Framework diagrams

23 diagrams of the mechanisms named in the essays.

By question

The whole position. Three Questions, Each Starting Where the Last Stops, The Whole Framework in One Map, Scale, Burden and Control, People Keep the Institution Running.

Distribution: who bears the burden. Burden Transfer Flow, When Covering Becomes the Plan, What the System Sees, Where the Complexity a Rule Drops Ends Up, Permission → Cost → Non-Use Loop, A Person Lives With the Consequences; the System Keeps Running. On the program page this is the distribution lens.

Correction: who can change the arrangement. Does the Harm Get Back to the Institution?, What It Takes for an Error to Change Anything, Correction Authority Is Divisible, Fixing the Case, the Rule, or Who Decides, More Evidence Puts More Under Review, A Settled Decision Should Say How It Can Be Reopened, What a Rule Does With the People It Fails, Internal Learning vs External Selection, Anticipated Sanction → Observed Consensus Loop, Good vs Bad Insulation, The Capacity to Act and the Capacity to Be Corrected. On the program page this is the transformation lens.

Jurisdiction: what the institution was entitled to decide. Being Right Doesn’t Make It Yours to Decide, Giving a System the Right Goal Doesn’t Give It Authority. On the program page this is the authority lens.

Reproduction: how an arrangement keeps itself in place. When Covering Becomes the Plan, Permission → Cost → Non-Use Loop. These sit in the groups above, because reproduction cuts across them. On the program page this is the reproduction lens.

New to the framework? Read the short guide first.

Program lenses against diagram groups, showing populated intersections and open gaps.
Program lensThe whole positionDistributionCorrectionJurisdiction
DistributionWhere does the cost go?No diagram in this intersectionBurden Transfer Flow
What the System Sees
No diagram in this intersectionNo diagram in this intersection
AuthorityWhere does authority stop being legitimate?No diagram in this intersectionNo diagram in this intersectionNo diagram in this intersectionBeing Right Doesn’t Make It Yours to Decide
Giving a System the Right Goal Doesn’t Give It Authority
ReproductionHow does the arrangement preserve itself?No diagram in this intersectionWhen Covering Becomes the Plan
Permission → Cost → Non-Use Loop
No diagram in this intersectionNo diagram in this intersection
TransformationWho can change the arrangement?No diagram in this intersectionNo diagram in this intersectionFixing the Case, the Rule, or Who Decides
Correction Authority Is Divisible
No diagram in this intersection

How to read a diagram

A part of a diagram is drawn by the job it does, and keeps that look in every diagram on the site. Yellow is always the person, red is what costs them or what forces a change, and solid ink is the way out.

A person meets a gate. One way runs through what forces a change to repair; the other waits, costs the person, and ends in failure.
A person meets a gate. One way runs through what forces a change to repair; the other waits, costs the person, and ends in failure.
View source
%% title: How to read a diagram
%% caption: A person meets a gate. One way runs through what forces a change to repair; the other waits, costs the person, and ends in failure.
flowchart TD
  P([The person]):::actor --> S[A step]:::system
  S --> G{A gate}:::gate
  G == passes ==> B[[What forces<br/>a change]]:::binding
  G -- held --> T([Pending]):::status
  B ==> R([Repair]):::repair
  T -. on paper .-> R
  T -- delay --> L[Burden]:::burden
  L --> F([Failure]):::failure
  R ~~~ C[A concept]:::concept
  • The person, in yellow: whoever lives with the outcome, often someone in your place.
  • A step, a plain box with a blue edge: routine work the system does, such as intake, review or keeping a record.
  • A gate, an ink edge and usually a diamond: a test or threshold where the path splits.
  • What forces a change, solid red: the mechanism that makes an institution give something up, the handle you pull on a safety card.
  • Repair, solid ink: real protection, or a way out the person survives.
  • Pending, dashed ink: a state left open or suspended, such as on hold or under review.
  • Burden, a red edge on a red tint: work or cost pushed onto the person. Some diagrams call it load.
  • Failure, dashed red: exclusion, attrition, or a case closed without an answer.
  • A concept, dotted grey: an idea laid over the picture, such as time or a metric.
  • Arrows: a thick arrow is an obligation that binds; a dotted one is a path that exists only on paper.

The whole position

Three Questions, Each Starting Where the Last Stops

Each question starts where the previous one stops: a fair distribution says nothing about whether a failure can be remedied, and a working remedy says nothing about whether the authority was the institution's to hold. The third question is partly open. Use this to place an argument about an institution before choosing a diagram: find the question it's really asking, then open that question's diagrams below.

Distribution, correction and jurisdiction each fail in their own way. An institution that passes the first can still fail the second, and one that passes both can still fail the third.
Distribution, correction and jurisdiction each fail in their own way. An institution that passes the first can still fail the second, and one that passes both can still fail the third. How to read this
View source
%% title: Three Questions, Each Starting Where the Last Stops
%% caption: Distribution, correction and jurisdiction each fail in their own way; passing one does not answer the next.
flowchart TD
  classDef concept stroke-width:2.4px,font-weight:bold;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
  classDef status stroke-dasharray:4 3;
  Q1[Distribution:<br/>who bears the burden?]:::concept --> F1[Fails when outcomes<br/>are funded by moving costs<br/>onto those served]:::failure
  Q1 -- even if burdens<br/>land fairly --> Q2[Correction:<br/>who can change<br/>the arrangement?]:::concept
  Q2 --> F2[Fails when people who<br/>can show the failure<br/>cannot get a remedy]:::failure
  Q2 -- even if mistakes<br/>get corrected --> Q3[Jurisdiction:<br/>what was it entitled<br/>to decide?]:::concept
  Q3 --> F3[Fails when it exercises<br/>authority it should<br/>never have held]:::failure
  Q3 -.-> O[Open: who sets the limits<br/>when the people it<br/>affects disagree]:::status

The whole position

The Whole Framework in One Map

One map of the whole position: error governance is the mechanism, consequential review is the test, and the absence of a permanently exempt layer is the design. Use this as orientation before any individual mechanism: locate the arrangement on the map, then descend into the stage of the path that decides its case.

Institutions will always make mistakes, and where power is unequal, so is the power to define errors and push them onto others — the risk of domination. The answer runs through detectability, contestability, and burden tracing into consequential review, and the correction mechanism itself can also be wrong and can also be challenged.
Institutions will always make mistakes, and where power is unequal, so is the power to define errors and push them onto others — the risk of domination. The answer runs through detectability, contestability, and burden tracing into consequential review, and the correction mechanism itself can also be wrong and can also be challenged. How to read this
View source
%% title: The Whole Framework in One Map
%% caption: Mistakes plus unequal power make domination possible, unless an error can have consequences without anyone's permission and the correction mechanism itself can be challenged.
flowchart TB
  A[Permanent fallibility,<br/>asymmetric institutions] --> B[Unequal power to define,<br/>absorb, externalize,<br/>and respond to error]
  B --> C[Risk of domination]
  C --> D[Detectability:<br/>who can see the error]
  C --> E[Contestability:<br/>who can define it]
  C --> F[Burden tracing:<br/>who pays for it]
  D --> G[Consequential review:<br/>decision, rule, structure,<br/>jurisdiction]
  E --> G
  F --> G
  G --> H[The correction mechanism<br/>is also fallible]
  H --> D

The whole position

Scale, Burden and Control

When uniform rules misclassify a case and officials can refuse correction, people downstream carry the error. Smaller reach can make cases easier to inspect. Binding challenges and funded repair can make a large institution answer for errors, while some losses remain. Use this to examine who can challenge a decision, what a successful challenge requires, and who ultimately funds repair. Compare evidence about actual arrangements instead of treating the triangle as a proof.

Uniform scale, protection from errors and insulated authority can pull against one another. This diagram compares proposed arrangements; it doesn't establish a mathematical impossibility.
Uniform scale, protection from errors and insulated authority can pull against one another. This diagram compares proposed arrangements; it doesn't establish a mathematical impossibility. How to read this
View source
%% title: Scale, Burden and Control
%% caption: A proposed trade-off to investigate: compare authority, error burdens and remedies at different scales. Liability and smaller reach do not guarantee the elimination of harm.
flowchart TB
  classDef binding stroke-width:2px;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
  classDef gate stroke-width:1.6px;
  classDef status stroke-dasharray:4 4;
  classDef repair stroke-width:2.4px,font-weight:bold;
  T[Scale, Burden and Control] --> S[1. Massive Scale & Throughput<br/>Legibility at Scale]:::binding
  T --> H[2. Protection from Errors<br/>Who Carries Residual Costs?]:::failure
  T --> A[3. Absolute Executive Authority<br/>Uncontested Execution]:::gate
  S --> D{Compare actual<br/>arrangements}
  H --> D
  A --> D
  D -. Insulated authority .-> CL[Errors can remain<br/>private costs]:::status
  D -. Smaller reach .-> LO[Cases easier to inspect<br/>No guarantee of safety]:::failure
  D == Binding challenges ==> OP[Institution must fund repair<br/>Trace the losses that remain]:::repair

The whole position

People Keep the Institution Running

Resilience cybernetics identifies the gap between work-as-imagined and work-as-done; social reproduction reveals the unpriced human energy maintaining life; institutional economics tracks how costs are dumped into bodies; administrative burden explains how friction disarms redress. At the center sits the Crumple Zone: institutions surviving on unrecorded human compensation. Use this when explaining what kind of theory is built here, and how its critique connects systems engineering, economics, care labor, and political power.

How four distinct traditions converge into one diagnosis: the gap between formal system design and human survival.
How four distinct traditions converge into one diagnosis: the gap between formal system design and human survival. How to read this
View source
%% title: People Keep the Institution Running
%% caption: How four distinct traditions converge into one diagnosis: the gap between formal system design and human survival.
flowchart TD
  classDef concept stroke-width:1.5px;
  classDef binding stroke-width:2.4px,font-weight:bold;
  A[Safety-Critical Cybernetics<br/>Work-as-Imagined vs Work-as-Done]:::concept --> C
  B[Feminist Social Reproduction<br/>Unpriced Affective & Relational Labor]:::concept --> C
  C[THE HUMAN SUBSIDY<br/>Compensated performance & epistemic erasure]:::binding
  C --> D[Institutional Economics<br/>Adaptation Externalities & Social Costs]:::concept
  C --> E[Administrative Burden & Non-Domination<br/>Attenuated Signals & Corrective Standing]:::concept

Distribution

Burden Transfer Flow

What looks efficient at the center may be strain that has been exported to someone with less leverage. Use this when an institution looks orderly only because somebody downstream is cleaning up after it.

A system often runs smoothly only because work, risk, delay, or emotional fallout has been pushed downstream, and the people absorbing it get called resilient.
A system often runs smoothly only because work, risk, delay, or emotional fallout has been pushed downstream, and the people absorbing it get called resilient. How to read this
View source
%% title: Burden Transfer Flow
%% caption: A system often runs smoothly only because work, risk, delay, or emotional fallout has been pushed downstream, and the people absorbing it get called resilient.
flowchart TD
  A[System wants smoothness] --> B[Cost gets exported]
  B --> C[Downstream person,<br/>team, or patient absorbs it]
  C --> D[Adaptation looks like<br/>professionalism or resilience]
  D --> E[Institution counts the surface as success]
  E --> A

Distribution

When Covering Becomes the Plan

Two loops keep the arrangement in place. The plan comes to count on the cover, so being good at the workaround makes it the job. The complaint channel logs the objection, so the institution can answer every case and change nothing. The dashed path is heard; the heavy one binds. Use this when an arrangement survives its known costs: staff keep covering and complaints keep getting answered, while the rule that produces both never comes up for review. The open question is who could be made to review it, and on what finding.

A worker covers for a known failure, the metric reads green, and the next plan counts on the cover. A complaint is heard and logged without power over the rule. Only a finding that binds the rule's owner leaves the loop.
A worker covers for a known failure, the metric reads green, and the next plan counts on the cover. A complaint is heard and logged without power over the rule. Only a finding that binds the rule's owner leaves the loop. How to read this
View source
%% title: When Covering Becomes the Plan
%% caption: A worker covers for a known failure and the next plan counts on the cover; a complaint is heard without power over the rule. Only a finding that binds the rule's owner leaves the loop.
flowchart TD
  classDef actor stroke-width:1.6px;
  classDef gate stroke-width:1.6px;
  classDef binding stroke-width:3px,font-weight:bold;
  classDef burden stroke-width:2px;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
  classDef repair stroke-width:2.4px,font-weight:bold;
  O([Rule owner: sets the<br/>process and the staffing]):::actor --> F[The process fails<br/>in a known way]
  F --> W([Worker covers it:<br/>rebooks, re-enters, stays late]):::burden
  W --> M[The metric reads green]
  M --> P[The next plan counts on the cover:<br/>staff cut, cover expected]:::failure
  P -- "the cover becomes the job" --> W
  P --> C[Worker or patient complains]
  C --> G{Can the complaint bind<br/>the rule owner?}:::gate
  G -.-> I[No: heard and logged,<br/>with no power over the rule]:::failure
  I -. "the loop holds" .-> P
  G == yes ==> R[[Owner must review:<br/>restaff, fix or stop the process]]:::binding
  R ==> X([The cover goes on the books<br/>or stops being needed]):::repair

Distribution

What the System Sees

The dashboard and the ledger record completed outcomes. The phone calls and workarounds that produced them are invisible to the model — which is exactly why the model's assessment of itself stays green while that work keeps growing. Use this when an institution reports lower cost or higher throughput and the open question is how much uncounted labor downstream made up the difference.

A system can look efficient because it counts outcomes and ignores the labor required to produce them.
A system can look efficient because it counts outcomes and ignores the labor required to produce them. How to read this
View source
%% title: What the System Sees
%% caption: A system can look efficient because it counts outcomes and ignores the labor required to produce them.
flowchart LR
  classDef system stroke-width:1.2px;
  classDef burden stroke-width:2.2px,stroke-dasharray:6 3;
  subgraph SEEN[What the system counts]
    direction TB
    A1[Application submitted]:::system
    A2[Appointment completed]:::system
    A3[Shift covered]:::system
    A4[Claim resolved]:::system
  end
  subgraph HIDDEN[What it may not see]
    direction TB
    B1[3 phone calls / 2 hours waiting]:::burden
    B2[Caregiver help / masking]:::burden
    B3[Unpaid overtime / workarounds]:::burden
    B4[Stress]:::burden
  end
  A1 -. uncounted .- B1
  A2 -. uncounted .- B2
  A3 -. uncounted .- B3
  A4 -. uncounted .- B4

Distribution

Where the Complexity a Rule Drops Ends Up

When an automated interface is simplified, complexity is conserved and dumped downstream. Power is the privilege of simplifying while forcing others to adapt. Use this when an institution claims an automated process cut handling time or unit cost, while claimants or staff downstream do the uncounted work of reconciling it with real cases.

Formal compression removes complexity from the model while the world keeps all of it. The unmodeled residual turns directly into adaptation work for the people downstream and power for whoever did the simplifying.
Formal compression removes complexity from the model while the world keeps all of it. The unmodeled residual turns directly into adaptation work for the people downstream and power for whoever did the simplifying. How to read this
View source
%% title: Where the Complexity a Rule Drops Ends Up
%% caption: Complexity is conserved. Formal compression leaves residual complexity, which lands as burden on some people and power for others.
flowchart TD
  classDef actor stroke-width:1.6px;
  classDef system stroke-width:1.2px;
  classDef gate stroke-width:1.6px;
  classDef repair stroke-width:2.4px,font-weight:bold;
  A[Formal Compression<br/>Reduction to schema]:::system --> B[Residual Complexity<br/>Complexity conserved]:::gate
  B --> C[Burden Allocation<br/>Downstream absorption]:::actor
  C --> D[Power<br/>Adaptation privilege]:::repair

Distribution

Permission → Cost → Non-Use Loop

A system can preserve the language of choice while making the lived use of that choice too costly to sustain. Use this when an institution points to low uptake, silence, or “continued participation” as evidence that the option was fair all along.

Formal permission fails publicly when the cost of use makes withdrawal predictable and institutions misread that withdrawal as proof nothing needed to change.
Formal permission fails publicly when the cost of use makes withdrawal predictable and institutions misread that withdrawal as proof nothing needed to change. How to read this
View source
%% title: Permission → Cost → Non-Use Loop
%% caption: Formal permission fails publicly when the cost of use makes withdrawal predictable and institutions misread that withdrawal as proof nothing needed to change.
flowchart TD
  A[Permission exists] --> B[Using it carries cost]
  B --> C[Withdrawal, delay,<br/>silence, masking]
  C --> D[Institution records non-use]
  D --> E[Non-use gets misread<br/>as consent or lack of need]
  E --> A

Distribution

A Person Lives With the Consequences; the System Keeps Running

People experience an action from the inside, but institutions tend to process it as throughput. Use this when a system calls something simple or available while the person inside it is the one exposed.

Humans have to inhabit actions with bodies, identities, and consequences, while systems keep executing until recourse, redesign, or refusal interrupts the loop.
Humans have to inhabit actions with bodies, identities, and consequences, while systems keep executing until recourse, redesign, or refusal interrupts the loop. How to read this
View source
%% title: A Person Lives With the Consequences; the System Keeps Running
%% caption: Humans have to inhabit actions with bodies, identities, and consequences, while systems keep executing until recourse, redesign, or refusal interrupts the loop.
flowchart TD
  A[Human] --> B[Must inhabit the action]
  B --> C[Feels risk, conscience,<br/>identity cost, exposure]
  C --> D[Needs recourse,<br/>redesign, or a stopping point]
  E[System] --> F[Executes rule or workflow]
  F --> G[Optimizes continuity,<br/>throughput, legibility]
  G --> H[Pressure to continue<br/>or normalize the cost]
  H --> D

Correction

Does the Harm Get Back to the Institution?

Dependence is legitimate where the defect it produces can travel back and change the system. It becomes domination where the defect stays with the person it landed on: the difference between a return channel and a dead end. Use this when an arrangement justifies itself by how much it relies on people's endurance. The open question is whether that reliance ever becomes a reason the institution must change.

Institutions depend on people's time, labor, and endurance. When the harm can travel back and force change, the dependence is legitimate; when it stays with the person, it is domination.
Institutions depend on people's time, labor, and endurance. When the harm can travel back and force change, the dependence is legitimate; when it stays with the person, it is domination. How to read this
View source
%% title: Does the Harm Get Back to the Institution?
%% caption: Institutions depend on people's time, labor, and endurance. When the harm can travel back and force change, the dependence is legitimate; when it stays with the person, it is domination.
flowchart TB
  A[Institution depends on<br/>people's time, labor, proof, endurance] --> B{Who absorbs the cost and<br/>who is licensed to repair it?}
  B --> C[Harm gets standing<br/>and an upstream return channel]
  C --> D[Correction binds the institution]
  D --> E[Power stays contestable]
  E --> A
  B --> F[Harm stays privately absorbed<br/>with no return channel]
  F --> G[Burden is normalized as resilience;<br/>non-use is misread as consent]
  G --> H[The institution's surface stays smooth]
  H --> A

Correction

What It Takes for an Error to Change Anything

Corrigibility depends on the whole path holding, from the first stage to the last. An institution can see an error and still change nothing. Each stage is a separate place where evidence can lose its hold on the institution. Use this to find where a failed correction died. Look at the person who could only plead, the one-off exception, the absorbed harm, and the advisory finding. Then look at the revision that stalled or came out patchwork, the acknowledgment that changed nothing, the institution grading its own exam, and the forgotten trial.

An error often gets dealt with only if it travels the whole path. On this path it goes from the person stuck with it, past the thresholds that make it count, into a consequence the institution can't refuse, and into a record that later revisions can find.
An error often gets dealt with only if it travels the whole path. On this path it goes from the person stuck with it, past the thresholds that make it count, into a consequence the institution can't refuse, and into a record that later revisions can find. How to read this
View source
%% title: What It Takes for an Error to Change Anything
%% caption: An error often gets dealt with only if it travels the whole path. Each stage is a separate place where evidence can lose its hold on the institution.
flowchart TD
  A[Standing:<br/>who may report] --> B[Learning rate:<br/>when variance indicts]
  B --> C[Burden tracing:<br/>who pays during review]
  C --> D[Correction authority:<br/>who can force revision]
  D --> E[Scale:<br/>how revision propagates]
  E --> F[Consequence:<br/>what must change]
  F --> G[Measurer:<br/>who defines the test]
  G --> H[Memory:<br/>where the record lives]
  H -.-> A

Correction

Correction Authority Is Divisible

The political fight is usually over which of the nine authorities an actor holds, rather than whether correction exists at all. A channel can grant the first four and still force nothing to change. Use this when a participatory process, appeal, or audit claims to give standing, and the open question is how far down the ladder the claimant's authority actually travels.

Correction authority is not one power but nine separable ones, running from observing to terminating. Being affected justifies strong rights in some without implying equal competence in all.
Correction authority is not one power but nine separable ones, running from observing to terminating. Being affected justifies strong rights in some without implying equal competence in all. How to read this
View source
%% title: Correction Authority Is Divisible
%% caption: Nine separable authorities run from observing to terminating. Being affected justifies strong rights in some without implying equal competence in all.
flowchart TB
  A[Observe the evidence] --> B[Report it]
  B --> C[Trigger review]
  C --> D[Contest the interpretation]
  D --> E[Adjudicate the error]
  E --> F[Specify the remedy]
  F --> G[Implement the technical response]
  G --> H[Reopen the jurisdiction]
  H --> I[Terminate or<br/>replace the institution]

Correction

Fixing the Case, the Rule, or Who Decides

A correction is as deep as the authority that makes it, and what it leaves running is the evidence for the next depth. An appeal restores one file and leaves the rule running on every file like it. A rule change reaches every file and can leave the saving it was built for to the next rule. Only a body above the program can narrow who may cut coverage off, and whoever holds that power next can still be wrong. The dashed path is a complaint that is heard and corrects nothing. Use this when a remedy is announced for a failure and the question is how deep it reached: which of the three was corrected, who had to sign it, and what the residual says about the level the source sits on. The correction ladder instrument walks all six rungs; this draws the three the arguments most often confuse.

One wrongful cutoff, corrected at three depths. Each correction has to be made by a different authority, and what each leaves running points to the next depth.
One wrongful cutoff, corrected at three depths. Each correction has to be made by a different authority, and what each leaves running points to the next depth. How to read this
View source
%% title: Fixing the Case, the Rule, or Who Decides
%% caption: One wrongful cutoff, corrected at three depths. Each correction has to be made by a different authority, and what each leaves running points to the next depth.
flowchart TD
  classDef actor stroke-width:1.6px;
  classDef binding stroke-width:3px,font-weight:bold;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
  classDef status stroke-dasharray:3 3;
  E([Coverage ends because<br/>a notice went unanswered]):::actor
  E -.-> H[Complaint line: heard,<br/>with no power to correct]:::failure
  E --> A[The case: an appeal<br/>finds the wrong address]
  A == "made by the appeal review" ==> A1[One family's coverage<br/>restored]:::binding
  A1 -- "still running" --> A2[The rule, on every<br/>file like hers]:::failure
  A2 --> B[The rule: silence stops<br/>ending coverage]
  B == "made by the rule's owner<br/>or a court" ==> B1[Every file under<br/>the rule changes]:::binding
  B1 -- "still running" --> B2[The saving the rule was for,<br/>sought by the next rule]:::failure
  B2 --> J[The authority: who may end<br/>coverage, on what proof]
  J == "made by a body above<br/>the program" ==> J1[The power to cut off<br/>is narrowed or withdrawn]:::binding
  J1 -- "still open" --> J2[Whoever holds it next<br/>can still be wrong]:::status

Correction

More Evidence Puts More Under Review

Not every error moves down the tree, and nothing moves automatically. The tree says when bigger questions become fair to ask, guarding against both overreaction and self-protective rigidity. Use this when an institution answers every adverse finding with a fix at the level the finding can't reach. The open question is what the pattern of repeat failures has already earned the right to put on trial.

The scope of review grows with the evidence. If a failure repeats, the rule goes on trial; if it survives reform, the jurisdiction itself is contested.
The scope of review grows with the evidence. If a failure repeats, the rule goes on trial; if it survives reform, the jurisdiction itself is contested. How to read this
View source
%% title: More Evidence Puts More Under Review
%% caption: What gets reviewed grows with the evidence, from implementation to jurisdiction to whether the institution should continue at all.
flowchart TB
  A[Isolated execution failure] --> B[Review implementation]
  B --> C{Repeats under<br/>the same rule?}
  C -- no --> D[Stop: fix held]
  C -- yes --> E[Review the rule]
  E --> F{Repeats across<br/>rules?}
  F -- no --> D
  F -- yes --> G[Review incentives and<br/>governance structure]
  G --> H{Persists<br/>after reform?}
  H -- no --> D
  H -- yes --> I[Review the jurisdiction]
  I --> J{Fundamentally<br/>unreformable?}
  J -- no --> D
  J -- yes --> K[Withdraw, replace,<br/>or terminate]

Correction

A Settled Decision Should Say How It Can Be Reopened

The settlement spells out its own reopening conditions: who can reopen it, on what evidence, at what cost. Non-finality is still a decision, one that states the terms on which it can be revised. Use this when an institution defends a closure as settled, and the open question is whether anyone was named with the standing to reopen it and whether the evidence conditions were written down anywhere.

A decision can be final enough to act on without being closed forever. Operational finality lets institutions act on settlements, while non-finality keeps the rule and its correction mechanism open to challenge over time.
A decision can be final enough to act on without being closed forever. Operational finality lets institutions act on settlements, while non-finality keeps the rule and its correction mechanism open to challenge over time. How to read this
View source
%% title: A Settled Decision Should Say How It Can Be Reopened
%% caption: A decision can be final enough to act on without being closed forever. The settlement states who can reopen it, on what evidence, at what cost.
flowchart TD
  A[Settlement:<br/>act on it now] --> B{Reopening<br/>conditions written?}
  B -- yes --> C[Named standing,<br/>evidence threshold,<br/>cost allocation]
  C --> D[Defeasible settlement:<br/>final for action,<br/>not immune forever]
  B -- no --> E[Exempt layer<br/>with a signature]
  E --> F[Institution cannot act<br/>without defending closure<br/>as permanence]

Correction

What a Rule Does With the People It Fails

The quadrant describes without grading. Coercive externalization, protective differentiation, extractive incorporation, and recursive reconstruction are four ways of handling the same misfit. Whether control is concentrated or distributed decides if any of them can be contested. Use this when an arrangement's rules fail people and the open question is what happened to the variance, and who had the authority to deal with it.

Two variables govern what an arrangement does with the people its rules fail: what happens to the variance, and whether the governing structure can be revised. A separate axis — who controls the response — decides whether any quadrant can be contested.
Two variables govern what an arrangement does with the people its rules fail: what happens to the variance, and whether the governing structure can be revised. A separate axis — who controls the response — decides whether any quadrant can be contested. How to read this
View source
%% title: What a Rule Does With the People It Fails
%% caption: What an arrangement does with the people its rules fail: variance externalized or incorporated, structure insulated or revisable, and a separate axis for who controls the response.
flowchart TB
  A{What happens to the<br/>rule's misfit?} --> B[Externalized:<br/>routed outside the rule]
  A --> C[Incorporated:<br/>absorbed as input]
  B --> D[Insulated structure:<br/>coercive externalization]
  B --> E[Revisable structure:<br/>protective differentiation]
  C --> F[Insulated structure:<br/>extractive incorporation]
  C --> G[Revisable structure:<br/>recursive reconstruction]
  D --> H{Who controls<br/>the response?}
  E --> H
  F --> H
  G --> H
  H --> I[Concentrated:<br/>no quadrant is contestable]
  H --> J[Distributed:<br/>no quadrant is a ranking]

Correction

Internal Learning vs External Selection

Neither mechanism works alone, and each fails in its own signature way. Internal learning risks incumbent capture, cosmetic reform, and organizational forgetting. External selection risks catastrophic exit costs, absent alternatives, and monopoly lock-in. Use this when an arrangement claims to correct itself, and the open question is whether its correction has a working rival that failure can't capture.

Internal learning preserves capacity but leaves insiders in charge of reading failure, while external selection replaces what fails but requires a real exit. Sturdy correction needs both, and each must be open to challenge.
Internal learning preserves capacity but leaves insiders in charge of reading failure, while external selection replaces what fails but requires a real exit. Sturdy correction needs both, and each must be open to challenge. How to read this
View source
%% title: Internal Learning vs External Selection
%% caption: Internal learning keeps capacity but leaves insiders in charge of how failure is read; external selection punishes failure but needs a real way out. Sturdy correction needs both, and each must be open to challenge.
flowchart TD
  A[Institution fails] --> B[Internal learning:<br/>revise model, rule,<br/>reallocation of authority]
  A --> C[External selection:<br/>exit, competition,<br/>loss of jurisdiction]
  B --> D[Signature failures:<br/>incumbent capture, cosmetic<br/>reform, forgetting]
  C --> E[Signature failures:<br/>exit too costly, no<br/>alternatives, lock-in]
  D --> F[Contestability of<br/>the internal channel]
  E --> G[Contestability of<br/>the external channel]
  F --> H[Redundant correction<br/>capacity]
  G --> H

Correction

Anticipated Sanction → Observed Consensus Loop

The willingness to disagree is shaped by the institution's own expected response, so the level of visible opposition cannot measure how much opposition exists. Use this when an institution points to a quiet comment section as evidence that its people feel free to object.

An institution reads the absence of dissent as evidence of openness, while the dissent it would have heard was filtered by what people expected disagreement to cost.
An institution reads the absence of dissent as evidence of openness, while the dissent it would have heard was filtered by what people expected disagreement to cost. How to read this
View source
%% title: Anticipated Sanction → Observed Consensus Loop
%% caption: An institution reads the absence of dissent as evidence of openness, while the dissent it would have heard was filtered by what people expected disagreement to cost.
flowchart TD
  A[Institution permits dissent] --> B[People estimate what<br/>using it would cost]
  B --> C[Criticism that would test<br/>the arrangement is withheld]
  C --> D[Record shows little<br/>fundamental opposition]
  D --> E[Low opposition is read<br/>as consent or openness]
  E --> A

Correction

Good vs Bad Insulation

A judge needs insulation from improper pressure. The same barrier becomes domination when it blocks the harm the institution caused from reaching the actor who can change the rule. Use this when an institution defends its unaccountability as 'independence' — the test is whether the insulation blocks improper pressure or blocks consequence.

Insulation from noise, retaliation, or panic protects a decision. The problem is insulation from evidence and consequence.
Insulation from noise, retaliation, or panic protects a decision. The problem is insulation from evidence and consequence. How to read this
View source
%% title: Good vs Bad Insulation
%% caption: The problem is not insulation from noise, retaliation, or panic. It is insulation from evidence and consequence.
flowchart TB
  classDef repair stroke-width:2.4px,font-weight:bold;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
  subgraph GOOD[Good insulation: blocks improper pressure]
    direction TB
    G1[Noise / retaliation / panic] -. blocked .- G2[Decision-maker]:::repair
    G2 --> G3[Still receives evidence + accountability]:::repair
  end
  subgraph BAD[Bad insulation: blocks evidence and consequence]
    direction TB
    B1[Harm / contradiction] -. blocked .- B2[Decision-maker]:::failure
    B2 --> B3[Keeps authority unchanged]:::failure
  end
  GOOD ~~~ BAD

Correction

The Capacity to Act and the Capacity to Be Corrected

Modernization measured on the first axis alone mistakes an unanswerable machine for a developed one. The corrigible-capability quadrant is the design target, and the capable-and-unanswerable quadrant is the trap development normally drifts into. Use this when an institution's growth story is told entirely in throughput, and the open question is whether the ability to contest it grew with the ability to act.

Administrative capacity and corrective capacity are separate axes. An arrangement can become more capable and less corrigible at once, and the quadrant decides what its power is worth.
Administrative capacity and corrective capacity are separate axes. An arrangement can become more capable and less corrigible at once, and the quadrant decides what its power is worth. How to read this
View source
%% title: The Capacity to Act and the Capacity to Be Corrected
%% caption: Administrative capacity and corrective capacity are separate axes. An arrangement can become more capable and less corrigible at once, and the quadrant decides what its power is worth.
flowchart TB
  classDef system stroke-width:1.2px;
  classDef gate stroke-width:1.6px;
  classDef repair stroke-width:2.4px,font-weight:bold;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
  A{Can it know, classify,<br/>decide, enforce?}:::gate --> B[High administrative<br/>capacity]:::system
  A --> C[Low administrative<br/>capacity]:::system
  B --> D{Can subjects detect, hear,<br/>reverse, repair?}:::gate
  C --> D
  D -- yes --> E[Corrigible capability:<br/>power that can be challenged]:::repair
  D -- no --> F[Capable and unanswerable:<br/>modernization without correction]:::failure
  C -- yes --> G[Small and answerable]:::system
  C -- no --> H[Weak and unreachable:<br/>where accidental freedom lives]:::system

Jurisdiction

Being Right Doesn’t Make It Yours to Decide

The valid path runs from a fact, through its relevance to this decision and a separate justification, to limited authority. The invalid path jumps from the fact straight to 'therefore we may govern more', and that jump is where domination gets in, dressed up as competence. Run in reverse, the rule protects the person: their adaptability is a fact about them, and the institution cannot treat it as a standing offer. Use this when an institution's claim to govern arrives as a record of what it knows and what it has survived, and the open question is where the argument for this particular authority begins.

Expertise, dependence, and successful output are facts, but none converts into authority without a separate justification. Likewise, a person's capacity to absorb harm is a fact about them, giving the institution no legitimate claim on their endurance.
Expertise, dependence, and successful output are facts, but none converts into authority without a separate justification. Likewise, a person's capacity to absorb harm is a fact about them, giving the institution no legitimate claim on their endurance. How to read this
View source
%% title: Being Right Doesn’t Make It Yours to Decide
%% caption: Facts do not convert into authority without independent justification. Expertise is not sovereignty; survival is not legitimacy; success is not competence.
flowchart TD
  A[Fact: expertise,<br/>dependence, necessity,<br/>participation, survival,<br/>success] --> B{Specific relevance<br/>to this decision?}
  B -- no --> C[No authority<br/>follows]
  B -- yes --> D{Independent<br/>justification<br/>for the extension?}
  D -- no --> C
  D -- yes --> E[Bounded authority,<br/>renewal conditions attached]
  A -.->|the invalid path:<br/>therefore we may<br/>govern more| F[Unbounded authority]

Jurisdiction

Giving a System the Right Goal Doesn’t Give It Authority

Mainstream alignment optimizes what the system should want (its principal's intent) and loses two other questions along the way: what the system has jurisdiction over, and who has corrective standing to challenge it. Use this when an AI safety framework claims alignment solves governance, ignoring what the law lets the system decide, who holds each decision, and how citizens get recourse.

Giving a system the right objective does not give it legitimate authority. Governing also means settling what the system may decide and how the people it affects can force a reversal.
Giving a system the right objective does not give it legitimate authority. Governing also means settling what the system may decide and how the people it affects can force a reversal. How to read this
View source
%% title: Giving a System the Right Goal Doesn’t Give It Authority
%% caption: The institutional trilemma: alignment optimizes the system's goal, but legitimate authority also needs jurisdiction and corrective standing.
flowchart TB
  classDef system stroke-width:1.2px;
  classDef gate stroke-width:1.6px;
  classDef repair stroke-width:2.4px,font-weight:bold;
  subgraph TEL [1. Teleological Alignment]
    A[Principal Objective]:::system --> B[Model Optimization]:::system
  end
  subgraph JUR [2. Bounded Jurisdiction]
    C[Statutory Mandate]:::gate --> D[Evidentiary Boundary]:::gate
  end
  subgraph COR [3. Institutional Corrigibility]
    E[Residual Harm]:::repair --> F[Corrective Standing]:::repair
    F --> G[Binding Reversal]:::repair
  end
  B -. Cannot Subsume .- C
  D -. Must Connect To .- E

Continue reading

New essays by email

One situation a week, the essays that explain it, and one question to put to the institution.

New essays by email

One situation a week, the essays that explain it, and one question to put to the institution.