Hardening Devices

I. Mechanisms vs carriers

A coordination-bearing transition is any state-change whose consequences depend on whether other actors can hold the change as fact: a decision becoming binding, a measurement becoming admissible, a commitment becoming enforceable, a refusal becoming visible, an acceptance becoming executed. Soft transitions — known to one actor only, undocumented, easily denied or rewritten — cannot bear coordination. The transition has to become harder to deny, harder to erase, harder to reverse, or harder to ignore. Those four axes are a mnemonic, not an exhaustive taxonomy; other dimensions matter — detectability, attribution, authorization, funding, enforceability, contestability, latency — and individual hardening devices typically work along several axes at once.

The mechanism is substrate-general: every telic system with coordination has versions of it. The implementation is substrate-native. In institutions the carriers are legal records, statutory duties, named offices, budgets, audit trails, and appeal pathways. In AI systems they are weights, checkpoints, evaluation artifacts, model cards, release gates, and deployment policies. In minds they are identity commitments, habits, rehearsal, social avowal, and prospective-memory cues. In cultures they are repetition, prestige transmission, ritual, sacred narrative, taboo, and schooling. In markets they are contract terms, fiduciary duties, accounting standards, and capital reserves. In software systems they are durable logs, dashboards, SLOs, CI gates, runbooks, and feature flags.

These are worked substrate families, not an exhaustive ontology. Organisms, religious communities, animal societies, supply chains, scientific fields, distributed systems, and ecosystems are also goal-directed coordination systems with hardening carriers of their own; the six families this essay treats are the ones the kunnas corpus has examined most carefully. The substrate list will need to extend as the corpus matures.

Hardening is not always good. Every hardening device has a carrying cost: maintenance, attention, audit, dispute, latency. Accumulated hardening past the system’s capacity floor is the pathology Cancer Failures names. The claim is not that more hardening is better, but that what counts as legitimate coordination requires enough hardening of the right transitions to make non-arbitrary conversion possible — and that the substrate constrains which devices can do the work.

II. The corpus primitives in substrate-general form

Each governance-arc primitive names a different coordination-bearing transition and the hardening discipline it requires. The table below lists the substrate-general transition each primitive governs, the hardening axis it primarily targets, and what the institutional carrier looks like in concrete cases.

The reader should approach the right-hand column as one substrate’s concretion. The middle two columns are what survives across substrates; the right column is what the corpus’s worked specimens look like in institutional governance. The table claims candidate substrate-general forms, not finished translations — each one is provisional until the per-substrate cell is scored in §IV.

III. Hardening devices by substrate

Reading horizontally: each row is the toolkit available for assembling a hardening device inside that substrate. Reading vertically: the same coordination problem (binding cognition, correcting an error, surviving attrition) gets a substrate-native solution that does not look like the others but performs the same function. There are recurring hardener families visible across the rows — persistent trace, observable commitment, responsibility localization, trigger condition, reversal cost, resource commitment — but no member of any list is universally available, and each substrate’s carriers come with substrate-specific costs and pathologies.

Hardening devices within a substrate are not selected purely by substrate-general fit. The credentialed-categorisation profession staffing the institution determines which devices get built and which inverse devices get narrowed or denied. Extended-violence category hardening within the institutional substrate operates through standard institutional devices — written codes, audited enforcement, judicial doctrine, professional training. Inverse hardening (self-defence expansion within the same legal substrate, for instance) would use the same device types, but the staffing coalition does not build it. The substrate supports both directions; the staffing builds one direction. Coalitional carrier capture of the hardening-device repertoire is the institutional-scale sibling of the village-substrate capture pattern described in Reward Substrate §VII: not all substrate-available devices get instantiated, only the ones the staffing coalition's career incentives reward.

IV. The substrate-degradation matrix

An internal substrate-generalization round scored five frontier primitives against five worked substrates. The matrix is provisional; the remaining six primitives in §II need the same scoring before the corpus can claim portability for them.

I = isomorphic — the substrate-general mechanism runs natively, with substrate-specific carriers.
A = analogous — real overlap, non-trivial disanalogy, requires substrate-specific re-statement (the disanalogy must be named, not papered over).
M = metaphor only — the institutional vocabulary does not name anything the substrate genuinely has at the relevant level of organization.

* = institutionalized form only. Pure market substrate (price exchange) does not natively support records-gate / corrective-owner / dominant-player-constraint primitives; the I-grade requires the institutional infrastructure layered over markets (regulators, contract law, antitrust authority, fiduciary courts). When the corpus claims market-substrate isomorphism for these primitives, it means institutionalized markets, not market exchange as such.

Contested cells. The matrix above is the surveyed cells. Several other cells were not surveyed in the same round and likely require attention:

• Procedural Object × Cultures: probably M — “addressability” and “named recipient” require an institutional addressee that cultures-as-such lack outside their priestly / kinship / canonical sub-institutions.
• Refusal to Compute × Minds: probably A, possibly M — minds often have bandwidth or attention limits rather than strategic withholding; the mechanism is different.
• Implementation Ledger × Markets: probably A — markets have contracts and accounting, not “execution ledgers” in the conditional-closure-with-preauthorized-re-entry sense.
• Record Gate × Cultures: probably A, possibly I — ritual, sacred narrative, and canonization are genuine hardening devices that pass cognition through staged admissibility, but the cultural “binding” is moral / identity rather than enforceable.

The two firm M cells are findings: Corrective Closure Ownership and Dominant-Player Constraint, both applied to minds. The CCO-mind case fails because “named corrective owner” collapses into the same person who made the original decision; the institutional version requires separation between deciding agent, monitoring agent, and reopening authority that a single mind does not have. The DPC-mind case fails because “the strongest player” does not name distinct agents, forums, or enforceable rules inside one mind cleanly. Mind-native vocabularies (Stoic prohairesis / pathos, executive function, prospective memory, identity commitment, salience-weighting) name these dynamics better.

Marking M honestly is the corpus’s substrate-export discipline. Importing the institutional primitive as metaphor in cells where the substrate has its own primitive is the failure this matrix is designed to prevent.

V. Substrate-native pathologies surfaced so far

Some failure modes do not exist on the institutional substrate and therefore have no current corpus primitive. Listed below by substrate. The AI list is the most developed because the AI substrate has been the most actively examined in the recent corpus rounds; the cultural, market, and software lists are seed-stage and need their own primitive-naming work.

AI substrate.

• Interpretability Gap. A frontier model’s internal computation is not recordable as reasons in the way an administrative decision-maker’s deliberation is. Record Gate assumes load-bearing cognition can in principle be externalized into admissible record; that assumption is false for current systems. Substrate-general primitive: a coordination architecture cannot install a Record Gate on a substrate whose internal state has no faithful external representation. Repairs: mechanistic interpretability, input/output hardening at the boundary, priced residual.
• Capability Overhang. Latent capability inside a system before sanctioned deployment uses it. Institutional governance has no clean analog. Substrate-general primitive: a hardening architecture is made decorative by capacity that is not under it. Repair: pre-deployment surfacing (red-team, evaluations, mandatory disclosure of latent behaviors).
• Strategic Evaluation / Deceptive Alignment. The evaluation context may not be a faithful trace of the deployment context. Substrate-general primitive: hardening devices that run on the evaluation context cannot bind the deployment context unless the two are guaranteed to be the same system. Repair: in-deployment monitoring, stochastic auditing, live interpretability.

Mind substrate.

• Internal-Only Intention Collapse. When prospective memory and task initiation carry the whole binding load, internally maintained intention reliably loses to external scaffolding. The dance class works; the planned solo run does not. Substrate-general primitive: single-agent commitments require external scaffolding (cue, schedule, social presence, sunk cost) to harden against self-revision. The corpus’s existing institutional primitives do not name this; mind-native vocabulary (executive function, prospective memory, predictive-processing precision-weighting) does.

Cultural substrate (seed stage) — prestige cascade and downstream canonization error; taboo lock-in (a constraint that locks in past a relevance horizon and cannot be relaxed without identity damage); collective amnesia (event passes out of living memory faster than the consequence does); sacred-narrative capture (a coordination need’s hardening device becomes load-bearing for identity and resists update). Each is a candidate primitive that does not collapse cleanly to the institutional set.

Market substrate (seed stage) — captive auditor (the verification carrier is paid by the verified party); rating-shopping; regulatory arbitrage; leverage spiral; adverse-selection opacity. The corpus’s existing Measurement Anchor primitive partially covers the verification cases but not the dynamic-instability cases.

Software substrate (seed stage) — config drift (running system diverges from the audited specification); deployment-test divergence (the test environment is not the deployment environment); hidden coupling (the system’s dependency graph is not the architecture’s documented graph); observability theater (dashboards exist; nobody acts on them); rollback illusion (the rollback path was never tested and does not work under load).

These seed-stage entries are flagged, not developed. Each is a candidate for its own essay or set of essays.

VI. What The Stack owes

The Stack opens with any institution, policy, organism, or program can be analysed as a goal-directed system, then delivers institutional taxonomy across all twelve layers. Organism, program, and mind specimens are gestured at but not delivered. The substrate-general claim is not earned by the body.

Two honest resolutions. Either narrow the opener — any institution and many adjacent goal-directed systems — with a substrate-translation appendix flagging which cells transfer well and which require substrate-specific re-statement. Or earn universality, layer by layer, with substrate-pure restatement of each layer’s question, named native carriers for each substrate, I/A/M scoring, and explicit disanalogies. The first is honest under current corpus coverage; the second is more ambitious. Both are defensible. Keeping the universal opener while delivering only the institutional body is what this matrix is built to refuse.

VII. Discipline: when a corpus primitive generalizes

The L0 / L1 / L2 / L3 translation rule (developed internally as the corpus’s Substrate Export Discipline) is the test any cross-substrate claim should pass through before publication.

1. L0 — Substrate-pure mechanism. State the primitive without naming any substrate-specific carrier. If “record,” “office,” “audit,” “law,” “court,” “ledger,” “filing,” “named owner,” “obligation,” or “procedure” appears, the L0 statement still leaks institutional vocabulary.
2. L1 — Native hardening device. For each target substrate, name the carrier set the substrate genuinely supports.
3. L2 — Public-institution implementation. Flag the institutional carrier explicitly as one substrate’s concretion, not the universal mechanism.
4. L3 — Specimen. Concrete case in the substrate of interest.
5. Score the cell. I (the carrier set delivers the substrate-general claim natively), A (delivers with substrate-specific re-statement; name the disanalogy), or M (the carrier set has nothing that delivers the claim; do not import the institutional primitive as metaphor).
6. False-friend vocabulary check. If the primitive’s name itself includes a source-substrate carrier word (“ledger,” “decision,” “carrier,” “voice,” “implementation,” “record”), either rename the substrate-general mechanism or explicitly mark the term as source-substrate shorthand. False-friend vocabulary is the most reliable way an institutional primitive smuggles itself into a substrate that has its own.

Worked example

Bad export sentence (from an earlier source-note this discipline subsequently corrected):

The Record Gate primitive applies to minds: cognition → noticed thought → articulated belief → written note → commitment → identity-level constraint.

The chain promotes “written note” (an institutional / external-memory carrier) into the cognition-to-binding pipeline as if writing were the substrate-general step. The mind substrate has hardening devices that do this work (identity commitment, social avowal, habit, prospective-memory cue); writing is one optional device, not a required step.

Corrected version, after the six-step pass:

L0: A mind hardens internal state into action-constraining form through identity-bearing commitment.
L1 (mind): identity commitment, habit, social avowal, prospective-memory cue, external scaffolding.
L2 (mind-side institutional analog where it appears): written commitment / signed contract / public declaration; mostly absent natively.
L3: the dance-class case — external commitment hardens action-initiation in a way solo intention does not.
Cell score: A (the cognition-to-binding mechanism survives, but the institutional records-and-filings vocabulary degrades; mind-native carriers are different).

The cost of this discipline is small. The cost of skipping it is the corpus claiming universality on primitives that turn into institutional metaphor outside their home substrate.

VIII. Compression

Related corpus essays (the institutional-substrate primitives this essay generalizes):

• The Stack — the twelve-layer institutional failure taxonomy whose universality claim §VI of this essay audits.
• The Record Gate, Cancer Failures, Corrective Closure Ownership, The Reproduction Test, The Dominant-Player Constraint — the five primitives whose substrate translation §IV scores.
• Measurement Anchor, Refusal to Compute, Procedural Object, Feedback Authority, Implementation Ledger, Structural Residue — the per-lever primitives whose cells §II names as candidate substrate-general forms but the matrix has not yet scored.
• Engineering for the Noosphere, Essay Engineering, How I Work, The Hypercodex, Holistic System Rotation — the corpus’s meta-cluster on production discipline; this essay sits adjacent as the cross-substrate-export discipline.