Memo 09 — Evolution of Normative Systems: Mutation Mechanisms and Feedback Loops

Series: The Landscape of Normative Systems (LNS) Memo number: 09 of 12 Primary JD questions: Q2 (the mutation mechanism slot of the six-part analysis, here treated comparatively across all sources), Q5 (identification of normative feedback loops). Secondary: Q4 (the dependency graph, whose back-edges are the loops of Q5), Q7 (relations among the domains, read dynamically). Upstream constraints: D0 methodology M1–M8; conventions §0.4 (epistemic tags), §0.5 (four standing categories), §0.6 (shared vocabulary: norm / source / validity vs legitimacy vs efficacy / Hohfeldian relations / deontic O/P/F). Per §0.2, D0 is immutable and is neither criticized nor defended here.

Scope note. This memo owns the dynamics of the series. Memos 01–08 photographed each source at rest — its origin, its legitimacy story, its enforcement organ, its hierarchy, its conflict rules. This memo studies the derivative: how a norm-set at time t becomes a different norm-set at time t+1, and how the outputs of a normative system loop back to become its own inputs. It has three jobs. First, it assembles a comparative catalogue of mutation mechanisms — formal amendment, judicial reinterpretation, legislative override, regulatory update, desuetude/obsolescence, revolution/rupture, drift, transplant/borrowing, codification of custom, precedent overruling, schism/forking, and market selection — and reads each as an operation on a norm-set. Second, it imports three cross-disciplinary models of norm change — variation–selection–retention (evolutionary, with memetics flagged strictly as HYPOTHESIS), path dependence and lock-in (institutional economics), and punctuated equilibrium — and asks which is testable and which is metaphor. Third, and most importantly for the compiler team, it discharges JD Q5 by drawing many normative feedback loops as explicit ASCII cycle diagrams and, for each, classifying the loop in control-theory terms: its delay (transport lag around the cycle), its loop gain (whether a perturbation is amplified or attenuated per revolution), and its sign (negative/stabilizing vs positive/amplifying feedback). We apply the mandatory six-part analysis (JD Q2) in full to precedent/tradition as a source, because precedent is the purest case of a norm-set that is its own change mechanism. We close with the computational implications that dynamics forces on D1/D2: versioning, migration, non-monotonic update, and loop stability/oscillation. All substantive claims carry epistemic tags per §0.4; when uncertain we tag down.

09.0 Orientation: why dynamics is a separate memo

INFERENCE A recurring temptation in a formalization program is to treat a normative system as a fixed object — a set of rules to be compiled once. Every source memo in this series contradicts that picture. Legislation is amended (Memo 01), precedent is overruled (Memo 02), moral consensus shifts (Memo 03), doctrine schisms (Memo 04), markets re-price and thereby re-norm (Memo 05), social conventions drift and lapse into desuetude (Memo 06), organizational policy is versioned (Memo 07), and treaties are renegotiated or denounced (Memo 08). A norm-set is not a value; it is a trajectory. The central methodological claim of this memo:

HYPOTHESIS Any faithful representation of a normative system must be a stateful, versioned, time-indexed object, and the mutation operators are as much a part of the system's identity as the norms themselves. A compiler that ingests a norm-set as a static snapshot has discarded exactly the information that distinguishes a living legal order from a museum exhibit. Stated as hypothesis because it is a design claim about D1/D2 that could be falsified if some class of governance turns out to be genuinely static; we know of no such class among the sources surveyed.

INFERENCE There is a second reason dynamics deserves its own memo. The JD's idealized dependency graph (Q4: politics → constitution → law → regulation → procedure → job description → KPI) is a directed acyclic picture. Reality is not acyclic. Almost every source memo reported a back-edge: law shapes the society that elects the legislature (Memo 01, Memo 06), doctrine shapes the ethics that later reforms the doctrine (Memo 03, Memo 04), prices shape the behavior that sets prices (Memo 05). The back-edges are the feedback loops of Q5. Dynamics and cyclicity are therefore the same subject viewed twice: mutation is what happens along an edge over time; a feedback loop is what happens when the edges close into a cycle. §09.1–09.5 treat mutation edge-by-edge; §09.6–09.9 treat the closed cycles.

A terminological discipline carried throughout. We keep the §0.6 triad — validity (formal membership), legitimacy (accepted-as-binding), efficacy (actually complied with) — separate in the time dimension, because they mutate on different clocks and often in opposite directions:

Three properties, three clocks (schematic; INFERENCE):

Divergence between the three clocks is the raw material of almost every mutation mechanism below:

• desuetude $= \text{efficacy} \to 0$ while validity stays $1$
• revolution $=$ legitimacy of old order $\to 0$, forcing validity reset
• codification $=$ efficacy + legitimacy high, validity catches up

INFERENCE This "three clocks" picture is the organizing intuition of the memo. A great deal of what follows is a taxonomy of the ways the three clocks can drift apart and be re-synchronized.

09.1 Established consensus

This section states what the relevant fields (comparative law, legal history, institutional economics, historical sociology, evolutionary theory as applied to culture) broadly agree on about how norm-sets change. Live disagreements — above all the status of evolutionary models — are deferred to §09.3.

09.1.1 Norm-sets change through a small, recurring repertoire of operations

FACT Across every source surveyed, the ways a norm-set changes are not infinitely various. The same handful of operations recur: add a norm, repeal a norm, modify a norm's content, re-scope its subjects or conditions, re-rank it in the hierarchy, and re-interpret it without changing its text. INFERENCE Reduced to the level a compiler cares about, this is a small operator set on a typed collection — an edit algebra over norm-sets. The mechanisms catalogued in §09.2 are institutional wrappers around these few primitive edits: they differ in who may apply the edit, under what procedure, with what delay, and with what authority to bind. This factorization (primitive edit vs institutional wrapper) is a load-bearing simplification carried to Handoff.

09.1.2 Change is governed by second-order rules (rules of change)

FACT It is common ground in analytic jurisprudence (the account associated with Hart, Memo 02) that mature systems contain rules of change: secondary rules conferring power (Hohfeldian) to alter the primary rules, specifying who may legislate, by what procedure, within what limits. INFERENCE Mutation, in the well-behaved case, is itself norm-governed: the system carries its own amendment interface. The pathological cases (revolution, drift, desuetude) are precisely those where change occurs outside or against the rules of change. This gives a clean primary partition of the whole field:

INFERENCE This A/B/C partition is more compiler-relevant than any discipline-based taxonomy, because it tells the machine whether the change can be validated against the system's own rules (A: yes; B: partly — the interpreter's authority is checkable, the interpretation is not; C: no — the change must be recorded as a fact, not derived as valid). It recurs at Handoff as the recommended top-level type tag on any mutation event.

09.1.3 Change is uneven in time: long stasis punctuated by bursts

FACT Legal and institutional historians broadly observe that normative change is not smooth: systems exhibit long periods of relative stability interrupted by concentrated bursts of change (constitutional moments, waves of codification, post-crisis regulatory expansion, doctrinal revolutions). INFERENCE Whether this pattern is explained by punctuated-equilibrium theory or merely described by it is contested (§09.3.3); the pattern itself — stickiness plus occasional rapid reconfiguration — is not seriously disputed.

09.1.4 History matters: the past constrains the reachable future

FACT It is broadly accepted in institutional economics and legal history that the sequence of past choices constrains which future norm-sets are reachable and at what cost — the phenomenon labelled path dependence (§09.3.2). INFERENCE In edit-algebra terms: the mutation operators available at time t are a function of the state at t, and some target states are reachable only through prohibitively long or costly edit sequences. Not all norm-sets are reachable from all others. This is a hard, recurring constraint and is the reason "just rewrite it cleanly" is usually not an available move.

09.1.5 Sources change on characteristic timescales

INFERENCE The sources differ systematically in how fast they mutate and how visibly. This is consensus at the level of gross ordering, even where exact rates are contested:

Characteristic mutation speed (fast → slow; INFERENCE, ordinal only):

INFERENCE The compiler-relevant consequence is a timescale-mismatch problem: sources that are coupled in a dependency graph mutate at rates that differ by many orders of magnitude. A fast source (prices) coupled to a slow source (constitution) produces the classic control-theory hazard of a fast loop nested inside a slow loop — stiffness. Flagged for §09.7 (loop analysis) and Handoff.

09.2 The comparative catalogue of mutation mechanisms

This is the core descriptive contribution of the memo: each mechanism read as an operation on a norm-set, tagged by the A/B/C class of §09.1.2, with its characteristic delay, its reversibility, and its computational signature (what a compiler must do to represent it). All mechanism descriptions are FACT as to the institution's existence and general operation; the reductions to edit-operations are INFERENCE.

09.2.1 Formal amendment (class A)

FACT The paradigm endogenous change: a constitution or statute specifies a procedure (supermajority, referendum, ratification thresholds) by which its own text may be altered. The change is dated, textual, and validity-preserving by construction — the amendment is valid because the rule of change was followed.

• Edit signature: add / repeal / modify, applied to a versioned text, producing a new version with a known predecessor. INFERENCE
• Delay: high and deliberately high — amendment thresholds are engineered friction (Memo 01). FACT that thresholds exist; [INFERENCE that their function is friction.]
• Reversibility: reversible only by a further amendment; the procedure is symmetric but slow.
• Computational signature: the cleanest case for versioning. A total order (or DAG) of versions, each a validated delta over its predecessor, each carrying an effective-date. This is the floor case that D1 must support and against which all messier mechanisms are measured. INFERENCE

09.2.2 Legislation and legislative override (class A)

FACT Ordinary lawmaking adds/repeals/modifies statutory norms via the legislature's power-conferring rules. Override is the special case where a later legislative act deliberately displaces a norm previously established by another organ — e.g., a statute reversing a (non-constitutional) judicial interpretation, or a legislature overriding an executive veto. FACT that both occur; the specific override channels are jurisdiction-specific.

• Edit signature: add/repeal/modify, plus an explicit derogation edge ("this act displaces norm X"). INFERENCE
• Delay: moderate (a legislative session cycle).
• Conflict interaction: override is where the lex posterior conflict rule (later law displaces earlier, Memo 02) becomes a mutation rule. Ordering-in-time is doing double duty: it resolves conflicts and it drives change. INFERENCE
• Computational signature: requires the version graph to carry typed edges (supersedes, derogates-from, amends) not just a linear history, because an override is a semantic relation between norms, not merely a timestamp. INFERENCE

09.2.3 Judicial reinterpretation (class B)

FACT A court changes the operative content of a norm without changing its text, by adopting a new interpretation. The statute's words are identical before and after; what the words do is different. This is the paradigm interpretive mutation (class B) and the mechanism most corrosive to the assumption that a norm equals its text.

• Edit signature: no edit to the norm text; an edit to the interpretation function that maps text+facts to outcomes. INFERENCE This is the crucial point for D1: the mutable object is not always the norm; sometimes it is the reader.
• Delay: case-driven and irregular — change occurs only when a suitable case arises and reaches an authorized court. Event-driven, not scheduled. INFERENCE
• Reversibility: by later reinterpretation, by overruling (§09.2.10), or by legislative override (§09.2.2).
• Computational signature: forces a separation between norm text (versioned) and interpretation (also versioned, and separately). A compiler that stores only the text will silently return stale verdicts across a reinterpretation event with no textual diff to alert it. This is one of the strongest arguments in the series for versioning the interpretation layer independently. Cross-ref Memo 02 (open texture, hard cases). INFERENCE

09.2.4 Regulatory update (class A, high-frequency)

FACT Administrative agencies issue, amend, and rescind regulations under delegated power, typically far faster and more frequently than primary legislation, often via a defined procedure (notice-and-comment or equivalent). FACT that delegated rule-making exists and is higher-frequency; procedures are jurisdiction-specific. Cross-ref Memo 07.

• Edit signature: add/repeal/modify on a subordinate norm layer whose validity is contingent on the enabling statute (a validity-conferral edge upward). INFERENCE
• Delay: low-to-moderate; designed to track changing technical conditions.
• Hierarchy interaction: a regulatory update can be invalidated retroactively if held ultra vires (outside the delegated power) — so its validity is defeasible in a way primary legislation's is not. INFERENCE The version history may need to record not only "superseded" but "voided ab initio".
• Computational signature: a fast-mutating layer anchored to a slow-mutating parent, i.e., the timescale-mismatch of §09.1.5 in miniature. Requires cascade/re-validation when the parent changes. INFERENCE

09.2.5 Desuetude / obsolescence (class C)

FACT A norm that remains formally valid but ceases to be enforced or complied with, until it is effectively dead though never repealed. Recognized as a doctrine in some legal traditions (a valid law may lapse by long disuse) and contested in others; as a social process (Memo 06) it is uncontroversial. FACT that the phenomenon occurs; [FACT that its legal recognition varies by jurisdiction.]

• Three-clocks reading: efficacy → 0 and legitimacy often → 0, while validity stays at 1. The clocks diverge and are never re-synchronized by any dated event. INFERENCE This is the paradigm of the three clocks coming apart.
• Edit signature: no edit occurs. The norm-set is textually unchanged; the change is entirely in the (unrecorded) efficacy/legitimacy state. This is why desuetude is so hard for a text-based system: there is no diff to observe. INFERENCE
• Delay: very long and undated — there is no moment at which the norm "dies"; it fades.
• Computational signature: the strongest argument in the memo for carrying efficacy as first-class, decaying state rather than assuming valid⇒live. A compiler that equates validity with applicability will confidently enforce dead letters. Cross-ref Memo 06 (social desuetude), §09.10.3. INFERENCE

09.2.6 Revolution / rupture (class C, discontinuous)

FACT A change of the validity criterion itself — the rule of recognition / Grundnorm (Memo 02) is replaced rather than followed. The new order's norms are valid by a new standard that the old order would not recognize; legally, the chain of validity is broken, not extended. FACT that such ruptures occur historically; the jurisprudential reading is [INFERENCE following Kelsen/Hart.]

• Three-clocks reading: legitimacy of the old order collapses first (or is declared to have collapsed); validity is then reset by fiat; efficacy of the new order is initially uncertain and must be established. INFERENCE
• Edit signature: not an edit within the version history but a new root — a discontinuity that the version graph cannot represent as a delta because there is no shared predecessor validity criterion. INFERENCE This is the single case that breaks the "linear/DAG of validated deltas" model.
• Reversibility: counter-revolution or restoration, itself a further rupture.
• Computational signature: forces the version model to admit multiple roots and discontinuous re-basing — you cannot assume a single unbroken validity chain back to a single origin. Any system assuming one Grundnorm-per-history is falsified by the historical record of ruptures. INFERENCE Cross-ref §09.10.1 (versioning), §09.9 (crisis→emergency loop).

09.2.7 Drift (class C, continuous, unlicensed)

FACT Slow, cumulative, mostly-unlegislated change in the operative content of norms as the surrounding practice, vocabulary, and background conditions shift. Most visible in social norms and constitutional conventions (Memo 06), but present everywhere as the gradual mismatch between a fixed text and a moving world. FACT that language and practice shift; [INFERENCE that this silently changes operative norm content.]

• Edit signature: many tiny, unrecorded interpretive shifts summing to large change — the integral of a small rate over long time. No single edit is attributable. INFERENCE
• Delay: the longest of all; drift is delay made visible.
• Computational signature: the hardest to represent because there is neither a diff (as in amendment) nor an authorized interpretive event (as in reinterpretation) to hook onto. Drift can only be detected by periodically re-measuring the gap between text and practice — i.e., it requires the efficacy-monitoring state of §09.2.5. INFERENCE

09.2.8 Transplant / borrowing (class C, usually)

FACT A norm or whole body of norms is imported from one system into another — legal transplants (a foreign code adopted wholesale), regulatory borrowing, doctrinal reception. FACT that transplants occur widely; the degree of "fit" or rejection is contested (§09.3) and associated with the transplant-debate literature.

• Edit signature: a bulk add of foreign norms, followed by a slow adaptation period in which the transplanted norms are re-interpreted (class B) to fit the host's other norms and background. INFERENCE
• Delay: the adoption is fast; the integration (or rejection) is slow.
• Conflict interaction: transplants routinely create conflicts with incumbent norms that the host's conflict-resolution machinery must then metabolize. INFERENCE
• Computational signature: cross-system norm import raises the provenance and namespace problems: the imported norm arrives with a foreign validity story that does not hold in the host. The host must re-anchor it (re-validate under local rules) or carry it as flagged foreign-origin. Cross-ref Memo 08 (international borrowing), §09.10.4 (migration). INFERENCE

09.2.9 Codification of custom (class C→A boundary)

FACT A norm that lived as efficacious, legitimate custom (Memo 06) is captured in authoritative text and thereby acquires formal validity — the informal is made formal. Historically a major source of legal content (customary law codified; commercial usage codified into commercial codes). FACT that this occurs; the exemplar association is with historical codification movements.

• Three-clocks reading: efficacy and legitimacy are already high; validity catches up. The reverse of desuetude. INFERENCE
• Edit signature: a promotion — a norm moves from the un-versioned social layer into the versioned legal layer, acquiring a canonical text it previously lacked. INFERENCE
• Risk: codification freezes a snapshot of something that was living and adaptive; the codified version can then diverge from the continuing custom (creating future drift/desuetude). Cross-ref Memo 06 (recognition "freezes" custom). INFERENCE
• Computational signature: the formalization event itself — the point where an un-compilable social norm becomes a compilable textual one. This is exactly the above-kernel→below-kernel transition the series cares about (Memo 12), viewed as a mutation. INFERENCE

09.2.10 Precedent overruling (class A-within-doctrine / class B hybrid)

FACT In precedent-based systems, a court with sufficient authority may overrule a prior precedent, replacing the rule it stood for. Distinguished from distinguishing (narrowing a precedent's scope without overruling) and from reversal (a higher court undoing a specific decision on appeal). FACT re common-law practice; cross-ref Memo 02.

• Edit signature: repeal+add on the precedent set, but with a distinctive property: the overruling is itself a precedent, so the change mechanism and the thing changed are the same kind of object. INFERENCE This self-application is what makes precedent the paradigm case for §09.4.
• Delay: event-driven and often very long (a precedent may stand for generations before a suitable case reaches a court willing to overrule).
• Retroactivity problem: overruling is typically declaratory — treated as if the law was "always" the new rule — which raises acute versioning problems (was conduct under the old precedent lawful?). Prospective overruling is the attempted fix. INFERENCE This is a direct challenge to naive versioning (§09.10.1).
• Computational signature: the version history of a precedent set is a DAG with self-modifying edges, and the retroactivity question means an edit can change the interpretation of past states, not just future ones. This breaks append-only versioning. INFERENCE

09.2.11 Schism / forking (class C, structural)

FACT A normative community splits into two communities, each carrying a divergent descendant of a once-common norm-set — religious schism (Memo 04), the split of a legal tradition, the forking of a standard or of an open-source governance charter. FACT that schisms occur across religious, political, and technical domains.

• Edit signature: a branch — one version history becomes two, sharing a common ancestor and diverging thereafter. The version graph is no longer a line or a single-rooted tree but genuinely branched, with no merge in the general case. INFERENCE
• Delay: the split can be sudden (a declared schism) or the culmination of long drift.
• Authority problem: after a fork each branch claims to be the legitimate continuation; there is no shared authority to adjudicate which fork is "the" system. INFERENCE Legitimacy becomes multi-valued.
• Computational signature: the version model must support branching without a canonical merge, and the identity question ("which branch is the system?") is not internally decidable — it is decided socially, above the kernel. This is the version-control analogy taken literally, and it is exact. Cross-ref Memo 04 (doctrinal schism), §09.10.1. INFERENCE

09.2.12 Market selection (class C, decentralized)

FACT In the economic family (Memo 05), norms/practices/standards change not by legislation but by differential survival: practices that yield advantage spread, those that don't are abandoned; standards win by adoption; contract terms that prove efficient become boilerplate. FACT that competitive selection of practices occurs; the efficiency of the outcome is contested (§09.3.2), associated with the Hayek/North threads.

• Edit signature: no authored edit; the norm-set changes as the population distribution over practices shifts. The "edit" is a change in relative frequency, not in a text. INFERENCE This is variation–selection–retention (§09.3.1) operating on norms directly.
• Delay: as fast as the market clears; the fastest mutation mechanism in the catalogue.
• Computational signature: requires representing a norm not as present/absent but as prevalent-to-what-degree — a distribution over a population — and updating by selection dynamics rather than by edit. This is continuous, statistical, and un-authored: the antithesis of the amendment floor case. Cross-ref Memo 05, Memo 06 (norm cascades). INFERENCE

09.2.13 Summary table of mechanisms

* "no diff" / "diff*": the textual diff, where one exists, does NOT capture the operative change (reinterpretation, declaratory overruling).

INFERENCE The single most compiler-relevant fact in this table is the DIFF? column: a large fraction of real normative change produces no textual diff, or a diff that misrepresents the change. A version-control model keyed on text diffs — the obvious first design — is therefore structurally blind to reinterpretation, desuetude, drift, and market selection. This is logged as the memo's primary warning to D1 and returned to at §09.10.

09.3 Competing theories: models of why and how norm-sets evolve

§09.2 catalogued what happens (the mechanisms). This section presents the three principal models offered to explain and predict norm change. Per §0.5 they are presented as live competitors without adjudication. The critical methodological task, under M1 (falsifiability) and M8 (theories are evidence, not authority), is to separate the testable core of each model from its metaphorical surplus. All three are, at minimum, useful lenses; the question is which parts are science and which are analogy.

09.3.1 Variation–Selection–Retention (the evolutionary model)

FACT There is a well-developed research tradition that models cultural and institutional change using the abstract Darwinian schema: a population of variants is generated (variation), some are differentially favored (selection), and the favored ones are retained/transmitted to the next period (retention/inheritance). Applied to norms: rules mutate (variation via drafting, error, reinterpretation, borrowing), some spread and some die (selection via enforcement cost, compliance, competitive advantage, adoption), and survivors are copied forward (retention via codification, teaching, precedent). FACT that this modeling tradition exists; [INFERENCE that norms exhibit all three sub-processes.]

The testable core (survives M1): if one can independently define the variation rate, the selection criterion, and the retention/copying fidelity, then the model makes falsifiable predictions about which variants spread and how fast. Where those three quantities are measurable (e.g., adoption of contract clauses, spread of a regulatory template across jurisdictions, diffusion of a constitutional provision), variation–selection–retention is a genuine, testable model. INFERENCE

Substrate-neutral: the schema does not care whether the "replicator" is a gene, a habit, or a statute. FACT re the abstract schema; its APPLICABILITY to norms is [HYPOTHESIS.]

Memetics — flagged strictly as HYPOTHESIS

HYPOTHESIS The strong form of the evolutionary model — memetics — posits that norms (and ideas generally) are replicators ("memes") that propagate between minds analogously to genes, with the associated apparatus of fidelity, fecundity, and longevity, and that cultural evolution is literally Darwinian selection over memes. Per the JD's explicit instruction and per M1, memetics is logged here only as hypothesis, and a weak one, for concrete reasons that the compiler team should record:

• There is no agreed unit of the meme (what exactly is the replicated entity? a rule? a phrasing? a practice?) — without a well-individuated replicator the theory resists measurement. INFERENCE — this is the standard objection.
• The copying-fidelity assumption is doubtful: norms are typically reconstructed by recipients (re-interpreted, adapted) rather than copied with high fidelity, which is more Lamarckian (inheritance of acquired modifications) than Darwinian. INFERENCE
• The theory is at risk of unfalsifiability: any spread can be redescribed post hoc as "high memetic fitness," which under M1 is a red flag. INFERENCE

INFERENCE The defensible residue for the compiler: the substrate-neutral variation–selection–retention schema is a legitimate modeling frame for norm dynamics; the meme-as-gene ontology is not needed and should not be built in. In edit-algebra terms, variation = the edit operators of §09.2; selection = a fitness function combining enforcement cost, compliance, and competitive advantage; retention = the versioning/copying machinery. We adopt the schema and decline the memetic ontology.

Falsification watch (M1): the evolutionary model is falsified for a given domain if norm change there is shown to be directed/teleological rather than variation-then-selection — i.e., if the "variation" is not blind but goal-aimed (as deliberate legislation arguably is). INFERENCE This is the single most serious objection: much normative variation is designed, not random, which makes the analogy to biological evolution partial at best. We therefore treat the model as one lens among several, strongest for un-authored change (market selection, drift, cascade) and weakest for authored change (amendment, legislation).

09.3.2 Path dependence and lock-in (the institutional-economics model)

FACT The tradition associated with institutional economics (the positions associated with Douglass North, Paul David, and W. Brian Arthur, cross-ref Memo 05) holds that institutional trajectories exhibit path dependence: outcomes depend on the sequence of prior states, small early events can have large durable consequences, and systems can become locked in to an arrangement that is stable but not necessarily efficient. The classic mechanism is increasing returns / self-reinforcement: the more a norm is adopted, the more costly it is to switch away (setup costs, learning effects, coordination effects, adaptive expectations). FACT that the model and its mechanisms are well-developed; the inefficiency claim in specific cases is contested.

(A POSITIVE feedback loop; see §09.7.)

Consequence: the reachable set of future norm-sets NARROWS over time. Early, cheap choices become late, expensive ones. "Lock-in" = the switching cost exceeds the efficiency gain of any reachable alternative. INFERENCE

INFERENCE Path dependence is, in control-theory terms (§09.7), a positive feedback loop on adoption — and this is why it belongs in this memo twice: as a model of evolution here, and as a feedback loop in §09.7. The compiler-relevant content is sharp and testable: the cost of a given mutation is not constant but increases with the incumbency of the norm being changed. Migration cost (§09.10.4) is therefore state-dependent and path-dependent, not a fixed property of the target norm-set.

Testable core (M1): path dependence predicts that (a) switching costs rise with adoption duration/breadth, and (b) systems can persist in demonstrably dominated arrangements. Both are measurable and have been observed. INFERENCE It is falsified for a domain if norm choices there are shown to be freely reversible at constant cost regardless of history — a condition the market-selection literature (§09.2.12) suggests holds only under strong assumptions.

Relation to the evolutionary model: path dependence constrains the selection step. Selection does not freely find the fittest norm; it finds the fittest norm reachable from the current state at acceptable cost. The two models are complementary, not rival, on this point. INFERENCE

09.3.3 Punctuated equilibrium (the tempo model)

FACT Borrowed into institutional and policy analysis from evolutionary biology (the tempo hypothesis associated with Eldredge and Gould) and developed in political science (the punctuated-equilibrium framework in policy studies), this model describes the tempo of change: long periods of stasis (stability, incremental adjustment) interrupted by short, intense punctuations (rapid, large-scale reconfiguration). FACT that the descriptive pattern is widely reported; cross-ref §09.1.3.

   PUNCTUATED EQUILIBRIUM (tempo of norm change)
   ---------------------------------------------
   change
   rate
     |                    ||                          ||
     |                    ||                          ||
     |____________________||__________________________||______ time
      long stasis      PUNCT.   long stasis        PUNCT.
      (incremental)    (rupture, crisis, constitutional
                        moment, cascade, doctrinal revolution)

   Stasis is maintained by NEGATIVE feedback (stabilizers: amendment
   friction, precedent, path dependence). Punctuation occurs when a
   stabilizer is overwhelmed and a POSITIVE feedback (cascade) takes
   over briefly, until a new equilibrium re-establishes negative
   feedback. [INFERENCE - this is the control-theory reading.]

INFERENCE The control-theory reading (developed in §09.7) is the memo's own synthesis and is the reason punctuated equilibrium, path dependence, and feedback analysis are the same subject: stasis = a regime dominated by negative (stabilizing) feedback; punctuation = a transient regime where a positive (amplifying) feedback loop dominates until saturation restores stability. The stabilizers are exactly the friction mechanisms (amendment thresholds, precedent, lock-in); the amplifiers are exactly the cascades (§09.7.6, Memo 06). This unifies the three models: variation–selection–retention supplies the micro-dynamics, path dependence supplies the stabilizing feedback that produces stasis, and punctuated equilibrium describes what the whole looks like from far away.

Testable core (M1): punctuated equilibrium predicts a specific statistical signature — the distribution of change magnitudes is leptokurtic (heavy-tailed: many tiny changes, few enormous ones, little in between), unlike the bell-shaped distribution a smooth-incrementalist model predicts. FACT that this leptokurtosis prediction is the standard falsifiable form of the policy-punctuation thesis. This is a genuinely falsifiable claim and has been tested in budget and policy data. INFERENCE that it is testable; we do not here assert the verdict.

09.3.4 How the three models divide the labor

INFERENCE They are not three answers to one question but three answers to three questions:

INFERENCE For D1/D2 the practical upshot is that no single dynamical model should be hard-coded. The architecture should be able to represent both slow incremental self-reinforcing change and rare discontinuous re-basing, because the historical record contains both and punctuated equilibrium says the same system produces both at different times.

09.4 The six-part analysis applied to precedent / tradition as a source

Per the conventions (JD Q2), each source discussed must receive the six-part analysis. This memo's designated source is precedent/tradition, chosen deliberately: precedent is the purest specimen of a source in which the norm-set is its own mutation mechanism. Where legislation separates the rule from the rule-changer, precedent fuses them — each application of the norm-set is simultaneously a potential edit of it. That self-application is why precedent belongs in the dynamics memo rather than only in Memo 02. We treat "precedent" (the common-law doctrine) and "tradition" (the broader social/religious analogue, cross-ref Memos 04, 06) as two grades of one underlying source-type: authority derived from accumulated past instances.

09.4.1 Origin

FACT Precedent as a legal source originates in the practice of deciding present cases by reference to how past like cases were decided — the doctrine of stare decisis ("stand by things decided"), central to common-law systems. FACT re common law; cross-ref Memo 02. Tradition, its generalization, originates in any community's practice of treating "how it has been done" as a reason for how it ought to be done — the transmission of a practice across generations as itself normative. FACT that traditions function this way across religious, professional, and social domains.

INFERENCE The structural origin common to both: a norm whose content is not promulgated by an authority in a canonical text but induced from a series of past instances. The norm is a generalization over cases, not a deduction from a statute. This is the inductive source par excellence, and it is why precedent/tradition sits opposite legislation on the taxonomy (Memo 10): legislation is norm-by-authorship, precedent is norm-by-accumulation.

09.4.2 Legitimacy

FACT The legitimacy story for precedent rests on several stated grounds in the doctrine: formal equality (like cases treated alike), predictability/reliance (subjects can plan because past decisions bind future ones), judicial humility/constraint (the present judge is bound by more than personal preference), and accumulated wisdom (the distilled experience of many past decisions outperforms any single fresh judgment). FACT that these are the standard justifications offered; cross-ref the Burkean position associated with tradition below.

INFERENCE For tradition, the legitimacy ground associated with the position of Edmund Burke is that inherited practice encodes tacit accumulated knowledge that no living reasoner could reconstruct from scratch, so deference to tradition is epistemic humility, not mere conservatism. This is structurally identical to the Hayekian argument for prices and spontaneous order (Memo 05): both claim that a distributed, evolved system carries more knowledge than any central designer. INFERENCE that the two arguments share a structure — this is a substantive cross-source observation.

INFERENCE Note the legitimacy/validity/efficacy separation (§0.6) is unusually tight here: a precedent's validity (is it binding?) depends on the deciding court's place in the hierarchy; its legitimacy depends on the equality/reliance/wisdom story; its efficacy depends on lower courts actually following it. Because the same institution (courts) governs all three, they mutate more in lockstep than for most sources — but not perfectly (a widely-criticized precedent retains validity while losing legitimacy, setting up overruling).

09.4.3 Enforcement mechanism

FACT Precedent is enforced through the court hierarchy: a lower court that ignores binding precedent is reversed on appeal. The enforcement of "follow precedent" is thus internal to the adjudication system itself — the sanction for non-compliance is the nullification of the deviating decision. FACT re appellate review; cross-ref Memo 02 rules of adjudication.

INFERENCE Tradition's enforcement is the diffuse social sanction of Memo 06 (disapproval, exclusion, loss of standing) plus, in religious traditions, the institutional sanction of the relevant authority (Memo 04). The key contrast: precedent has a dedicated appellate organ enforcing consistency; bare tradition does not, which is why traditions drift (§09.2.7) far more freely than precedent does.

09.4.4 Mutation mechanism

This is the heart of the matter and the reason precedent is in this memo. Precedent mutates through a distinctive, self-applied repertoire:

• FACT Distinguishing — a court confronts a binding precedent but holds the present case materially different, so the precedent does not control. The precedent is not changed, but its scope is narrowed. Over many cases, cumulative distinguishing can hollow out a precedent without ever overruling it. FACT re the technique.
• FACT Overruling (§09.2.10) — a court with authority declares the prior precedent no longer good law and substitutes a new rule.
• FACT Extending/analogizing — a precedent is applied to a new situation by analogy, broadening its scope.
• INFERENCE Drift by re-characterization — the ratio decidendi (the binding principle) of a case is itself a matter of later interpretation; successive courts re-describe what an old case "really held," slowly migrating its content (a class-B interpretive mutation, §09.2.3, applied to precedent).

KEY PROPERTY: every mutation of the precedent-set is PERFORMED BY an instance of the precedent-set (a judicial decision). The change mechanism is a member of the thing changed. This is self-reference, and it is exactly the Hart "rule of recognition" / Luhmann "autopoiesis" structure (Memo 02, Memo 06) seen dynamically. INFERENCE

INFERENCE Tradition mutates by the same operators minus the dedicated organ: practices are silently narrowed ("we no longer do X on occasion Y"), extended, or abandoned (desuetude), and the "meaning" of the tradition is continuously re-characterized by each generation that inherits it. The absence of an appellate organ means tradition has distinguishing and drift in abundance but no crisp overruling — traditions rarely die by decree; they fade or fork (schism, §09.2.11).

09.4.5 Hierarchy

FACT Precedent is embedded in a strict vertical hierarchy: decisions of higher courts bind lower courts (vertical stare decisis), and a court's treatment of its own past decisions (horizontal stare decisis) is governed by weaker, discretionary rules. The precedential weight of a decision is a function of the deciding court's rank, the size of the majority, the age of the decision, and how often it has since been followed or doubted. FACT re the weighting factors, though their exact operation varies by jurisdiction.

INFERENCE This yields a crucial computational property absent from statute: precedents are not simply valid/invalid but carry a graded, mutable weight — a scalar (or vector) of authority that itself changes over time as the precedent is followed (weight up) or doubted/distinguished (weight down). A precedent can decay in authority for decades before being formally overruled. This graded, decaying authority is the same decaying-efficacy state flagged for desuetude (§09.2.5) and is a strong argument for representing normative authority as a continuous, time-varying quantity, not a boolean. INFERENCE

09.4.6 Conflicts

FACT Precedent generates conflicts of several kinds: two precedents pointing to opposite outcomes for the present case (resolved by hierarchy, recency, or by distinguishing one away); a precedent conflicting with a statute (statute prevails, subject to constitutional limits — cross-ref Memo 02 hierarchy); and lines of precedent diverging across different courts (a "split" resolved, if at all, by a higher court). FACT re these conflict types and their general resolution.

INFERENCE The distinctive conflict dynamic: because precedent both is the norm-set and changes it, a conflict between precedents is simultaneously a substantive dispute and a mutation event — resolving the conflict (by choosing which line to follow) edits the precedent-set. Conflict resolution and mutation are the same operation here, unlike in legislation where a conflict is resolved by applying a stable meta-rule (lex posterior, lex superior) that is not itself thereby changed. INFERENCE This fusion of "resolve" and "edit" is the deepest structural peculiarity of precedent as a source.

09.4.7 Computational implications of precedent specifically

INFERENCE Collecting the above, precedent hands D1/D2 the following distinctive requirements, several of which no other source demands:

1. Norms as inductive generalizations over a case-set, not as canonical texts — the norm's content is derived from stored instances and changes when the instance-set changes. This is case-based reasoning, not rule application, and it is a fundamentally different compilation target. INFERENCE
2. Graded, time-varying authority weight per norm (§09.4.5), decaying and accreting continuously.
3. Self-modifying rule-set: the change operators are themselves members of the set (§09.4.4) — the system edits itself as it runs. Any static compile-once model is inadequate. INFERENCE
4. Fused conflict-resolution/mutation: resolving a precedent conflict is an edit (§09.4.6); the two cannot be separated in the data model.
5. Retroactive/declaratory change (§09.2.10): an overruling can alter the operative meaning of past states, breaking append-only versioning (§09.10.1).

INFERENCE These make precedent the hardest source in the series to reconcile with a clean versioned-text model, and simultaneously the most illuminating, because a system that can represent precedent's self-modifying, graded, retroactive dynamics can represent every other source's mutation as a restriction of it.

09.5 Interlude: the control-theory vocabulary for reading loops

Before drawing the loops (JD Q5), we fix the cybernetic vocabulary used to classify each one. This is imported from control theory and systems dynamics and applied to norm dynamics; the application is INFERENCE throughout, the definitions are FACT.

• Feedback loop. A cycle in the dependency graph: the output of a process eventually returns as an input to that same process. FACT re the definition.
• Sign of feedback.
• Negative feedback (stabilizing). The loop opposes the perturbation: a deviation triggers a response that pushes the system back toward its prior state. Negative feedback produces homeostasis, regulation, equilibrium. FACT Example intuition: a thermostat.
• Positive feedback (amplifying). The loop reinforces the perturbation: a deviation triggers a response that pushes further in the same direction. Positive feedback produces runaway, cascade, lock-in, collapse, or explosive growth until some nonlinearity saturates it. FACT Example intuition: a bank run.
• Loop gain. The multiplicative factor by which a perturbation is scaled per trip around the loop. |gain| < 1 → perturbations shrink (convergence); |gain| > 1 → perturbations grow (divergence); |gain| ≈ 1 → sustained oscillation or marginal stability. FACT re the control-theory meaning; assigning a numeric gain to a normative loop is [HYPOTHESIS — we assign only qualitative gain (low/near-unity/high) and a sign.]
• Delay (transport lag). The time it takes a perturbation to travel once around the loop. Delay is the single most dangerous parameter: a stabilizing (negative) loop with a long delay can overshoot and oscillate, because the correction arrives after conditions have changed. FACT re control theory — delayed negative feedback is the classic source of oscillation and instability. This is the key insight the memo imports for governance: a well-intentioned stabilizer with a long lag is an oscillator. INFERENCE
• Setpoint / reference. The target state the negative loop regulates toward. In governance, the "setpoint" is itself contested and mutable (unlike a thermostat's), which is a fundamental disanalogy flagged below. INFERENCE
• Saturation / nonlinearity. Real loops are not linear; positive feedback runs until it hits a limit (resource exhaustion, backlash, a stabilizer switching on). Punctuation (§09.3.3) is positive feedback running to saturation. INFERENCE

With DELAY T added to a NEGATIVE loop: correction for state(t) arrives at t+T, when state has moved on => overshoot => oscillation; if gain high enough, instability.

Three honest disanalogies (M3 hostile emergence; M8). The compiler team must not over-trust the control-theory frame. Three ways governance loops differ from engineered control loops, each INFERENCE:

1. The setpoint is endogenous and contested. A thermostat's target is set from outside the loop; a polity's "target state" is itself produced by the political loop it is supposed to regulate. There is no external reference. This means governance loops can regulate toward a moving, self-chosen target — closer to an adaptive/self-referential controller than a fixed-setpoint one. INFERENCE
2. Gains and delays are not stationary. They change as the system changes (a crisis shortens some delays and lengthens others; media technology changes loop speed — Memo 06). The system is non-stationary, so stability analysis is local-in-time at best. INFERENCE
3. The "signal" is partly interpretive. What counts as a "deviation" requiring correction is itself a normative judgment (Memo 06 injunctive component, Memo 02 interpretation), not an objective measurement. The sensor is inside the ideology, above the kernel. INFERENCE

INFERENCE With those caveats, the frame is still the most useful available for reasoning about whether a given normative arrangement settles, oscillates, or runs away. Each loop below is tagged with (sign, gain, delay) and a one-line stability verdict. The verdicts are HYPOTHESIS-grade engineering judgments, not measured results, and are labelled as such.

09.6 The catalogue of normative feedback loops (JD Q5)

This section discharges JD Q5. Each loop is drawn as an explicit ASCII cycle and analyzed as (sign, gain, delay, stability verdict). The loops are grouped: §09.6 primary constitutional/social loops; §09.7 self-reinforcing and stabilizing structural loops; §09.8 organizational/economic loops; §09.9 the crisis loop and its pathologies. All diagrams are FACT as to the existence of the named couplings; the control-theory classifications are INFERENCE/HYPOTHESIS as marked.

A reading convention for every diagram:

LOOP SIGN = product of edge signs: even number of [-] => positive/amplifying loop; odd number of [-] => negative/stabilizing loop. FACT re sign algebra

09.6.1 The electoral loop: Election → Government → Law → Society → Election

• Sign: designed as negative/stabilizing INFERENCE. The intended function is regulation: if government output diverges from society's preferences, the next election corrects it (throw the rascals out). Odd number of opposing links in the intended design → negative loop.
• Gain: near-unity and variable HYPOTHESIS. Electoral systems differ in how strongly a preference shift translates into a government change (proportional vs majoritarian — Memo 01). High gain = responsive but jumpy; low gain = stable but unresponsive.
• Delay: long and quantized — one electoral cycle (years). This is the critical parameter. INFERENCE
• Stability verdict HYPOTHESIS: a stabilizing loop with a long, quantized delay is prone to overshoot and oscillation — the well-known "policy pendulum" where governments over-correct their predecessors because the correction signal (the election) arrives years after the grievance formed and operates in coarse on/off steps. The Society→Election link also carries a positive sub-loop (law shapes preferences that then re-elect the same government — see §09.6.2), which can flip the loop's effective sign. INFERENCE
• Compiler note: the SOCIETY→ELECTION back-edge is exactly the one Memo 06 insisted the dependency graph must include; it makes the "politics→law" chain a cycle, not a DAG.

09.6.2 The preference-formation loop (endogenous preferences): Law → Society → Preferences → Law

• Sign: positive/amplifying INFERENCE. Law shapes what people regard as normal (habituation, expressive function of law); the changed preferences then demand more of the same law. Even number of opposing links → positive loop.
• Gain: variable; high where law strongly shapes preferences (the "law as teacher" effect associated with the expressive theory of law), low where preferences are law-resistant. HYPOTHESIS
• Delay: generational — preference change is one of the slowest clocks (§09.1.5).
• Stability verdict HYPOTHESIS: amplifying with very long delay → slow lock-in of a normative regime (this is path dependence, §09.3.2, viewed as a loop). It is the mechanism by which a legal order becomes self-legitimating over generations, and equally the mechanism by which entrenched orders resist reform. It can destabilize the electoral loop (§09.6.1) by making "society" a product of "government" rather than an independent check. INFERENCE — this is the Foucault/normalization concern of Memo 06 seen dynamically.

09.6.3 The precedent loop: Precedent → Case → Ruling → Precedent

• Sign: predominantly negative/stabilizing INFERENCE. Stare decisis is a homeostat: rulings mostly confirm and reinforce existing precedent, damping deviation and producing doctrinal stability. But it contains an embedded positive sub-loop: a novel ruling that is followed becomes stronger precedent that attracts more followers (§09.4.5 weight accretion). So the loop is mostly stabilizing with occasional amplifying excursions (overruling/new doctrine) — a textbook punctuated-equilibrium generator. INFERENCE
• Gain: normally < 1 (deviations get distinguished away and damped); transiently > 1 during a doctrinal shift when a new line of cases self-reinforces. HYPOTHESIS
• Delay: event-driven and long — the loop only turns when a suitable case arises and is litigated to a court with authority (§09.2.10). Irregular sampling.
• Stability verdict HYPOTHESIS: normally strongly stable (this is precedent's whole point — predictability); the danger is the retroactivity discontinuity (overruling changes past states, §09.2.10) and the possibility of a self-reinforcing doctrinal cascade escaping the damping. This is the self-referential loop of §09.4.4 drawn as a cycle.

09.6.4 The religion–ethics–culture–politics–law loop

• Sign: mixed, net positive/amplifying over the long run HYPOTHESIS. Each link mostly transmits and reinforces a value-orientation around the cycle; a society's dominant value-complex tends to reproduce itself across religion→ethics→culture→politics→law and back. Where law suppresses or reforms the source religion, that link is negative and the loop can become stabilizing or even reverse.
• Gain: high in tightly-integrated value communities (theocracy, Memo 01/04), low in pluralist ones where the links are weak and contested. HYPOTHESIS
• Delay: the longest loop in the catalogue — multi-generational to secular. Deep culture and doctrine are the slowest clocks (§09.1.5).
• Stability verdict HYPOTHESIS: so slow and high-gain that it behaves less like a regulator and more like a civilizational attractor: it defines the equilibrium the faster loops oscillate around. Its practical significance for the compiler is that it produces the background value parameterization that Memo 06 flagged as above-kernel and non-compilable — the loop that generates the priors, not the rules. INFERENCE

09.6.5 The moral-progress / reform loop: Norm → Moral Critique → Reform → Norm

• Sign: negative/stabilizing relative to a moral setpoint INFERENCE — but note the setpoint is the moral ideal, not the status quo, so this loop drives the norm-set toward the moral reference and holds it there, which looks like directional "progress" rather than homeostasis. This is the clearest example of disanalogy #1 (§09.5): the setpoint (the moral ideal) is itself contested and mutable.
• Gain: highly variable; depends on the moral movement's mobilizing capacity vs the incumbent norm's lock-in (§09.3.2). HYPOTHESIS
• Delay: long (a reform generation) and event-gated by political opportunity.
• Stability verdict HYPOTHESIS: stabilizing toward a moving target; because the target (moral consensus) also drifts (§09.1.5), the loop chases a moving setpoint and never fully settles. Cross-ref Memo 03 (is-ought, moral realism debate — whether the "setpoint" is objective is exactly the unresolved question that determines whether this loop converges).

09.7 Structural loops: enforcement homeostats, lock-in amplifiers, legitimacy

09.7.1 The compliance homeostat: Norm-violation → Sanction → Compliance → Norm

• Sign: negative/stabilizing INFERENCE — the canonical regulator. Deviation triggers a sanction that suppresses deviation. This is the enforcement mechanism of essentially every source (Memo 02 courts, Memo 05 market loss, Memo 06 social sanction, Memo 07 organizational discipline) rendered as a control loop.
• Gain: the deterrent gain — how much a unit of detected violation reduces future violation. Depends on sanction severity × probability of detection (the standard deterrence product). HYPOTHESIS
• Delay: detection lag + adjudication lag + enforcement lag. Short in tight social groups (Memo 06), very long in formal legal systems (investigation→trial→appeal).
• Stability verdict HYPOTHESIS: stable when gain is adequate and delay short. Two pathologies, both control-theoretic: (a) long delay → oscillation: sanctions arriving long after the violation cause enforcement cycles (crackdown/relapse); (b) inadequate gain → desuetude: if detection×severity is too low, efficacy decays and the norm dies unenforced (§09.2.5). Desuetude is thus loop failure — the homeostat's gain fell below the threshold needed to hold the setpoint. INFERENCE — this connects desuetude to the loop framework precisely.

09.7.2 The lock-in amplifier: Adoption → Switching-cost → Adoption (path dependence)

• Sign: positive/amplifying FACT re increasing-returns mechanism; §09.3.2. This is path dependence as a loop.
• Gain: > 1 in the increasing-returns regime (that is what "increasing returns" means).
• Delay: moderate; each round of adoption raises the switching cost for the next.
• Stability verdict HYPOTHESIS: destabilizing in the sense of driving to a corner — it does not oscillate; it runs to a locked-in fixed point and stays there, possibly a suboptimal one. It is the stabilizer of the equilibrium and the obstacle to reform: it is why the moral-reform loop (§09.6.5) faces rising resistance and why migration cost is state-dependent (§09.10.4). Positive feedback here produces stasis, not runaway — because it saturates at total adoption. This is the microfoundation of punctuated-equilibrium stasis (§09.3.3). INFERENCE

09.7.3 The legitimacy loop: Compliance → Legitimacy → Compliance

• Sign: positive/amplifying INFERENCE. Compliance breeds perceived legitimacy (descriptive→injunctive slippage, Memo 06); legitimacy breeds compliance. A self-fulfilling loop.
• Gain: high; this is a strong loop, which is why legitimacy is "sticky" in both directions.
• Delay: medium-long (legitimacy is a slow state variable, §09.0).
• Stability verdict HYPOTHESIS: bistable — the hallmark of strong positive feedback. It has two stable attractors: a high basin (widely obeyed → seen as legitimate → widely obeyed) and a low basin (widely flouted → seen as illegitimate → widely flouted → desuetude/collapse). Small perturbations near the tipping point flip the system between basins — this is the mechanism of legitimacy cascades and sudden regime collapse (the crowd discovers "no one believes in this anymore"). Directly relevant to revolution (§09.2.6) and to Memo 06 norm cascades. INFERENCE The compiler must not model legitimacy as a smooth linear variable; it is a bistable state with tipping points.

09.7.4 The constitutional-review loop: Legislation → Review → Constraint → Legislation

• Sign: negative/stabilizing INFERENCE — a homeostat that holds legislation within constitutional bounds (Memo 02 hierarchy, Memo 01). Deviation from the constitution triggers invalidation, pushing legislation back inside the permitted region.
• Gain: depends on how aggressively review is exercised and how easily the legislature can re-route around a strike-down. HYPOTHESIS
• Delay: long (litigation must reach the reviewing court).
• Stability verdict HYPOTHESIS: stabilizing, but with a slow-moving second loop underneath: the constitution's meaning itself drifts (§09.2.3, §09.2.7) and the composition of the reviewing court changes, so the "setpoint" this homeostat regulates toward is itself mutating — disanalogy #1 again. A stabilizer whose own reference drifts produces slow doctrinal migration masquerading as stability. INFERENCE

09.7.5 The bureaucratic-elaboration loop: Regulation → Complexity → More Regulation

• Sign: positive/amplifying HYPOTHESIS — the regulatory-complexity ratchet. Each rule creates edge cases and adaptive avoidance, prompting further rules. Cross-ref Memo 07.
• Gain: > 1 where regulated actors actively game rules (adversarial adaptation).
• Delay: the regulatory update cycle (§09.2.4).
• Stability verdict HYPOTHESIS: amplifying → monotone growth of regulatory volume (a ratchet with weak reverse — repeal is rarer than enactment). Saturates only against enforcement capacity or political backlash (deregulation punctuations). This is a positive loop that produces slow accretion rather than oscillation, and it is the dynamic analogue of the non-monotonic-update problem (§09.10.3): the norm-set only grows, and exceptions pile on exceptions. INFERENCE

09.7.6 The cascade loop (rapid norm change): Threshold → Defection → Threshold

• Sign: positive/amplifying INFERENCE; the Granovetter-style threshold cascade (cross-ref Memo 06).
• Gain: > 1 once the tipping threshold is crossed; ≈ 0 below it (nothing happens).
• Delay: short once triggered — this is the fastest structural loop, which is why punctuations are sudden.
• Stability verdict HYPOTHESIS: below threshold, dormant; above threshold, explosive to a new equilibrium. This is the punctuation in punctuated equilibrium (§09.3.3) and the mechanism by which the legitimacy loop (§09.7.3) flips basins. It is the positive feedback that briefly overwhelms the stabilizers during a rupture. INFERENCE

09.8 Economic and organizational loops

09.8.1 The market loop: Market → Price → Behavior → Market

• Sign: negative/stabilizing FACT re the standard price-adjustment account; cross-ref Memo 05. Price rises → demand falls / supply rises → price falls back. The paradigm decentralized homeostat and the fastest loop in the whole catalogue.
• Gain: governed by elasticities; near-unity in well-behaved markets.
• Delay: minutes to seasons depending on the good; the shortest delays in the series.
• Stability verdict HYPOTHESIS: normally strongly self-stabilizing — the reason markets need no legislator to generate norms (Memo 05). Pathologies from delay: production lags produce the classic cobweb oscillation (agricultural boom/bust); and a sign flip can occur when expectations dominate (price rise → expectation of further rise → more buying → further rise), turning the stabilizing loop into the amplifying bubble loop below. INFERENCE

09.8.2 The bubble loop (market positive-feedback pathology): Price → Expectation → Demand → Price

• Sign: positive/amplifying INFERENCE — the same market whose fundamental loop is stabilizing (§09.8.1) contains this embedded amplifier when the expectations channel dominates the substitution channel.
• Stability verdict HYPOTHESIS: runaway until saturation (a crash — the positive loop reverses sign violently). Same structure as the legitimacy cascade (§09.7.3): a bistable/tipping system. Illustrates the general lesson: the same coupling can be stabilizing or amplifying depending on which channel dominates, so loop sign is not a fixed property of an edge but of the operating regime. INFERENCE Critical warning for the compiler: you cannot label an edge's sign statically.

09.8.3 The organizational KPI/incentive loop: Target → Behavior → Metric → Target

• Sign: intended negative/stabilizing (adjust target toward performance), actual often positive/degenerate via Goodhart's law. INFERENCE "When a measure becomes a target, it ceases to be a good measure" (the effect associated with Goodhart/Campbell) is a feedback pathology: optimizing the proxy decouples it from the goal. Cross-ref Memo 07.
• Delay: the reporting cycle (quarterly, etc.).
• Stability verdict HYPOTHESIS: stable if the metric stays coupled to the real objective; degenerate if agents optimize the metric at the objective's expense (gaming). This is the organizational analogue of §09.7.5 (rules create gaming). The compiler-relevant lesson: a governance loop that regulates a proxy will be gamed unless the proxy is re-validated against the objective — an argument for keeping the objective (above kernel) distinct from the metric (below kernel). INFERENCE

09.8.4 The transplant-adaptation loop: Import → Misfit → Re-interpretation → Fit

• Sign: negative/stabilizing INFERENCE — friction drives adaptation that reduces friction, settling the transplant into the host. Or, if the misfit is too great, the transplant is rejected (efficacy→0, desuetude), the loop failing to converge.
• Delay: long (the "integration" phase of §09.2.8).
• Stability verdict HYPOTHESIS: converges to fit or to rejection depending on the depth of incompatibility with the host's deep culture (§09.6.4). A stabilizing loop with a possible non-convergent (rejection) basin. Cross-ref Memo 08.

09.9 The crisis loop and its pathologies

09.9.1 Crisis → Emergency-law → Normalization → Constitution

• Sign: positive/amplifying (ratchet) in the pathological case HYPOTHESIS. The intended design is a negative loop — emergency powers are meant to be temporary and to return the system to baseline once the crisis passes (a homeostat around "normalcy"). The pathology, associated with the analysis of the state of exception (the position associated with Carl Schmitt on the sovereign exception, and separately with critical analyses of emergency powers), is that the return link fails: emergency measures normalize and the baseline ratchets, so each crisis leaves a permanently expanded set of extraordinary powers. Cross-ref Memo 01, Memo 11 (Schmitt as falsification thread).
• Gain: the ratchet gain — the fraction of each emergency's expanded powers that persists after the crisis. If 0, the loop is a clean stabilizer; if > 0, it is a ratchet; the historical concern is that it is reliably > 0. HYPOTHESIS
• Delay: crisis-fast on the way up (powers invoked quickly), slow-or-never on the way down (return to baseline lags or never completes). Asymmetric delay is the core pathology — fast to expand, slow to contract. INFERENCE
• Stability verdict HYPOTHESIS: an asymmetric-delay loop with incomplete return is a ratchet: it does not oscillate around a fixed baseline; it walks the baseline in one direction. This is the most dangerous loop-topology in the catalogue for a governance system, because it edits the validity criterion itself (the constitution) under conditions (emergency) designed to bypass the normal rules of change. It is revolution (§09.2.6) in slow motion and by legal means. The compiler-relevant flag: any loop with asymmetric expand/contract delays on a validity-altering edge is a structural hazard and must be detectable as such. INFERENCE

09.9.2 The backlash loop (self-limiting reform): Reform → Backlash → Counter-reform

• Sign: negative/stabilizing INFERENCE — a societal homeostat resisting rapid change; the faster/deeper the reform, the stronger the restoring force. Related to path-dependence lock-in (§09.7.2) as the active resistance complement to the passive switching-cost resistance.
• Delay: medium (mobilization time).
• Stability verdict HYPOTHESIS: damps reform; with long delay can produce reform/counter-reform oscillation (the policy pendulum again, §09.6.1). Explains why punctuations (§09.3.3) are often followed by partial regression to a new intermediate equilibrium rather than full adoption of the reform. INFERENCE

09.9.3 Synthesis: loops are coupled, nested, and non-stationary

INFERENCE The loops above do not run in isolation; they are wired together, and the system's behavior is a property of the coupled set, not of any single loop. Three structural facts the compiler team must carry:

(a) Nesting and timescale separation (stiffness). Fast loops (market §09.8.1, cascade §09.7.6) are nested inside slow loops (electoral §09.6.1, cultural §09.6.4). In control terms this is a stiff system — dynamics on wildly separated timescales — which is numerically and conceptually hard: the fast loop looks like instantaneous noise to the slow loop, and the slow loop looks like a fixed constraint to the fast loop, yet over long horizons the slow loop moves and invalidates the fast loop's assumed constants. INFERENCE A simulator that fixes the slow variables to analyze the fast loop will be right locally and wrong globally.

Each inner loop sees the outer as ~constant; each outer loop sees the inner as ~equilibrium/noise. True only locally in time.

(b) Sign is regime-dependent, not edge-intrinsic. The market loop (§09.8.1 stabilizing) and the bubble loop (§09.8.2 amplifying) are the same edges operating in different regimes; likewise the electoral loop's sign depends on whether the preference-formation sub-loop (§09.6.2) dominates. INFERENCE Therefore a static signed dependency graph is insufficient; the sign of a loop is a function of the operating point. This is one of the strongest reasons a compiler cannot precompute stability once and cache it.

(c) Non-stationarity. Gains and delays change over time (§09.5 disanalogy #2) — media technology shortens the cascade loop, judicial appointments change the review loop's setpoint, crises re-wire delays. INFERENCE Stability is a local-in-time property; a loop that is stable this decade may not be next decade with the same topology but different parameters.

INFERENCE Net: the normative system is a coupled, nested, non-stationary, regime-switching feedback network with stabilizers (negative loops: enforcement, market, review, backlash, precedent) and amplifiers (positive loops: lock-in, legitimacy, cascade, bubble, bureaucratic ratchet, crisis ratchet). Stasis is when stabilizers dominate; punctuation is when an amplifier transiently wins. This is the memo's unified dynamical picture and the single most important thing it hands downstream.

09.10 Computational implications of dynamics

This section translates the descriptive findings into requirements on D1/D2. All items are INFERENCE from the preceding sections unless marked otherwise; they are design constraints, not results.

09.10.1 Versioning

The mechanisms of §09.2 impose a versioning model far richer than a linear history of text diffs. Requirements, in increasing order of difficulty:

1. Time-indexed norm-sets with effective-dating. Every norm and every norm-set must be queryable as of a date ("what was the law on 2019-03-01?"). Enactment date, effective date, and repeal date are distinct and all three matter. This is the floor case (amendment, §09.2.1).
2. Typed version edges, not a bare timeline. Edges must carry types — amends, supersedes, derogates-from, interprets, overrules, branches-from (§09.2.2, §09.2.10, §09.2.11). A timestamp alone loses the semantic relation between versions.
3. Independent versioning of the interpretation layer. Because reinterpretation (§09.2.3) and drift (§09.2.7) change operative content with no textual diff, the mapping text+facts→outcome must be versioned separately from the text. A system that versions only text is blind to the largest class of real change (the DIFF? column, §09.2.13). INFERENCE — flagged as the primary versioning warning.
4. Multiple roots / re-basing. Revolution (§09.2.6) breaks the validity chain; the model must admit more than one root and discontinuous re-basing rather than assuming a single unbroken lineage to one origin.
5. Branching without canonical merge. Schism/forking (§09.2.11) produces genuine branches with no merge and no internal fact of the matter about which branch is "the" system. The identity decision is above-kernel (social). The version model is thus a DAG that may have multiple roots and unmergeable branches — closer to a distributed version-control graph than to a linear ledger.
6. Non-append-only history (the hard one). Declaratory overruling and retroactive reinterpretation (§09.2.10) can change the operative meaning of past states. This breaks the comfortable assumption that history is append-only/immutable. The model must distinguish what the record said at time t from what we now hold the law to have been at time t — i.e., a bitemporal model (valid-time × transaction-time) at minimum, and arguably a tri-temporal one adding "as-interpreted-time." INFERENCE — bitemporality is the known database technique; its sufficiency here is [HYPOTHESIS.]

   MINIMUM VERSION MODEL (composite of 09.2 requirements)
   ------------------------------------------------------
   NORM = { id, text_version*, interpretation_version*,
            authority_weight(t), effective_interval,
            provenance/source, validity_chain_root }
   EDGES(typed) = amends | supersedes | derogates | interprets |
                  overrules | branches-from | promotes(custom->law)
   GRAPH = multi-rooted DAG, branchable, NOT append-only
   TIME  = bitemporal minimum (valid-time x transaction-time),
           tri-temporal if interpretation-time is tracked

09.10.2 Migration

INFERENCE When the norm-set changes version, existing state (pending cases, ongoing obligations, prior acts, cached verdicts) must be migrated to the new version — exactly the schema-migration problem in software, here with legal semantics.

• Transitional rules are first-class. Real systems carry transitional/savings provisions ("acts done under the old rule remain valid", "pending cases proceed under the old law"). These are the migration scripts of the normative system and must be representable, not hard-coded. INFERENCE
• Migration cost is state-dependent and path-dependent (§09.3.2, §09.7.2): the cost of moving from norm-set A to norm-set B depends on how entrenched A is, not just on the size of the A→B diff. The reachable-set narrows over time. A migration planner cannot assume constant edit cost.
• Retroactivity is a migration policy choice, not a default: does the new rule apply to past acts (retroactive), only to future acts (prospective), or to pending-but-not-closed matters (a middle case)? Each is a different migration semantics and must be an explicit parameter, because the choice is contested and above-kernel. INFERENCE
• Transplant migration = namespace re-anchoring (§09.2.8): an imported norm arrives with foreign provenance and validity that do not hold locally; migration must re-anchor it under local rules or carry it flagged as foreign-origin.

09.10.3 Non-monotonic update

INFERENCE This is the deepest logical implication and connects to Memo 10's treatment of defeasibility. Normative change is non-monotonic: adding a norm can remove conclusions that previously held (a new exception defeats a prior obligation), and repeal can restore them. The consequence set is not monotone in the norm-set.

• Classical monotonic logic is inadequate for the update dynamics: in monotonic logic, adding premises only adds conclusions. Normative update routinely retracts conclusions (an amendment carves an exception; an overruling withdraws a rule; desuetude silently withdraws efficacy). A representation must support retraction and defeasible/default reasoning (the family associated with non-monotonic logics, default logic, defeasible deontic logic — cross-ref Memo 10). INFERENCE
• Belief-revision discipline. Updating a norm-set is structurally the AGM-style belief-revision problem (the framework associated with Alchourrón, Gärdenfors, and Makinson — notably Alchourrón also worked on the logic of legal change): on receiving a new norm that conflicts with the current set, one must revise (accommodate the new norm while giving up the minimum necessary) rather than merely expand. The minimal-mutilation and priority/entrenchment ordering questions from belief revision map directly onto legal derogation and hierarchy. INFERENCE — this correspondence is one of the memo's concrete handoffs.
• The bureaucratic ratchet (§09.7.5) is monotone growth of the text but non-monotone in conclusions: piling exception on exception grows the norm-set while shrinking and complicating what actually holds. The compiler must compute the current defeasible consequence set, which is not simply the union of all un-repealed norms.

Required: defeasible reasoning + belief-revision update, NOT set-union over valid norms.

09.10.4 Loop stability and oscillation

INFERENCE The feedback analysis (§09.5–09.9) imposes requirements that go beyond representing static norms — they concern representing and reasoning about the dynamics the norm-set participates in.

• Represent loops as first-class objects. The dependency graph handed to D1 (Memo 12, JD Q4) must carry its back-edges (the loops) explicitly, each annotated with (sign, qualitative gain, delay-order). A DAG-only representation is unfaithful (§09.0, Memo 06). INFERENCE
• Sign is regime-dependent (§09.9.3b). An edge's sign cannot be a static attribute; it depends on the operating regime (market vs bubble, electoral vs preference-locked). The representation must allow conditional edge signs and therefore regime detection. INFERENCE This alone defeats any "compute stability once" approach.
• Delay is the primary hazard. Delayed negative feedback oscillates (§09.5). Any governance mechanism D2 executes that closes a stabilizing loop with a long lag (a sanction regime, a review process, an electoral correction) risks inducing oscillation. D2 should expose loop delays so that oscillation risk is analyzable before deployment. INFERENCE
• Detect ratchets and bistability. Asymmetric-delay validity-altering loops (crisis ratchet §09.9.1) and bistable legitimacy loops (§09.7.3) are structural hazards; the ability to detect these topologies in a proposed governance design is a concrete, buildable analysis (it is graph-structural, not value-laden). INFERENCE — flagged as a research opportunity, §09.12.
• Non-stationarity forecloses permanent guarantees. Because gains/delays drift (§09.9.3c), any stability certificate is local-in-time and must carry an expiry / re-check obligation. D2 must not treat a stability proof as permanent. INFERENCE
• Simulation, not closed-form. Given nonlinearity, regime-switching, and stiffness (§09.9.3a), analytic stability results will be rare; the realistic tool is simulation of the coupled loop network under perturbation, with the honest caveat (M1) that the model's gains/delays are estimates. HYPOTHESIS that simulation is the right tool; it is a design bet, not a result.

INFERENCE The through-line of §09.10: dynamics turns the norm-set from a database into a versioned, bitemporal, non-monotonically-updated, self-modifying object embedded in a regime-switching feedback network. Each adjective corresponds to a mechanism (§09.2) or a loop family (§09.6–09.9). None is optional; each is forced by the evidence from the source memos.

09.11 Open questions

Per §0.5, problems this memo cannot resolve and which D1/D2 must treat as unresolved.

1. OPEN Is normative variation ever "blind"? The evolutionary model (§09.3.1) requires variation independent of selection. But much normative variation is designed toward anticipated selection (legislators draft with the courts in mind). If variation is systematically directed, the Darwinian analogy is at best partial and the model's predictive core is weakened. We cannot settle how much normative variation is blind vs directed, and the mix likely differs by source (high blindness in drift/market; low in legislation).

2. OPEN Can loop sign and gain be measured, or only assigned? Every gain/sign classification in §09.6–09.9 that is not purely structural is HYPOTHESIS-grade. Whether the gains and delays of normative feedback loops can be empirically estimated with enough reliability to support quantitative stability analysis — or whether they remain qualitative judgments — is unresolved. If only qualitative, D2's "stability analysis" is closer to hazard-spotting than to control design.

3. OPEN What is the correct temporal model? §09.10.1 argues for at least bitemporality and possibly tri-temporality (adding interpretation-time). Whether even tri-temporal modeling captures declaratory overruling and drift faithfully — or whether operative meaning is genuinely not reconstructable "as of a past date" because past interpreters are gone — is open. This is the temporal face of the Luhmann worry (Memo 06/11): perhaps "what the law was" is not a well-defined function of stored state.

4. OPEN Where is the boundary between mutation-the-system-licenses and mutation-that-happens-to-it? The A/B/C partition (§09.1.2) is clean in the paradigm cases and blurry at the edges: aggressive reinterpretation shades into unlicensed amendment; sustained non-enforcement (class C desuetude) can be tacitly licensed. Whether a crisp validity-preserving/validity-breaking line exists, or whether it is inherently contested (a Schmitt-flavored worry, Memo 11), is unresolved.

5. OPEN Does a self-modifying rule-set (precedent, §09.4) have a stable fixed-point semantics? A system whose change operators are members of the system it changes raises a self-reference question (cross-ref Gödel/Church/Turing threads, Memo 05/11). Whether "the current content of the precedent-set" is always well-defined, or whether it can be genuinely indeterminate/undecidable at a given moment, is open.

6. OPEN Can the "setpoint" of a stabilizing normative loop be represented at all without importing ideology? Every negative loop regulates toward a reference (constitutional baseline, moral ideal, market-clearing, "normalcy"). Each reference is value-laden and above-kernel (§09.5 disanalogy #1, Memo 12). Whether the loop's stabilizing structure can be represented content-neutrally while the setpoint is carried as an exogenous flagged input — or whether the two are inseparable — is the version of the kernel-boundary question that dynamics raises.

09.12 Research opportunities

Per §0.5, where D1/D2 work could contribute or must decide.

1. Build the typed, bitemporal version graph (§09.10.1) as the D1 substrate. This is concrete, buildable, and content-neutral. The typed-edge set (amends/supersedes/derogates/interprets/overrules/branches/promotes) is a finite, source-agnostic vocabulary derived directly from §09.2 and is a strong candidate interface element for Memo 12.

2. Formalize the mutation edit-algebra (§09.1.1). The claim that all mechanisms reduce to a small operator set (add/repeal/modify/re-scope/re-rank/re-interpret) wrapped in institutional procedures is testable: attempt to express every §09.2 mechanism as (primitive edit + procedural wrapper + delay + authority-check). Success would give D1 a compact, source-neutral change model; failure would localize exactly which mechanism resists reduction.

3. Adopt a defeasible + belief-revision update engine (§09.10.3). The AGM/legal-change correspondence (Alchourrón et al.) is a ready-made, formal starting point for non-monotonic update with priority orderings that map onto lex superior/posterior/specialis (Memo 02). This is perhaps the single most tractable, high-value formal import in the memo.

4. Build a structural loop-hazard analyzer (§09.10.4). Detecting dangerous loop topologies — asymmetric-delay validity-altering loops (crisis ratchet), bistable legitimacy loops, delayed-negative oscillators, positive lock-in loops — is a graph-structural computation that does not require resolving contested values. It could serve as a "linter" for proposed governance designs. High value, clearly buildable, content-neutral.

5. Represent authority/efficacy as continuous decaying state (§09.4.5, §09.7.1). Modeling precedent weight and norm efficacy as time-varying quantities (with decay when un-followed/un-enforced) would let the system predict desuetude and detect dead letters instead of confidently enforcing them. Requires an input feed of enforcement/compliance observations — flag the data dependency.

6. Simulate the coupled loop network (§09.10.4). A systems-dynamics simulator over the annotated dependency graph (with regime-switching edge signs) would let D2 explore oscillation, ratchet, and cascade risk before deployment — with the M1 caveat that outputs are only as good as the estimated gains/delays. Frame as decision-support/hazard-exploration, not prediction.

7. Specify transitional-rule (migration) semantics as first-class (§09.10.2). Making savings/transitional provisions and the retroactive/prospective choice explicit, parameterized objects — rather than ad hoc text — is a bounded, high-value deliverable that directly serves migration.

09.13 Summary of the four standing categories

ESTABLISHED CONSENSUS (09.1)

• norm-sets change via a small recurring edit repertoire
• change is governed by second-order rules where well-behaved
• change is uneven in time (stasis + bursts)
• history constrains the reachable future (path dependence)
• sources mutate on characteristic, widely-separated timescales

COMPETING THEORIES (09.3)

• variation-selection-retention (testable frame; memetic ontology rejected as weak HYPOTHESIS)
• path dependence / lock-in (= positive feedback on adoption)
• punctuated equilibrium (testable via leptokurtosis)
• the three are complementary, dividing micro/constraint/tempo

OPEN QUESTIONS (09.11)

• blind vs directed variation; measurability of loop gains; correct temporal model; the licensed/unlicensed boundary; fixed-point semantics of self-modifying rule-sets; representability of contested setpoints

RESEARCH OPPORTUNITIES (09.12)

• typed bitemporal version graph; mutation edit-algebra; defeasible + belief-revision update; structural loop-hazard analyzer; continuous decaying authority/efficacy state; coupled-loop simulator; first-class migration semantics

09.14 Handoff to D1/D2

Concrete computational implications for the compiler architects. All INFERENCE from this memo unless noted; each cites its section.

1. Model every norm-set as a trajectory, not a value (§09.0). The mutation operators are part of the system's identity. A compile-once static snapshot discards what distinguishes a living order from a museum exhibit.

2. Adopt the A/B/C mutation partition as the top-level tag on every change event (§09.1.2): endogenous/licensed (validity preserved), interpretive (text preserved, content migrates), exogenous/unlicensed (validity broken/bypassed/outlived). This tells the machine whether a change can be validated against the system's own rules or must be recorded as a fact.

3. Do not key versioning on textual diffs (§09.2.13, §09.10.1). A large fraction of real change — reinterpretation, desuetude, drift, market selection — produces no textual diff or a misleading one. Version the interpretation layer independently of the text.

4. Implement the typed, multi-rooted, branchable, non-append-only version graph (§09.10.1): typed edges (amends/supersedes/derogates/interprets/overrules/branches/promotes), multiple roots (revolution), unmergeable branches (schism), and bitemporal-minimum time (valid-time × transaction-time; add interpretation-time if feasible) to handle declaratory/retroactive change.

5. Carry efficacy and authority as continuous, decaying, time-varying state (§09.2.5, §09.4.5, §09.7.1) — not as the boolean "valid⇒live". Otherwise the system enforces dead letters (desuetude) and cannot represent graded precedent weight. Desuetude is loop failure (enforcement gain below setpoint-holding threshold), and should be detectable as such.

6. Use non-monotonic (defeasible) update with belief-revision discipline (§09.10.3), not set-union over valid norms. Adding a norm can retract prior conclusions; the AGM framework (Alchourrón/Gärdenfors/Makinson) and defeasible deontic logic are the ready formal imports, with lex superior/posterior/specialis as the priority/entrenchment ordering.

7. Make migration first-class (§09.10.2): transitional/savings provisions are migration scripts; the retroactive/prospective/pending choice is an explicit parameter (contested, above-kernel); transplant import requires namespace re-anchoring of foreign provenance; and migration cost is state- and path-dependent, not a function of diff size alone.

8. Represent the dependency graph WITH its back-edges (loops) as first-class (§09.10.4, JD Q4/Q5). Annotate each loop with (sign, qualitative gain, delay-order). A DAG-only graph is unfaithful (concurs with Memo 06's norm↔law back-edge).

9. Treat loop sign as regime-dependent, not edge-intrinsic (§09.9.3b): the same edges are stabilizing in one regime (market) and amplifying in another (bubble); the electoral loop's sign depends on the preference-formation sub-loop. Static signed graphs are insufficient; stability cannot be precomputed once and cached.

10. Treat delay as the primary instability hazard (§09.5, §09.10.4): delayed negative feedback oscillates. Expose loop delays so oscillation risk is analyzable before deployment. Build a structural loop-hazard analyzer (§09.12.4) that flags asymmetric-delay validity-altering loops (crisis ratchet, §09.9.1), bistable legitimacy loops (§09.7.3), and lock-in amplifiers (§09.7.2) — all detectable from graph structure without resolving contested values.

11. Attach expiry/re-check obligations to any stability claim (§09.9.3c, §09.10.4): gains and delays are non-stationary, so stability certificates are local-in-time. Never treat a stability proof as permanent.

12. Isolate the setpoint from the loop structure (§09.5, §09.11.6): a stabilizing loop's structure is candidate-content-neutral (buildable below the kernel), but its reference/setpoint (constitutional baseline, moral ideal, "normalcy") is value-laden and above-kernel. Carry setpoints as exogenous, flagged, provenance-bearing inputs; do not let the system manufacture them. Whether structure and setpoint are truly separable is OPEN and routed to Memo 12.

13. Log the dynamics-specific falsification threats (§09.11) into the Memo 11 risk register: self-modifying rule-sets may lack well-defined fixed-point semantics (Gödel/Church/Turing-adjacent); "what the law was as of date t" may not be a well-defined function of stored state (Luhmann-adjacent); the licensed/unlicensed mutation boundary may be inherently contested (Schmitt-adjacent). The architecture should degrade gracefully if any holds.

14. The safe residue from dynamics (§09.13): the typed edit-algebra, the typed version graph, and defeasible update with a priority ordering are the three dynamics constructs the kernel can rely on as content-neutral. Loop setpoints, loop gains, and the selection criterion of the evolutionary model all live above the kernel and enter as flagged inputs or estimates.

End of Memo 09. Cross-references: Memo 01 (political generation of law, emergency powers, policy pendulum), Memo 02 (rules of change, precedent, open texture, conflict/hierarchy rules as priority orderings), Memo 03 (moral-reform setpoint, is-ought), Memo 04 (doctrinal schism/forking, revelation and slow doctrinal drift), Memo 05 (path dependence, market selection, transaction cost of enforcement, Hayek/North, Arrow), Memo 06 (drift, desuetude, cascades, norm↔law back-edge, Luhmann autopoiesis, Foucault normalization), Memo 07 (regulatory update, KPI/Goodhart loop, policy versioning), Memo 08 (transplant/borrowing, international norm life cycle), Memo 10 (defeasibility, non-monotonic logic, taxonomy of norm-by-precedent/evolution), Memo 11 (falsification: self-reference, Schmitt on the exception, computability limits), Memo 12 (dependency graph with loops, kernel/above-kernel boundary for setpoints, invariant interface).