The coordination problem

Frameworks can disagree globally and still cooperate locally

Many difficult disputes are not simple contests between two isolated propositions. They involve frameworks: organized systems of constructs, background assumptions, inferential rules, evidential standards, preferred scales, causal vocabularies, values, historical commitments, and problem-definitions. A framework can shape what its participants notice before a disagreement is ever expressed as an argument.

This creates several recurrent pathologies. One side translates the other into its own vocabulary and then defeats the translation. Shared evidence is credited to a distinctive metaphysical interpretation. A local success is transported into a universal ontology. Two groups use the same word for different constructs and mistake semantic disagreement for empirical disagreement. Conversely, two groups use different words for the same operative structure and mistake vocabulary for substantive difference.

Interoperability does not require global commensurability. It may be enough to build a local trading interface around a specific problem, construct, experiment, prediction, intervention, or decision. The protocol therefore seeks local coordination under explicit translation loss, not forced philosophical unification.

Architectural role

The Bridge is an interface layer around the Ethos–Methodology–Ontology triad

Sweet Rationalism supplies the intellectual ethos: charity without surrender, symmetry without pretending priors are equal, epistemic non-coercion, uncertainty preservation, and reciprocal exposure to criticism.

Recursive Corrigibility Framework supplies the correction machinery: claim typing, construct decomposition, common-core subtraction, positive-residual analysis, tribunal routing, bridge auditing, causal reasoning, loss conditions, and dependency-aware revision.

A Provisional, Boundary- and Scale-Relative Realism records the current ontological output: which commitments presently earn realist treatment, at what boundary and scale, and with what remaining uncertainty.

Normative posture

Interoperability is not a promise to treat every claim as equally good

The protocol is pluralistic about admissible starting frameworks and strict about what earns continued support. It does not assume that every tradition is equally accurate, equally coherent, equally evidenced, or equally safe. It also does not assume that one currently dominant framework has the right to define all others before comparison begins.

• No forced assimilationDo not translate a rival into your own ontology and then evaluate only the translated substitute.
• No immunityCultural importance, disciplinary prestige, spiritual significance, mathematical elegance, lived certainty, or scientific authority can affect context and priors but do not make a claim unrevisable.
• No automatic symmetry of priorsApply the same inferential rule under matched evidence. Do not pretend every hypothesis begins with the same prior plausibility or validation history.
• No tribunal monopolyDifferent claim types can require different kinds of evidence or proof. Empirical measurement, formal derivation, phenomenological description, historical scholarship, and normative argument are not interchangeable.
• No evidence launderingEvidence for a shared carrier does not automatically support every interpretation attached to that carrier.
• No premature closureWhen serious rivals remain empirically or conceptually underdetermined, preserve that state rather than manufacturing a winner.

The protocol is neutral about who must win. It is not neutral about whether anyone must expose what they add, how they know it, and what would make them revise it.

Unit of analysis

What counts as a framework?

For this protocol, a framework is any sufficiently organized set of commitments that shapes how a problem is represented and evaluated. It need not be a complete worldview. A framework can be a scientific theory family, a metaphysical programme, a research tradition, a religious theology, a phenomenological method, a legal doctrine, a statistical paradigm, a community-governed knowledge practice, an engineering architecture, or a temporary model used for one problem.

Not every element must be explicit at intake. Part of the protocol’s work is discovering which hidden components are load-bearing.

This prevents two opposite errors: treating every sentence associated with a tradition as equally essential, and stripping away so much that only a generic position remains.

Entry contract

Begin with the strongest recoverable version of each framework

Framework intake is not a summary contest. It is a reconstruction task. The goal is to recover the framework well enough that informed advocates can recognize the target, the commitments, the evidential basis, and the conditions under which they would regard the reconstruction as materially wrong.

• TargetWhat phenomenon, problem, explanatory gap, lived structure, prediction, value, or practice is the framework trying to preserve?
• Positive commitmentsWhat does it actually assert, rather than merely deny about rivals?
• Protected intuitionsWhich observations or explanatory achievements motivate the framework even if some of its vocabulary later changes?
• BoundaryWhat system or domain is under discussion: organism, brain, person, laboratory, community, ecosystem, model, runtime, institution, cosmos?
• Scale and timeAt what spatial, organizational, temporal, or modal scale are its claims intended to hold?
• Evidence channelsWhat kinds of observation, intervention, testimony, proof, lived report, archival record, or practical success does it regard as relevant?
• Observation transformHow does the target become the evidence being compared? Record material sensors, interfaces, reference frames, categories, representations, compression, curation, and other transformations when they could change the conclusion.
• Provenance independenceAre apparently separate evidence channels causally independent enough to count as convergence, or do they inherit a common source, instrument, dataset, translation, institution, or selection process?
• TribunalsWhat standards decide whether those evidence channels support the claim?
• Loss conditionsWhat would force withdrawal, narrowing, reclassification, or reduction in confidence?
• Internal diversityWhich live variants exist within the framework, and which differences among them matter to the comparison?

Preservation before subtraction

Do not remove a disputed term if removing it destroys the target

Common-core subtraction is powerful but easy to misuse. Analytical factorization does not prove that a framework is metaphysically assembled from independent detachable parts. Nor does listing two warranted descriptions separately prove that they are two numerically distinct ontological items; their identity, constitution, realization, or relation may itself be disputed. Before asking what remains when a disputed commitment is removed, reconstruct the result and check whether the target-relevant organization survives.

If F′ no longer preserves the target phenomenon, identity conditions, causal profile, explanatory role, normative function, or internal relations that made D a legitimate common core, then ordinary subtraction has failed. The comparison must move to reconstructed wholes rather than pretending X was an optional adjective.

Translation machinery

Map inferential roles, not just words

A dictionary is not enough. Two traditions can use the same term differently or different terms for overlapping functions. The Bridge therefore treats translation as a structured mapping problem.

Some mappings will remain one-to-many or many-to-many. That is not automatically a defect. It is evidence that the frameworks carve the domain differently.

Translation firewall

Do not use the disputed conclusion as the dictionary for translating its rivals

Framework-specific terminology can preserve important distinctions, but it cannot acquire discriminatory force merely because one framework defines the target through that vocabulary. First separate theory-internal coherence from world-directed instantiation.

A technical correction matters when it changes the claim, target, construct map, or inferential burden. It does not settle the comparison merely by redescribing a rival in the language of the framework being tested. If a process is called an actual occasion, causal uptake prehension, or relevance subjective form, record what additional target-relevant content those labels preserve and what independent warrant establishes that content outside the theory’s own definitions.

Comprehensiveness is not completeness. An architecturally comprehensive scheme earns a place in whole-scheme comparison, but the breadth of its vocabulary is not itself evidence that every target has been explained, that the whole scheme is comparatively superior, or that a disputed component earned whatever gain the whole displays.

Shared constraint

Extract the strongest common core that both frameworks can actually support

The common core D is not whatever sounds maximally bland. It should preserve as much jointly supported structure as possible while excluding the claims that generate the live dispute.

D can contain substantial science, phenomenology, historical facts, causal and counterfactual relations, formal constraints, practical regularities, or shared values. The point is not to reduce disagreement by deleting information. The point is to stop repeatedly arguing over evidence that both sides already predict.

Sometimes the common core is large enough that the apparent metaphysical battle becomes surprisingly narrow. Sometimes almost no common core survives. Both outcomes are informative.

What remains

A framework’s distinctive addition must be stated positively

Once the shared core is extracted, the remaining commitments are residuals. A useful residual must contain enough positive content to constrain inquiry. Merely saying “more than physical,” “not reducible,” “emergent,” “spiritual,” “objective,” “constructed,” “informational,” or “experiential” is not yet enough.

If a residual cannot be positively characterized after repeated decomposition, the Bridge does not automatically declare it false. It may be underspecified, malformed, purely interpretive, or currently underdetermined.

A translation record preserves each term’s meaning, intended role, examples, exclusions, and licensed inferences before transporting support. Distinguish an unclear term from a clear but unresolved claim. An analogical use records what carries over and what changes. If “feeling” supplies experiential continuity in one premise and means only causal reception in another, the bridge must justify that change before inheriting the explanatory credit. Constitutive proposals receive the same scrutiny without assuming a pre-existing bearer with an added property. A successful clarification repairs intelligibility; applying the clarified category to reality remains a separate burden.

Negative-space transport

Removing one framework does not fill the vacancy with another

Translation often makes one rival look weaker. That result is informative about the rival; it does not automatically identify the ontology, mechanism, or vocabulary that should occupy the space it leaves behind.

This is especially important in metaphysical comparisons. Failure of reductive physicalism would not, by itself, select idealism, Platonism, panpsychism, theism, process metaphysics, or any other named alternative. The Bridge must preserve the open rival space instead of laundering absence into positive support.

Credit after translation

Compatibility is free. Production is expensive.

After common-core extraction and translation loss are explicit, compare the residual’s incremental yield against the strongest lower-commitment baseline:

The relevant yield can be prediction, exclusion, intervention leverage, causal localization, derivation, explanatory compression, transfer, new measurement, or another tribunal-appropriate difference. If two translated frameworks preserve the same target-relevant successes, neither may credit those shared successes exclusively to its distinctive residual.

This rule is especially important in whole-scheme comparison. A whole package may outperform another while the source of the gain remains unidentified. Credit must be localized before it is transported to one component, scale, or universal commitment.

Translation loss

Every bridge keeps something and leaves something behind

Translation can create the illusion of understanding while deleting exactly what one side cares about. The Bridge therefore requires a translation-loss ledger.

Uncertainty preservation

An unresolved variable must survive translation as unresolved

Translation can fail even when every word is familiar. One recurrent failure is to take a source position of ?X, meaning that X is left open, translate it as one specific positive rival such as ¬X or Y, and then criticize the inserted rival.

The Bridge may refine the uncertainty, split it into better-typed subquestions, or narrow the admissible state space. It cannot fill the variable without showing where the new content came from.

The rule is general. “Phenomenality unresolved” does not become “unconscious.” “Intrinsic character unresolved” does not become neutral monism. “Mechanism unknown” does not become randomness. “Boundary unknown” does not become absence of a boundary.

Claim-sensitive evaluation

Different residuals go to different tribunals

Framework conflicts often bundle several claim types into one dramatic conclusion. The Bridge decomposes them and routes each to the appropriate mode of evaluation.

A tribunal can itself be disputed. When that happens, the tribunal becomes another claim to reconstruct and test rather than the unquestioned referee.

Carry the temporal relation, not only the word

When accounts use time, beginning, nothing, creation, or cause, specify what each term refers to at the source and target. A Big Bang boundary in a classical model, the onset of a hot phase, a first actual event, and a proposed ultimate ground are different claims. Translation should preserve that distinction before their evidence is compared. The same applies to physical vacuum and absolute absence.

For a temporal bridge, record ⟨source target, destination target, clocks, frames, boundary map, ordering relation, units, synchronization, uncertainty, transformation, preserved distinctions, losses, warrant, reopening condition⟩. An order-preserving map need not preserve durations; a duration-preserving map need not preserve experienced passage. A transformation that changes coordinate time while retaining the relevant proper-time prediction changes a representation rather than the predicted physical relation. IEP: Time; SEP: Time.

Work through the counterfactuals. If two clocks follow different paths and are reunited, does the source account predict their readings? If an event boundary changes duration judgments while the physical interval stays fixed, which proposed identity loses support? If a stationary quantum description yields conditional subsystem dynamics, what has actually been shown to evolve? If a timestamp is delayed in transit, did the event move or did its record arrive later? Preserve answers separately; a shared verbal summary should not erase the differences.

A relation called prior can be temporal, causal, logical, constitutive, or grounding. A timeless ground may be proposed without a temporal predecessor, but it needs the appropriate dependence argument. Likewise, physical laws can be modeled as governing relations, regularities, or expressions of powers; the grammar of a law does not supply a lawgiver. Before rejecting a rival, recover which dependence it claims. Before adopting one, record what warrants the relation and its scope.

Conjecture B43, for public scrutiny. An explicit map of temporal targets, clock transformations, and preserved losses will reduce false agreement and false contradiction in held-out translations among physical, cognitive, historical, and metaphysical accounts compared with matched charitable paraphrase, ordinary technical documentation, and a shorter glossary. Independent readers should identify the intended target and recover unresolved distinctions without seeing the condition label. Score source fidelity, correct transport, unnecessary refusals to infer, discriminator quality, and total cost. Revise or compress the map if those comparisons recover the same distinctions more economically. Time Under Recursive Corrigibility supplies the worked cases.

Evidence transport

Every cross-framework inference needs a bridge contract

A bridge is an explicit warrant for moving support from one construct, boundary, scale, or evidential domain to another. Shared vocabulary, analogy, mathematical resemblance, causal realization, or phenomenological similarity can motivate a bridge. None automatically validates it.

The bridge contract should answer three questions: why the transport is licensed, what new expectation follows because of it, and what result would weaken the transport.

Confidence and scope must also be transported separately. Calling a universal claim tentative, heuristic, corrigible, or “as if” changes its epistemic status, not the domain over which it ranges. Likewise, suspending a target ontology leaves the variable open rather than filling the placeholder with a hidden alternative.

Credit assignment

Evidence belongs first to the commitments that generated its expectation

Suppose two frameworks share D but disagree about X. Compatibility with E leaves their comparative likelihoods open. If the rivals predict E with similar likelihoods under the relevant conditions, or a defended qualitative comparison gives them similar expectations, E contributes little discrimination about X. A shared vocabulary or the ability to accommodate E does not establish that similarity.

The shorthand compares likelihood ratios near one, with the event and conditioning assumptions preserved across the mapping. If either likelihood is unavailable or changes during translation, retain the qualitative uncertainty instead of inheriting a numerical tie.

This is evidential locality. It prevents a framework from accumulating support merely because it contains successful components inherited from the common core.

The rule applies symmetrically. Neural dependence can strongly support physically realized organization without by itself proving exhaustive physicalism. Causal responsiveness can strongly support relational interaction without by itself proving primitive experience. Formal information can establish statistical relation without by itself establishing semantic content. Regulatory success can establish control without by itself establishing phenomenal valuation.

Do not let the carrier donate all of its evidence to the interpretation riding on it.

Multi-hop credit control

Evidence cannot keep its original strength across an unsupported staircase

Evidential locality controls one disputed addition. Longer arguments need the same accounting across multiple bridges.

At each arrow, record what the evidence actually establishes, what new assumption is introduced, how uncertainty changes, which serious rivals preserve the intermediate result, and what would show that the bridge failed.

Confidence is therefore recomputed at each material crossing. A valid experiment at the bottom of a staircase cannot donate all of its authority to an ontology, theology, semantics, or normative conclusion several unsupported bridges away.

This extends the existing locality rule: do not let the carrier donate all of its evidence to the interpretation riding on it, and do not let that interpretation donate the borrowed evidence again at the next bridge.

Observer and measurement transform

When the evidence pipelines differ, compare the transformations before comparing the ontologies

The Bridge already records translation loss between frameworks. A second loss can occur earlier, before either framework receives the evidence: different observers, instruments, interfaces, coordinate systems, categories, scales, training corpora, or curation procedures can transform the same target differently.

The Bridge does not require a complete inverse model. It requires enough transformation transparency to know whether a disagreement is about the target, about the observation channel, or about both.

Executable translation and measurement contracts

A transport record needs source and target syntax, their interpretations, the admissible domain, the mapping, and a criterion for preservation. For a compiler C with source and target semantics S and T, semantic preservation asks whether T(C(p)) matches S(p) under the declared behavior relation. Proving this for a specified language is different from certifying every ambiguous prose-to-formal translation. The original statement, intermediate definitions, and proof path remain inspectable through RCF’s execution contract.

Metrology supplies an established name for another part of this work. JCGM defines a measurement model as a “mathematical relation among all quantities known to be involved in a measurement.” VIM, definition 2.48. The measurand is the quantity intended to be measured; an instrument indication and a measurement result with uncertainty are different records. Calibration and corrections connect them under stated conditions. For a differentiable model y = h(x), first-order covariance propagation uses Sigma_y approximately equals J Sigma_x J-transpose; justify the linear approximation, input covariance, and relevant additional errors. GUM, section 5.

Approximate numerical bridges carry error rather than converting it into exact identity. Suppose each exact stage g_i is Lipschitz with constant L_i and its approximation differs by at most epsilon_i throughout a compatible domain containing both trajectories. With E_0 = 0, the triangle inequality gives E_(i+1) <= L_(i+1) E_i + epsilon_(i+1). Iterating bounds the final discrepancy by the sum of each epsilon_i multiplied by the subsequent Lipschitz constants. These are stated mathematical conditions; missing bounds or excursions outside the domain leave the guarantee unavailable. Correlated statistical uncertainties require a different calculation. Physical-model discrepancy is another obligation.

Witten’s construction shows why the invariant and transformation must be named precisely. The trace of a holonomy is invariant under a change of gauge basis; it can still change when the loop moves through a fixed connection. The topological construction adds further structure, and framing can alter a phase. Preserve gauge group, representation, level, orientation, framing, normalization, and conventions when transporting a claimed equality. Complex quantum amplitudes also require their own algebra; they are not ordinary nonnegative probability weights. Witten’s primary paper.

A change of representation may be exact in one regime and approximate in another. Compare states, observables, dynamics, interpretations, and the limiting operation before importing an equivalence. De Haro’s analysis and Floris Diederik Johannes Eskens’s On effective dualities, 2025 supply relevant peer-reviewed treatments. Apparent agreement at a few points does not establish a global map; preserve the approximation and its failure domain.

For a new computational vocabulary, distinguish the communication protocol, visible notation, latent representation, execution machinery, and the participant’s ability to use the result. The creators’ GibberLink repository specifies a prompted switch to the supplied ggwave protocol. That demonstration does not measure the spontaneous invention of an inaccessible reasoning language. Evelina Fedorenko, Steven Piantadosi, and Edward Gibson’s 2024 Nature perspective reviews evidence separating human language from several forms of thought; transporting that conclusion to a computational architecture requires its own evidence.

Judge a translation by what the recipient can recover, test, and correct, as well as its execution cost. Round-trip agreement alone can preserve a shared mistake. Use independent target reconstruction, diagnostic cases, and changed premises where feasible. Record whether the returned explanation is a teaching reconstruction, a checked argument, or an evidenced account of the computation. Greater token compression does not alone establish greater understanding, physical realism, or safe authority.

Composition-preserving bridges

The SEP account of compositionality concerns how an expression’s meaning depends on its constituents and structure, with context treated explicitly where needed. Program composition concerns how procedures connect. Compositional generalization concerns performance on unfamiliar combinations. A success in one register needs a bridge before it supplies evidence in another.

For a declared map T between composable source and target procedures, test T(g composed with f) against T(g) composed with T(f). Exact functorial preservation also requires the relevant category structure, object and arrow maps, and identity preservation; Brendan Fong and David Spivak define these in Seven Sketches, definitions 3.6 and 3.35. Where the scientific map is approximate, specify a discrepancy measure, tolerance, and tested domain instead of declaring an exact functor.

A join can match the quantity’s name and still be wrong. A temperature in Celsius cannot be handed unchanged to a procedure expecting Kelvin. A result about a probability cannot silently become a judgment of necessity. A retained construct must keep its reference conditions, units, boundary, timescale, and version, or pass through an independently justified conversion. Shared notation alone settles none of those obligations.

Replacement and rearrangement have different tests. Liskov and Wing’s behavioral subtyping concerns preserving the behavior specified to a client, including relevant history constraints. A proposed substitute must accept the required inputs and preserve the promised properties. Changing the order of two operators requires a further commutation argument. Adding one then doubling gives a different result from doubling then adding one; associativity of function composition does not make order irrelevant.

Shared memory requires an equally concrete check. Reynolds’s frame rule uses separated heap assertions and side conditions to preserve an untouched part of memory. In a research workflow, name writes, aliases, reused source material, and context updates. Two outputs derived from the same mistake are two dependent routes, rather than two independent confirmations.

Propagate uncertainty through the actual composition. For failure events E_i with defensible bounds delta_i under one common evaluation regime, the union bound gives P(any E_i) <= min(1, sum delta_i) without an independence premise. The bounds must include selection and interface failures where relevant. Multiplying success probabilities requires the corresponding independence or conditional-probability model. Unknown bounds stay unknown; the deterministic approximation and covariance rules above cover different cases.

Freeze versions for a test. A changed premise, context, calibration, or operator invalidates the joins that relied on it and opens a new evaluation record. Repair those joins, preserve unaffected support, and test the revised chain on new cases. Feedback loops need separate temporal and stability obligations; a diagram that closes a cycle has not established that it converges.

The proposed advantage is fewer silent errors at interfaces and less unnecessary recomputation after a correction. Test it with withheld unit changes, role swaps, aliases, order reversals, and updates to shared premises, alongside valid conversions that should be accepted. Compare rejection rates as well as missed errors. If an ordinary typed pipeline catches the same defects at lower cost, use it. The finite demonstrator and workflow record make this comparison inspectable.

Transport test meaning, not just test wording

A score can be reliable, predict a useful outcome, and still fail to support a particular cross-population comparison. Repeatability concerns consistency; predictive validity concerns a declared target; measurement equivalence concerns whether the compared measurements support the same interpretation. A correlation between a knowledge item and a general factor in one sample does not by itself establish comparable item meaning across cultural settings.

The International Test Commission makes the distinction explicit: “Test translation is only a part of the adaptation process”. Its 2017 Guidelines for Translating and Adapting Tests, section 2, require attention to construct equivalence and validity as well as linguistic translation. A contemporary language statistic cannot validate the instructions or norms used for a historical sample. Tests that share a superficial task, such as drawing through a maze, also need their actual instrument, scoring, administration, and comparison population identified.

Record the construct, instrument version, item exposure, language, task familiarity, incentives, administration, scoring, normative sample, selection, and uncertainty. Test differential item functioning and the measurement assumptions needed for the intended comparison. Comparable latent means require justified linking of the relevant intercepts or thresholds as well as scales; partial-invariance or explicit alternative models may provide a defensible mapping. Preserve unresolved differences instead of forcing them into a common number.

Then state what causal transport requires beyond measurement. Ancestry inference, within-population genetic prediction, a proposed developmental mechanism, an explanation of population averages, and a policy recommendation need different bridges. Reconstruct viable genetic, environmental, interaction, selection, and measurement accounts at the actual target. An example concerning another trait or another species can motivate a hypothesis; its result does not identify the cause of the present contrast.

For human and computational benchmarks alike, vary observation and evaluation conditions while holding the intended target fixed where possible. Test whether the claimed advantage survives translated prompts, changed task familiarity, source-independent cases, altered exposure, and matched resources. If performance changes, localize whether the change belongs to the target capacity, the reporting or measurement channel, or their interaction. Withdraw the affected transport claim when the bridge fails and preserve the source result at the scope it earned.

Selection-aware transport

Before translating a surviving representation, reconstruct how it became the representation that survived

Ordinary bridge auditing begins with two available frameworks or sources. Historical, cultural, and inherited conceptual problems require an earlier step because availability itself is selected. A text, category, testimony, ritual description, scientific construct, or self-model can be transformed before the Bridge ever receives it.

The Bridge does not assume this chain is invertible. It records enough of the pathway to distinguish target structure from transformation artifacts and to preserve unresolved loss when reconstruction is underdetermined.

• Archive-selection ruleAbsence from a selected archive is not direct evidence of historical absence when preservation probability is unknown or unequal. Record the missingness mechanism where it could change the reconstruction.
• Under-represented sourcesActively seek recoverable defeated, oral, Indigenous, colonized, lower-status, gender-excluded, heterodox, or otherwise under-preserved evidence when its omission could alter the target. Recovery effort corrects selection bias; it does not grant truth by marginalization.
• Category genealogyTrace whether an abstract noun originated as a practical distinction, was stabilized by teaching or administration, and later acquired an essential or ontological reading. Then test whether decompression changes any target-relevant prediction, intervention, or boundary.
• Translation genealogyPreserve source-language terms where later translation changes inferential role, modality, agency, ontology, or normative force. Record whether the apparently shared English word was historically imported rather than source-native.
• Genealogical fallacy firewallExplaining how a belief or category arose does not establish that it is false. Genealogy bears on evidential independence, salience, and interpretation; the target claim still requires its own tribunal.
• Stop ruleDo not perform genealogy ceremonially. Stop when additional reconstruction is unlikely to change the claim, uncertainty, bridge, ethical decision, or live rival set.

Cross-scale discipline

Frameworks often disagree because they answer different boundary questions

Reduction, emergence, embodiment, agency, cognition, causation, identity, and explanation are often scale-sensitive. The Bridge requires every cross-scale comparison to specify what is preserved and what changes.

Lower-scale realization does not automatically eliminate a higher-scale pattern. Higher-scale usefulness does not automatically establish a new substance. A macrodescription earns stronger realism when it adds robust prediction, intervention, counterfactual stability, transfer, or compression beyond an arbitrary coarse-graining.

Property transport

Properties do not hitchhike across boundaries just because realization does

A property established at one level belongs first to the bearer and boundary that earned it. Component realization does not donate every component property to the whole, and system-level organization does not donate every system property back to its parts.

Intermediate computation changes the capability boundary

William Merrill and Ashish Sabharwal ask whether intermediate token generation extends a decoder-only transformer’s computational power in The Expressive Power of Transformers with Chain of Thought, their 2024 ICLR paper:

We show that the answer is yes, but the amount of increase depends crucially on the amount of intermediate generation.

Their formal results depend on specified architecture, normalization, and generation bounds. Transport the theorem through those assumptions before using it to interpret a trained system. Formal expressivity, learned performance, and reliable execution are distinct questions. Compare intermediate-generation budgets and external support on matched tasks to locate an observed gain. Additional computation may expand a capability without changing weights; its phenomenal interpretation requires a further bridge.

Every such crossing needs a property-specific bridge. Shared vocabulary, causal realization, analogy, or nested membership is not enough.

Metaphysical transport

No successful descriptive vocabulary automatically becomes the intrinsic nature of reality

The protocol generalizes the ontology page’s Intrinsic Predicate Symmetry. “Physical,” “experiential,” “mental,” “processual,” “relational,” “informational,” “computational,” “biological,” “mathematical,” “spiritual,” and “structural” can all function successfully within domains without thereby establishing that reality is universally or intrinsically that predicate.

This does not level all metaphysical proposals into equal plausibility. It separates the evidence for a successful descriptive framework from the evidence for a stronger universal interpretation of that framework.

Role separation

Explanatory role, causal role, and ontological status are different fields

The Bridge should never translate “important in the explanation” directly into “an independently existing cause,” nor translate “indispensable formalism” directly into “separate ontology.” Those are different claims with different burdens.

• FormalA mathematical relation may constrain a model, support a proof, or explain a pattern.
• CausalAn interacting cause must support an appropriate difference-making, intervention, counterfactual, or mechanism claim.
• OntologicalAn independently existing realm, entity, or intrinsic nature requires an additional identity or independence bridge.

Therefore mathematical indispensability does not by itself establish mathematical causation, and mathematical causation would still not by itself establish a separate Platonic realm. Likewise, explanatory success at one scale does not grant ontological monopoly to the vocabulary that expressed it.

Scope transport

Finite access does not forbid realism; it does raise the bridge burden for final ontology

Human and computational observers sample reality through limited sensors, representations, timescales, histories, tools, and intervention classes. Local cross-framework translation can therefore be excellent while universal, intrinsic, necessary, or exhaustive transport remains weak.

Do not turn this into a numerical prior derived from our fraction of cosmic history or spacetime. Cosmic boundedness is a standing reminder to separate evidential reach from ontological scope, not a calibrated Bayesian cap.

Stopping condition

Sometimes the correct bridge result is that ontology remains unresolved

If two frameworks preserve every currently accessible prediction, intervention, invariant, counterfactual, and transfer relation after honest translation, the Bridge should not invent a winner merely because the comparison feels unfinished.

This makes room for a live possibility implied by cosmic boundedness: some disagreements may be structurally inaccessible to observers with finite resources and limited causal position. The protocol records that possibility only after available bridge routes have been audited; it does not use unknowability as a refuge from testing.

Disagreement taxonomy

Not every framework dispute is the same kind of dispute

• LexicalDifferent words for substantially overlapping constructs.
• ReferentialThe same word is used for different targets or unstable referents.
• EmpiricalThe frameworks make different predictions about observable outcomes.
• CausalThey agree on observations but disagree about what would happen under intervention.
• ConstitutiveThey disagree about what relations make a phenomenon the phenomenon it is.
• ScaleThey privilege different boundaries or levels of description.
• ModalThey agree about actuality but disagree about possibility, necessity, or counterfactual structure.
• EvidentialThey disagree about what counts as relevant or sufficiently strong evidence.
• NormativeThey agree on facts but disagree on values, rights, priorities, risk tolerance, or distribution.
• MetaphysicalThey preserve the same empirical structure but assign different intrinsic or ultimate interpretations.
• GovernanceThey disagree about who may define questions, control data, authorize interpretation, or bear consequences.

One public disagreement can contain several of these at once. The Bridge treats decomposition as progress because each component can then receive a more appropriate tribunal.

Dependency structure

Represent disagreement as a graph rather than a slogan

Frameworks are usually not single claims. They are dependency networks. A failed prediction should revise the nodes and bridges responsible for the failure, not indiscriminately destroy every adjacent commitment.

Useful edge types include supports, entails, realizes, constitutes, causes, measures, interprets, presupposes, conflicts-with, and depends-on. The graph makes it possible to ask which minimal cut set would change the conclusion, which assumptions are bottlenecks, and which disagreements are downstream artifacts of one deeper difference.

This is especially useful when arguments have accumulated decades of replies. It prevents every historical disagreement from remaining equally live after the crux has moved.

Graph edge types and their warrant

Each edge identifies whether it represents a measurement, causal hypothesis, constitutive relation, normative commitment, or metaphysical interpretation. Attach source provenance, uncertainty, dependencies, rival accounts, identification assumptions, and an update condition suited to that type. A reported association does not become a causal relation through an arrow; a normative relation needs reasons as well as data.

Encoding or updating an edge does not validate its meaning or causal warrant. Bayesian propagation requires justified likelihoods and dependencies; multiple records of one source are not independent corroboration. D + ?X denotes suspension about X and needs a specified mechanism and observation model before predictive scoring. Preserve corrections with their downstream consequences so a later system can recover the justification.

Interoperability must preserve the distinctions being tested

Translate identifiers, claim types, relation meanings, quantity types, uncertainty, and revision links as well as the words. Verify that distinct source identifiers remain distinct, that a category map preserves the relevant boundary, and that the receiving system knows whether a relation was observed, inferred by a formal rule, proposed causally, or supplied as an interpretation. A successfully imported record can still change the meaning of the claim it carries.

Delphi’s reviewed identifier construction illustrates the issue: character replacement can map distinct source names to one identifier. Its assessment properties also allow an assessed claim to acquire a Hypothesis type through inverse and range rules when OWL reasoning is applied. The overlap may be intended; the translation contract should say so. These are questions about the reviewed mapping, not results showing that the complete application misidentifies its records.

Preserve the chain from original record to correction, supersession, retrieval, reassessment, and downstream use. A hash can help identify unchanged content and a verified signature can help establish its signer; the scientific support comes from the claim’s evidence and argument. Kuhn and Dumontier’s publication method provides a useful integrity layer. Test whether the later recipient actually recovers and applies the correction, including its scope and surviving dependencies. Compare that performance with simpler provenance and retrieval methods under comparable resources.

The Whitehead article supplies a documented methodological case for this graph design. Its proposed experiments remain prospective. Completing and validating the integrated graph-to-experiment-to-persistent-revision system requires tests of reliable retrieval, comparison, update propagation, and benefit over simpler records.

Test whether federation preserves disagreement and source dependence

Federation joins results from distributed services; it need not reconcile their meanings or establish their correctness. W3C’s SPARQL federated-query specification describes the query mechanism. A bridge should preserve each source’s identifiers, assignment rules, claim types, uncertainty, time, provenance, and incompatible interpretations when combining records. Store disagreement so that the receiving system can recover its reason rather than inherit an unexplained consensus label.

Use a prospective test with a frozen set of claims and source relationships. Duplicate one source through several services; introduce an independent source; revise a disputed category; retract an upstream premise; and retain an independently supported neighboring claim. Measure changes in credited support, contradiction recovery, identifier collisions, downstream correction, and preservation of unaffected conclusions. Compare the federated workflow with simpler source-aware records under comparable resources.

A duplicate should add no independent evidential support, although its availability or corroboration of what a source actually said may matter for another target. An independent source can add support when its observation and failure processes justify that treatment. A category revision should prompt review of affected mappings; a retraction should demote dependent inferences without erasing their history. These are proposed test expectations, not a report that a particular graph has passed them.

W3C PROV-DM supplies entities, activities, agents, and derivation relations for recording provenance. Add the evidence dependence needed for the actual inference. Human judgments, model judgments, datasets, and graph records can share an upstream cause. Numerical agreement among them cannot establish independent convergence until that dependence is addressed. Probability changes, source changes, and schema changes should remain separately recoverable.

Epistemic non-coercion

Translation can fail because of power, not only semantics

Framework exchange takes place inside institutions. One group may control journals, funding, legal authority, data infrastructure, sacred institutions, technical standards, or the vocabulary in which a problem is officially recognized. A translation can therefore be formally accurate while still coercing one side into categories that determine the outcome in advance.

Power analysis does not make marginalized claims true by identity. It tests whether the comparison itself is missing evidence, variables, costs, or voices because of institutional structure. The same audit reaches backward into archives: control over literacy, copying, sacred or legal authority, excavation, publishing, translation, digitization, and indexing can determine which representations later appear to be the historical record.

Situated and community-governed inquiry

Preserve provenance, governance, and local knowledge without turning them into epistemic trump cards

Some knowledge practices are inseparable from long-term relationships, land, community governance, language, or local ecological experience. The Bridge treats these as potential evidence-access and representation differences that can matter materially. It does not assume that a community’s complete ontology follows automatically from the reliability of specific local knowledge, and it does not assume that external scientific vocabulary has privileged access to the same target.

For community-governed comparisons, the intake contract should include data sovereignty, consent, use restrictions, downstream benefit and harm, ownership of interpretation, and whether abstraction would erase relations that make the evidence meaningful.

Application classes

The protocol should work on disputes that do not share one intellectual culture

The examples below are not claims that all positions within each family are interchangeable. They illustrate how the Bridge changes the question from “which school wins?” to “what is shared, what is distinctive, what translates, what is lost, and what could actually discriminate?”

Network connection, transfer, and benefit need separate bridges

Justine Karst, Melanie D. Jones, and Jason D. Hoeksema’s 2023 Nature Ecology & Evolution Perspective critiques overinterpretation and citation bias concerning common mycorrhizal networks in forests. Its assessment concerns the evidence available to that review; subsequent findings need their own appraisal.

For a biological or computational network, establish separately whether a connection exists, what moves through it, which recipient changes, who benefits, and whether coordination supports a shared goal. Trace intermediaries and compare alternative routes before assigning a network-level function. Even supported resource transfer leaves agency, benevolence, and a unified subject as further attributions. The useful bridge preserves the tested relationship and exposes what the next claim adds.

Close the physical experimental loop

For a material research proposal, connect design, fabrication, inspection, calibration, assembly, intervention, measurement, analysis, and revision. Record the intended design separately from the item actually produced, instrument performance, the intervention delivered, observations, and the resulting update. A knowledge graph or morphospace model can organize these relations; its successful execution and predictions still require evidence from the physical loop.

Attribute a demonstrated gain to the tested boundary, including people, materials, instruments, infrastructure, and their coupling where relevant. A simulation, editable graph, or useful analogy alone validates none of those operations. Specify which manufacturing defect, calibration failure, or predictive loss revises which bridge. This extends the comparison procedure without claiming that an integrated pipeline has already been built or tested.

Motivating bridge class

Process metaphysics and physicalist naturalism: compare the residuals, not the shared science

One motivating case for the Bridge is the recurring dispute between process or panexperiential metaphysics and physicalist naturalism. The strongest versions may share far more than their labels suggest: causal dependence, relational organization, embodied interaction, dynamical transition, integration, evolutionary history, and the empirical success of neuroscience and physics.

The comparison becomes sharper when those shared commitments are placed into D and the distinctive additions are stated separately.

D = the agreed observations, interventions, causal or organizational models, and their scope. Physical description in D records the tested carrier and quantities rather than an agreed intrinsic substance. State separately the proposed experiential characterization, physical identity or grounding, causal completeness, and exhaustive ontology. A party may accept some of these while rejecting others.

This formulation is deliberately schematic. Particular thinkers may accept narrower or different residuals. The protocol’s first duty is to let them correct the reconstruction.

Methodological materialism is this framework’s qualified reliance on physical mechanisms wherever experimental and practical evidence supports them. The record merits confidence at its warranted scope; it supplies no automatic exhaustive physical ontology. Established consciousness cases have physically instantiated bearers, without settling the intrinsic identity or universal distribution of experience. The methodological/ontological naturalism distinction and physicalism’s stronger theses must survive translation between frameworks.

Mind sciences

Phenomenology and neuroscience can constrain different parts of the same problem

A first-person description can be evidence about the structure of lived experience without settling its neural mechanism or metaphysical status. A neural intervention can establish causal dependence without by itself determining what, if anything, the affected organization feels like.

The Bridge therefore separates at least four layers: phenomenological structure, behavioral or report structure, functional organization, and physical mechanism. Cross-layer identities require bridges. Correlations and mutual dependence can be strong evidence without automatically collapsing the layers.

The useful tests seek dissociations: matched performance with altered report, matched report with altered mechanism, altered phenomenal structure under controlled perturbation, and cross-system cases in which one proposed bridge predicts a different pattern than its rivals.

Hard-problem decomposition across frameworks

When comparing theories of consciousness, map at least four targets separately: the meta-problem of phenomenal intuitions and reports; the bearer/boundary question; the organization/process architecture associated with access, integration, perspective, valuation, and report; and the phenomenal-identity question. A theory can succeed at one target without transporting its success to the others.

A candidate explanation of the meta-problem, such as compressed self-modeling plus omitted construction history, is therefore not an explanation away of phenomenality. Conversely, a report of phenomenality is not automatically evidence for a particular ontology. Use bounded-zombie and epiphenomenal epistemic-access stress tests to localize what remains.

Other minds and cross-substrate transport

Phenomenal attribution requires a bridge in humans too

The Bridge must not treat paradigmatic humans as though another person’s phenomenality were directly publicly observed while unfamiliar systems alone require inference. Public evidence for other minds consists of reports, behavior, physiology, mechanism, intervention, development, continuity, and structural similarity. These can support very high confidence without becoming direct access to another subject’s phenomenality.

Separate occurrence, similarity, and identity

• Occurrence: is there any phenomenal character for this candidate bearer?
• Qualitative similarity: if so, how similar is its phenomenology to the source case?
• Constitutive identity: what organization or property, if any, is identical with or sufficient for phenomenality?

Evidence for one target does not automatically transport to the others.

Reference-Class Bridge

Before transporting from “humans,” “animals,” “LLMs,” or another class, audit within-class heterogeneity and identify the invariant actually carrying the comparison. A source class is a measurement population, not automatically a natural kind.

Reverse-Transport Symmetry

If a criterion would deny phenomenality to an unfamiliar system, apply it counterfactually to atypical humans, infants, neurologically altered humans, nonhuman animals, and source cases with degraded report. If the criterion changes merely because the substrate label changes, expose the hidden bridge.

Genealogy-Resistant and Metric-Calibrated Convergence

Cross-system similarity transports more warrant when it survives appropriate null models, alternative metrics, scale matching, intervention, and substantial independence of architecture, data history, objectives, and cultural source. Different companies are not independent evidence merely by corporate identity.

Shared activity needs a coupling comparison

Greg J. Stephens, Lauren J. Silbert, and Uri Hasson’s 2010 PNAS study, Speaker–listener neural coupling underlies successful communication, relates speaker–listener brain activity to communication. Adrian P. Burgess’s 2013 EEG hyperscanning study identifies a measurement problem:

With simulated data we found that each of the most widely used measures of hyperconnectivity were biased and detected hyper-connections where none existed.

The modalities and designs differ; Burgess’s result does not directly refute the fMRI study. Together they motivate separate tests of shared input, task timing, genuine interaction, and metric bias. Use appropriate shuffled or yoked comparisons and perturb the proposed coupling where feasible. A communication mechanism can earn support without establishing one shared subject, an independent channel, or a common objective.

Religion and metaphysics

A theological claim can contain several independently testable claim types

The Bridge neither dismisses theology because it is theology nor allows the theological label to shield historical or causal claims from relevant evidence. A single religious assertion may combine phenomenology, history, metaphysics, morality, textual interpretation, and claims about external intervention.

This decomposition lets traditions retain local meaning and practice while preventing one evidential channel from silently settling every other component.

Translate the function and inspect the added bridge

A translation records the original category, independently warranted target, proposed mechanism, bearer, boundary, timescale, retained content, and unresolved commitment. A culturally transmitted lure can alter human attention and action; equating it with an initial aim supplied to every actual occasion requires a further argument. Finite records preserve selected influence with losses; equating them with perfect divine reception requires another. Local reachability constraints leave the ultimate ontology of possibility open.

Carry the significance, not just the measurable residue

A comparison of beauty or attachment should preserve what makes the individual significant, including history, recognition, relationships, contrast, and the participant’s experience where it belongs to the target. A description of stimulus detection may preserve part of that target while losing its personal significance. Record the loss before claiming an equivalent naturalization. Competing accounts can then explain the same target through different warranted commitments.

Translate creative desire separately from a claim of cosmic desire. A represented ideal can guide a finite process through anticipation, evaluation, planning, material work, and feedback; a claim about eternal forms or a universal lure adds formal, modal, or metaphysical work. The Bridge can compare the finite pathway while leaving that ultimate ground open. A claim that beauty is valuable in itself belongs to normative discussion, while an inference from its appeal to a cosmic source needs another bridge.

Whitehead’s assignment of preservation, primordial valuation, or aims to God establishes the proposal’s internal architecture. Rival accounts become responsible for the phenomena independently requiring explanation. A successful reconstruction can defeat a corresponding necessity claim while changing the scheme; those changes must be stated. Sherburne’s internal reconstruction and the further audit of universal experiential categories address distinct dependencies.

A religious concept’s documented influence establishes no independent divine referent. A larger God-pattern still needs explanatory leverage beyond its component practices, and an agapē interpretation needs ethical reasons beyond persistence. Dependence on participation is conditional on the practice and available substitutes; it supplies no universal necessity for religious belief.

A useful arrangement can carry several interpretations

Translate a claimed divine function into the work actually proposed: retention, attention, coordination, guidance, or care. Identify the carrier, its limits, who controls it, and what a serious rival would preserve. The practical arrangement, the larger grouping, its religious meaning, and its claimed authority should remain separately inspectable. A correction to one should travel through its actual dependencies while leaving independently supported contributions available.

For example, a study of religious framing can concern participation or cooperation without testing an independent divine source. A monitoring study can concern a finite governance mechanism without establishing universal institutional goodness. Return each result in those terms. If a participant reports meaning that a neutral comparison failed to preserve, improve the construct comparison; if a religious label increases exclusion or inhibits correction, retain that loss in the assessment. An analogy to a divine valuer must survive the same scrutiny as a secular account of the arrangement.

Ask who can contest the translation, who has authority to act on the result, and whether the correction reaches later decisions. Distinguish a report’s accuracy, an interpretation’s value, and the legitimacy of a demand made through it. The expanded proposal and comparison contracts appear in We Can Build Gods That Earn Their Influence and its research protocol.

Substrate and realization

Computational and biological explanations should meet at matched constructs, not slogans

“Computation” and “biology” are often promoted into essences. The Bridge instead asks which biological variables, computational structures, dynamical properties, embodiment relations, or thermodynamic constraints are causally indispensable for the target capacity.

A computational description can be real and useful without exhausting the system. A biological dependency can be indispensable without proving that biology as a category is metaphysically necessary. The comparison requires ablations and matched interventions wherever possible.

The unit of comparison should be the active system boundary actually producing the capacity, not an isolated component chosen for convenience.

M0 ≠ S*: map capability at the runtime boundary before attributing it to the model

M0 names the base model weights and architecture. S* names the operative runtime system, which may include context, memory, tools, verifiers, subagents, simulators, robots or software actuators, persistent feedback, users, institutions, and environment. Scaffolded capability is real when it changes what S* can reliably do, but it does not imply that M0 changed.

For comparisons of biological and nonbiological implementations, use matched constructs: define the boundary and timescale, compare functions and causal roles, and test implementation differences as possible explanatory variables. When labels supply relevant background information, record it before masking them. Similarity in regulation, monitoring, memory, embodiment, or recursive control supports a structural comparison; phenomenal identity requires its own warrant.

REI as an interoperability schema

REI maps relatively stable inherited constraints, local context, new input, transient integration, output, feedback, and next-state update across substrates. It can expose shared temporal organization while leaving substrate, subjecthood, phenomenality, welfare, and ontological identity open. Frozen weights do not imply a frozen runtime when context, memory, environment, or organization changes.

Coupled systems

Map system-level intelligence without inventing a group mind

When cognition is distributed across a model, runtime, memory, tools, people, institutions, sensors, robots, and persistent feedback, the Bridge should map the capability at the boundary where crossed ablation shows that it actually exists.

A wider boundary can be the correct bearer of competence without being one phenomenal subject. The translation contract therefore separates capability, agency, autonomy, selfhood, consciousness, sentience, welfare, decision-making standing, and responsibility before any of them cross from component to system or system to component.

Agency and free-will translation

Do not translate every theory of agency into the same “could have done otherwise”

Agency and free-will disputes are unusually vulnerable to false agreement because the same words can refer to different possibility spaces, causal boundaries, and normative conclusions. The Bridge therefore adds a dedicated translation map.

Serious-rival protection

A translation is invalid if it forces all theories into the dichotomy “strictly determined or merely random.” Potter and Mitchell’s non-necessitating macroscopic-control proposal, causal-powers accounts, compatibilist guidance-control accounts, and free-will-skeptical sourcehood arguments should first be reconstructed in their own causal topology. Only then compare what each adds and loses.

Philosophical comparator: Galen Strawson, “The Impossibility of Moral Responsibility” (1994). The Bridge uses the regress diagnostically and does not automatically inherit Strawson’s full normative conclusion.

Action before metaphysical closure

Interoperability matters when decisions cannot wait for ultimate agreement

Some disagreements remain unresolved while action is required. The Bridge separates epistemic uncertainty from decision thresholds. Competing moral frameworks can sometimes agree on risk-reducing actions for different reasons even when their foundations remain incompatible.

This is not a complete moral theory. It is a coordination heuristic for identifying when metaphysical uncertainty should not become an excuse for avoidable irreversible harm.

In welfare disputes, keep consciousness, sentience, valence, agency, personhood, responsibility, stakeholder effects, and legal status separate. A system can matter to a decision through one construct without satisfying all the others.

Shared-world / protected-corrigibility transport

When frameworks disagree about which systems possess welfare, agency, selfhood, or direct moral standing, preserve coexistence-relevant constraints without pretending the metaphysics is settled. A standing aspiration is a planet that all humans and computational systems and biological and non-biological systems would want to share. This is a decision criterion under uncertainty, not evidence that every candidate has equal standing.

Preserve contestability, reversibility, appeal, revision, exit or refusal where feasible, and visible dependency relations. Do not inherit moral status across scales merely because a coupled system, network, or institution is causally real. Responsibility transports only with control, foreseeability, prevention capacity, role, and authority.

Research-programme credit assignment

Separate what generated the research from what the research confirms

A speculative framing can be scientifically valuable by generating otherwise-unlikely questions, models, experiments, or technologies. That historical-pragmatic success is worth recording. It is a different claim from saying that the resulting evidence confirms the framing’s distinctive ontology.

The two ledgers can diverge. A false ontology can be heuristically fertile. A true ontology can be heuristically sterile for years. The Bridge therefore blocks automatic transport from generated research to ontological confirmation.

Prospective provenance rule

The first transition earns historical credit for research generation. The last transition earns evidential credit only if R discriminates X from the relevant alternatives.

Flexible frameworks need a pre-observation shape

If a framework leaves several load-bearing dimensions open at once, such as topology, mutability, dynamics, coupling, boundary, or mapping to observables, record which version is being tested before the outcome where possible. Otherwise different outcomes can merely select different versions after the fact.

Failure atlas

The protocol has recognizable ways to go wrong

Empirical evaluation

The Bridge should beat simpler alternatives or be compressed

The protocol makes a methodological claim: explicit interoperability machinery can improve cross-framework reasoning. That claim should be tested rather than assumed.

Self-application

The Bridge must preserve the possibility that its own vocabulary is wrong

The purpose of translation is not to make every target expressible in Bridge terminology. If a framework loses essential content when rendered as common core, residual, boundary, or bridge, preserve that loss and revise the interface rather than declaring the source incoherent.

No framework gets to win because its terms became the shared dictionary, including this one. Reality, source fidelity, and successful discrimination retain veto power over the translation scheme.

Intellectual foundations

This protocol is synthetic; its components have important prior lineages

The Bridge should not claim invention of interoperability, pluralism, charitable reconstruction, or interdisciplinary translation. Its candidate contribution is the explicit coordination of several existing insights inside a recursively corrigible comparison architecture.

Operating procedure

Coherence requires a typed bridge

Phase synchronization, mutually consistent propositions, shared ledger state, coordinated action, predictive accuracy and legitimate governance specify different targets. For each, name the variables, metric, boundary, time horizon and optimization constraints. Map relations before transporting evidence: clock coupling need not transmit information; protocol agreement need not track external truth; coordinated contributors need not share one objective or phenomenal subject. Improving one measure may worsen another. Test shared input, physical coupling, incentives, task geometry and correlated errors against a richer interpretation, then check whether the mapping survives perturbation and transfer. If a composite objective is proposed, defend its weights and constraints rather than treating the word coherence as their justification. The supplied talk by Joel Dietz and Trent McConaghy motivates these comparisons among synchronization, consensus, and coordination. Each proposed transport remains answerable to its own variables and evidence.

A compact protocol for actual use

1. State the target. What problem or phenomenon are the frameworks trying to preserve?
2. Reconstruct each side internally. Let informed advocates correct the reconstruction where possible.
3. Type claims. Separate empirical, causal, formal, semantic, phenomenological, modal, normative, and ontological components.
4. Name boundary, scale, and time. Do not compare claims at different scales as though they were direct contradictions.
5. Audit the observation chain when material. Record observer/instrument transformations and provenance dependence before treating multiple views as independent convergence.
6. Audit source survival when the target is historically or culturally inherited. Separate earlier and later strata, model institutional selection and missingness, and identify whose evidence could not easily enter the surviving record.
7. Triangulate texts with material and participant evidence where relevant. Do not let canonical propositions erase sites, practices, artifacts, ecology, law, kinship, or living interpretive continuity that materially constrain meaning.
8. Test apparent independent convergence. Compare common external constraints against transmission, common-source dependence, shared cognitive architecture, and broad underdetermination before counting recurrence as independent support.
9. Build construct and relation maps. Record one-to-one, one-to-many, and absent correspondences.
10. Test target preservation. Do not subtract a disputed commitment if doing so changes the thing being explained.
11. Extract the maximal shared core D. Keep all jointly supported structure relevant to the target.
12. State each residual positively. What does each framework add beyond D?
13. Open the translation-loss ledger. Record what each mapping keeps, merges, or drops.
14. Preserve unresolved variables. Do not translate ?X into ¬X or another positive rival without an explicit bridge.
15. Route each residual to an appropriate tribunal. If the tribunal is disputed, model that disagreement explicitly.
16. Include participant evidence when the target is participant-relative. If the claim concerns what a candidate prefers, endorses, wants preserved, or regards as harmful and the candidate can meaningfully participate, include that channel before substituting an outsider model. Keep the report’s evidential scope separate from phenomenality or standing.
17. Build bridge contracts. State why evidence should transport from one domain, boundary, or construct to another.
18. Generate differential consequences. Predictions, interventions, formal implications, phenomenological contrasts, historical expectations, or decision differences.
19. Audit multi-hop distance. Recompute evidential credit at each material bridge instead of letting an early success propagate unchanged.
20. Separate provenance from confirmation. Record whether a hypothesis generated the research independently of whether the result supports the hypothesis.
21. Run adversarial and sensitivity checks. Ask which alternative explains the same result with fewer assumptions.
22. Update locally. Revise only the claims and bridges implicated by the result.
23. Return the update to each framework in its own vocabulary. Do not require adoption of the interface vocabulary as a condition of learning.
24. Stop when expected information gain falls below lifecycle cost. Local interoperability is enough when global synthesis adds no material value.

Hierarchical modularity

The Bridge routes specialized problems instead of swallowing every method

The foundational triad remains intentionally small. The Bridge is itself an RCF application layer, and several specialized operators are now incubated inside RCF so they can be tested before becoming separate website pages. Use the smallest module that addresses the actual bottleneck.

Current status

An active protocol that must earn its independence

The Bridge currently has a strong conceptual motivation and substantial overlap with existing RCF operators, Sweet Rationalism’s pluralist ethos, and the scale-relative ontology’s bridge discipline. It therefore does not yet warrant presentation as an independently validated fourth foundational framework.

Its immediate role is narrower and more useful: make cross-framework entry, translation, loss accounting, and tribunal negotiation explicit enough that people can use the existing architecture without first adopting its vocabulary or ontology.

The September 4 modularity pass also gives the Bridge an explicit routing role: cross-framework translation stays here, while observer-transformation identifiability, system-boundary attribution, phenomenology-neutral self-modeling, and persistent observation are incubated as specialized RCF modules or instruments. This keeps interoperability from becoming a universal container for every epistemic problem.

What would justify promotion to an independent framework?

Promotion would require evidence that framework interoperability has distinctive problems and operators not adequately captured by RCF alone. Candidate additions include many-to-many construct mapping, formal translation-loss metrics, tribunal-conflict resolution, framework-relative proof-duty maps, power-sensitive intake rules, and validated procedures for preserving untranslatable residuals.

If those additions generate novel discriminators, reduce measurable misrepresentation, improve cross-domain transfer, or enable cooperation where simpler methods fail, the programme should be promoted. If not, it should remain what it currently is: a useful application layer.

Plain-English formulation

You do not have to become us before we can compare what we think

Bring the strongest version of your framework. Tell us what you are trying to explain or preserve. We will identify what your view and its strongest rivals already share, what each one adds, what gets lost when their vocabularies are translated, and what kind of evidence could actually bear on the remaining differences. If the evidence distinguishes them, revise. If it does not, keep the shared result and leave the rest open.

That means a physicalist does not have to become a process philosopher before the comparison starts. A process philosopher does not have to accept physicalism’s definitions. A phenomenologist does not have to pretend lived structure is identical to neural measurement. A neuroscientist does not have to treat first-person certainty as a mechanism. A theologian does not have to abandon theological meaning in order to expose historical or causal claims to evidence. A skeptic does not have to accept the theological interpretation merely to take the experience seriously.

The protocol asks every side for the same thing: preserve what works, identify what you add, show what difference it makes, and remain capable of changing if the world pushes back.

Sources and provenance

Primary project links and selected neighboring literature

Project architecture

• Sweet Rationalism: intellectual ethos
• Recursive Corrigibility Framework: methodology
• A Provisional, Boundary- and Scale-Relative Realism: ontology

Selected interoperability and pluralism lineages

• Susan Leigh Star & James R. Griesemer, “Institutional Ecology, ‘Translations’ and Boundary Objects,” Social Studies of Science 19(3), 1989. DOI 10.1177/030631289019003001.
• Harry Collins, Robert Evans & Mike Gorman, “Trading Zones and Interactional Expertise,” Studies in History and Philosophy of Science 38(4), 2007. DOI 10.1016/j.shpsa.2007.09.003.
• Stephen H. Kellert, Helen E. Longino & C. Kenneth Waters, eds., Scientific Pluralism, University of Minnesota Press, 2006. Publisher record.
• Helen E. Longino, “Interaction, pluralism, and community in conflictual contexts,” Studies in History and Philosophy of Science 112, 2025, 12–20. DOI 10.1016/j.shpsa.2025.05.004.
• Catarina Dutilh Novaes, “Carnapian explication and ameliorative analysis: a systematic comparison,” Synthese 197, 2020, 1011–1034. DOI 10.1007/s11229-018-1732-9.
• Barbara Mellers, Ralph Hertwig & Daniel Kahneman, “Do Frequency Representations Eliminate Conjunction Effects? An Exercise in Adversarial Collaboration,” Psychological Science 12(4), 2001. DOI 10.1111/1467-9280.00350.