The Stack

I. The twelve layers

A given problem rarely sits at one layer alone — but the canonical failure mode usually does. The diagnostic move is to ask, layer by layer, where the binding failure is. A policy that has the right mechanism (1) and the right ledger (5) but no decision-side enforcement (8) will fail at the decision layer; a policy whose decision-side is healthy but whose execution layer (9) has been hollowed out will fail at execution. The vocabulary lets the post-mortem be specific.

II. The layers, with corpus pointers

0. Telos — what is the system for?

The question. What is the system actually selecting for, explicitly or implicitly? An institution without an owned, articulated purpose is being optimised by selection pressures it does not see.

Healthy primitive. Explicit telos, explicit tradeoffs, named objective function.

Failure family. No owner of the purpose; drift; capture by proxy objective; comfort-and-preservation as terminal value; multi-objective incoherence.

In this corpus.

• The Telos Gap — real, mechanism-produced harms that fail to enter any institution's mandate, budget, KPI, or procedural trigger
• The Selection Question — what is the population compounding toward?
• The Axiological Malthusian Trap — Foundry → Abundance → Hospice transition
• The Hospice AI Problem — AI optimised for comfort rather than aliveness
• Values Aren't Subjective, Values Are Ecology, Values Are Replicators — the three-essay treatment of how values are selected, not chosen
• Belonging Is Axiology — coordination boundaries as objective-function boundaries
• The Question Nobody Asks — what does physics actually require for a complex system to persist?

1. Mechanism — how does it actually work?

The question. What rule, incentive, constraint, or architecture is supposed to produce the outcome? "We will pass a law about X" is not a mechanism; "the law changes the cost of Y, which causes actor A to substitute toward Z" is.

Healthy primitive. Explicit mechanism claim with named actor, changed constraint, expected response, target output, time horizon.

Failure family. Intent treated as causal force; legal text assumed to compile into operational behaviour; moralisation replaces computation.

In this corpus.

• Mechanism Realism — outcomes are caused by mechanisms, not intentions
• The Mechanism Analysis — the missing artifact that tests the causal machine before enactment
• How Mechanism Analyses Are Made — production discipline behind the artifact
• Mechanism Space — the space of available mechanisms, not just the proposed one
• The Halting Problem of Law — legal text must be compiled into operational behaviour
• Laws Are the Wrong Abstraction — incentives and architecture do the work, not the law's words
• The Environment Is the Author — the environment writes behaviour the policy can only nudge
• The Mechanist Tradition — the lineage of mechanism-first thinking
• Physics to Practice — translating physical constraint into institutional design

2. Response — what do agents do next?

The question. Once the rule changes, what does the next actor find rational, cheap, safe, or unavoidable? Policy lands inside an existing strategic landscape and triggers a response. The response is a property of the actor and the architecture, not of the legislator's intent.

Healthy primitive. Mapped response channels; predicted dominant response; counterfactual against status quo.

Failure family. Pressure routes through the cheapest channel — compliance gaming, formal substitution, withdrawal, exit, lobbying, reclassification, passive non-uptake.

In this corpus.

• The Response Vector — pressure routes into target response, base loss, formal gaming, or incidence shift
• Bad Equilibria Are Not One Thing — different stuck states require different repairs
• The Capability Trap — the system consumes the capacity it claims to build
• Trapped Equilibria — multi-input production-function bottlenecks
• Optionality Has No Router — exit-as-response when the system has no inside repair
• Generator-Substitute AI — AI bypassing or replacing the practitioner chain (cross-listed at Reproduction)

3. Carrier — who can carry harm into the decision system?

The question. Which harms become visible because they have someone or something to carry them into the institution that could repair them, and which harms remain invisible because no carrier exists?

Healthy primitive. Standing for the silent class; explicit bearer of diffuse cost; statistical victim given a procedural voice.

Failure family. One side has a carrier (legally, organisationally, narratively); the other is silent. The silent side absorbs the cost. The carrier-equipped side writes the rules.

In this corpus.

• The Asymmetric Carrier Problem — the structural form of silent-side harm
• The Attenuation Layer — costs route toward the lowest-attribution carrier
• The Copenhagen Trap — action creates liability; inaction does not
• Theatrical Accountability — response-form replaces repair

4. Measurement — what signal is treated as reality?

The question. What does the system observe, and what does it treat as evidence of success or failure? Measurement is never neutral: once a measure is attached to reward, punishment, or legitimacy, the actors inside the system optimise the measure, not the underlying reality it was meant to track. Measurement failure is upstream of ledger failure; before you can count the wrong stock, you can measure the wrong proxy.

Healthy primitive. Measurements that remain coupled to the underlying reality; proxy treated explicitly as proxy; audit path from metric back to territory; periodic recalibration when the relationship between proxy and reality drifts.

Failure family. Goodharting; proxy-target divergence; cargo-cult epistemology; metric-compliance replacing reality contact; map detaching from territory.

In this corpus.

• Cargo Cult Epistemology — epistemic signals used as status or tribal markers after decoupling from reality-mapping
• The Severed Map — the map detached from the territory it claimed to model
• The Information Gradient — distance degrades signal quality before the system can measure, decide, or correct

5. Ledger — what gets counted as cost?

The question. Which capital stocks does this intervention draw down to fund the visible benefit, and are those stocks on a ledger anyone consults? Most of what a civilisation runs on is booked at zero by default; that booking decides what looks like a gain.

Healthy primitive. Full accounting across all load-bearing stocks; stock + flow + recovery-horizon for each.

Failure family. Hidden depletion masked by visible flow; "hard to measure" booked as zero; balance-sheet theft at civilisational scale.

In this corpus.

• Full Accounting — why "hard to measure" is not the same as zero
• Capital Stocks — the reference page for the dozen stocks Full Accounting names
• Flourishing Is Maximum Safety Margin — Liebig's Law applied to the consolidated balance sheet
• Fertility Requires Margin — demographic capital as multi-input production function
• The Tyranny of the Present — civilisation-scale hyperbolic discounting
• Ultimate Waste — the cost of optionality consumed
• Retirement Is Anti-Life — human-capital depletion treated as a category error

6. Compilation — does the right decision-frame exist?

The question. Have the relevant academic substrate, statistical evidence, and policy considerations been assembled into a public-usable decision frame that a generalist can actually pick up and use? Three distinct objects often get confused: academic synthesis, institutional compilation outputs, and the publicly-pickable decision-frame artifact. The third one is the one that is frequently missing. Compilation is the supply-side question; whether the frame is actually used once it exists is the Computation/adoption layer below.

Healthy primitive. Parameterised public frame with named variables, causal structure, an update rule, and a policy-evaluation test.

Failure family. The pieces exist; the assembled frame does not.

In this corpus.

• Non-Compilation — supply-side failure: the artifact is missing
• The Unpopulated Meta — the higher-order frame nobody is writing
• The Framing Machine — institutional incentives that reward frame-production over mechanism-building
• Complexity Laundering — unmodelled causal chains let costs vanish before they enter any decision frame
• The Epicycles of Debt — the intellectual apparatus for denying what compilation would reveal

7. Computation / adoption — does the system actually use the frame?

The question. Once a decision frame exists, does the system actually compute with it — or does it convert the frame into slogans, factional ammunition, or moral cover? Compilation and computation are separable problems: a perfectly compiled frame can land in a discourse architecture that refuses to compute with it. The two failures call for different repairs.

Healthy primitive. Parameterised reasoning made procedurally consequential; disagreement attached to named variables; values separated from empirical claims; institutional teeth that make ignoring the frame expensive.

Failure family. Computation refused; empirical claim hidden as value disagreement; frame captured by faction; frame sloganised; public debate remains slogan-shaped despite the compiled artifact existing.

In this corpus.

• Calculemus — most legislative disagreement is uncomputed empirical disagreement masked as values
• Non-Compilation — supply-side sibling: the frame may not exist at all (cross-listed from Compilation)
• What "Vote on Values" Actually Does — democratic preference aggregation as replicator-fitness measurement rather than value computation

8. Decision — who must choose, repair, or override?

The question. Once the mechanism failure is visible and the frame exists, who is forced to decide, and how is invisible evasion prevented? Diagnosis without a forced decision is decorative.

Healthy primitive. Repair-or-override loop with named owner; political override permitted but visible; no quiet decay.

Failure family. No one is forced to act; the warning is filed; the system continues. Or the decision is delegated until no owner remains.

In this corpus.

• Powerless Intelligence — analysis exists; authority does not
• The Fourth Branch — the constitutional architecture for a forced-decision institution
• Telocracy — the objective function that makes the decision actionable
• The Governance Alignment Problem — why the system selects against honest decision-making
• The Alignment Problem in Your Government — AI-safety framings applied to government
• Steering Power — capacity × coupling × salience; why some warnings move nations and others drift
• Proportional Representation: The Panic Architecture — decision-layer failure under coalitional politics

9. Execution — does the system actually implement the repair?

The question. Once a decision is made, does the receiving institution have the capacity, authority, and feedback loop to execute the repair without producing worse distortion than the original problem?

Healthy primitive. Implementation ledger with owner, trigger, and a movement test that distinguishes activity from result.

Failure family. Critique absorbed without repair; reform-form without transformation; smart mechanism failing because implementation capacity is missing.

In this corpus.

• Constructive Diagnosis — diagnosis must compile into a repair specification, not stop at critique
• The Containment Pattern — institutional immune response that absorbs critique without changing
• Simulated Metamorphosis — reform-form without transformation
• Capability Architecture — what implementation capacity actually requires
• Libertarianism Is an Incomplete Solution — why "more / less state" misses the execution-architecture question

10. Feedback — does reality update the system?

The question. After the policy is enacted, does observed failure feed back into mechanism correction — or only into narrative? Most evaluation produces reports; very little produces correction.

Healthy primitive. Post-enactment measured movement test; correction duty with consequence; visible feedback loop from outcome to rule.

Failure family. Decorative evaluation; advisory reports that bind nothing; audits that observe but do not alter the mechanism.

In this corpus.

• The Legitimacy Gate — what feedback must pass before the system updates
• Legitimacy Came Before Cognition — why feedback architectures often pre-date the rationalisation that justifies them
• Nullius in Verba — the Royal Society's wager and what it requires institutionally
• When Does Reform Actually Happen? — the five empirical vectors of structural change
• Why Science Advances Funeral by Funeral — when generational replacement is the only feedback path

11. Reproduction — does the generator-chain continue?

The question. Does the system reproduce the capacity that made its outputs possible — the apprenticeship, the lineage, the standards, the tacit knowledge? Output can continue for years after the generator is dead. The persistence is misleading: the corpse is still standing.

Healthy primitive. Live generator-chain; apprenticeship and intergenerational transmission; explicit standards; spore strategies for hostile environments.

Failure family. Output without generator; chain breakage; institutional shell surviving the competence that filled it; AI substituting for the practitioner-chain that would have produced the next generation of practitioners.

In this corpus.

• Sterile Generativity — output persists while the generator chain dies
• The Spore Strategy — preserving generators when macro-selection is hostile
• The Original Sin of History — the founding-cohort effect on what reproduces
• The Nokia Phase Transition — generator-chain collapse as national capability loss
• The Replicator Immune System — what protects a healthy chain from invasive forms
• Generator-Substitute AI — AI produces artifacts while bypassing future practitioners

III. The structural law underneath

One sentence keeps surfacing across the twelve layers and is worth naming as the master form:

The twelve layers are twelve different ways the institution can fail to own a channel through which reality enters. Unowned purpose produces telos gaps. Unowned mechanism produces intent-substitution and moralised causal disputes. Unowned actor response produces the four-channel pressure routing. Unowned harm produces silent-side absorption. Unowned measurement produces cargo-cult epistemology and the severed map. Unowned cost produces false-zero accounting. Unowned decision frame produces non-compilation. Unowned computation produces the Calculemus slogan war. Unowned decision produces powerless intelligence. Unowned repair produces institutional containment. Unowned feedback produces theatrical accountability. Unowned generator-chain produces sterile generativity.

The diagnostic question for any failure: which channel is unowned, and by what authority would it have to be owned? That is the cross-layer question the stack is built to support.

IV. Using the map

The page is built to be a working object, not an essay to be read once. Three intended uses:

Triage. For any concrete failure — a policy that obviously isn't working, an institution that obviously isn't delivering, an organisation's capability that is obviously eroding — walk the twelve layers in order. Where is the binding failure? Multiple layers usually contribute, but one is usually load-bearing for the current state. Naming the layer is the first move; the corpus essays at that layer are the second.

Cross-referencing. When two essays in the corpus seem to overlap but are not the same thing, the layer assignment usually clarifies why. Non-Compilation at the Compilation layer (6) and Calculemus at the Computation/adoption layer (7) treat the supply side and demand side of the same problem, but at different layers — Compilation asks whether the frame exists; Computation asks whether the system uses it. Cargo Cult Epistemology at the Measurement layer (4) and Full Accounting at the Ledger layer (5) also treat related but distinct failures: measurement asks what signal counts as evidence; ledger asks what stock counts as cost. The layer assignments make the distinctions visible.

Gap-finding. Where a layer has fewer essays than its weight in the failure stack would predict, the gap is informative. Several layers currently have known under-treatments. Measurement (4) is freshly separated from Compilation in this version of the stack and probably wants a dedicated anchor essay (working title: The Metric Is Not the Territory or Proxy Realities) that gathers the Goodhart family explicitly. Computation/adoption (7) anchors on Calculemus but could use a sharper stack-specific piece (working title: The Refusal to Compute) that distinguishes value disagreement from empirical-parameter refusal. Execution (9) is sparser than it should be relative to how often execution capacity is the binding failure; an Implementation Ledger essay would be load-bearing. Feedback (10) covers feedback architectures but under-treats the typology of feedback authority — a Feedback Authority essay distinguishing decorative / advisory / reputational / procedural / budgetary / statutory / constitutional binding force is the canonical missing slot. Reproduction (11) has strong coverage on chain breakage and weaker coverage on chain formation; a How Generator-Chains Form essay is the non-trivial missing half. The stack is a working map; the gaps are part of what the map is for.

The stack is not exhaustive of the corpus. Several essays are foundational (the physics layer beneath the stack) — The Four Axiomatic Dilemmas, The Physics of Intelligence, The Physics of Moloch, The Thermodynamics of Power. Others are domain-applied (AI alignment, Finnish-specific institutional analysis, the alignment-via-physics treatments). Others are about the craft of writing the essays themselves (Essay Engineering, Optimal Prose, How I Work). Those are listed on the main essays index rather than here, because the stack is a failure-mode taxonomy for goal-directed systems and they sit alongside it rather than inside it.

Related references:

• Full Accounting — the conceptual treatment of the ledger layer
• Capital Stocks — the operational reference for the dozen stocks that need a ledger
• The Layer Walk — the operational sibling: how to walk a real failure through the layers
• Essays index — the full corpus organised by topic cluster rather than by layer