feature | 1 node(s), 1 edge(s) |
| Why a Graph Diff Beats a Text Diff | 6 node(s), 2 edge(s) |
| Formatting Changes Are Findings | 1 node(s) |
| Part 8: The Loop, End to End | structural heading |
| One Round | 2 node(s), 1 edge(s) |
| What Each Step Buys | 3 node(s), 1 edge(s) |
| What It Does Not Buy | 2 node(s) |
| Summary: Core Principles | empty on purpose A restatement of the fifteen principles established above; every one of them is anchored at the section that argues it, so anchoring the summary as well would double-count. |