Ontic caching and the ontological cleaver
A theory of how ideas become so deeply built into a logic system that it starts resisting evidence that they may be wrong
I have published two articles today on my Recognition–Reconstruction–Reality (R–R–R) framework:
A simple introduction to the idea that reality is compressed into recognised tokens, and that the meaning of these can drift — so we need reconstruction processes to re-ground and correct them.
A practical AI audit tool that uses those insights, together with a knowledge base of “bad patterns”, to look for breaks in the authority “chain of custody” across an official document set.
Now comes the third part.
As I was reading one of the AI outputs, I had a genuine “jaw on the floor” moment. I think I may have spotted a hairline philosophical crack between epistemology and ontology — between what we know and what we take to exist as the objects of our reasoning.
The problem is surprisingly simple:
Before we can reason about something, we have already had to recognise it as something.
A court. An order. A diagnosis. A danger. A fact. A doctrine.
That recognition step normally disappears from view. We stop seeing “something recognised as a court” and simply see “the court”. From then on, perfectly competent reasoning can proceed inside the world created by that recognition — even if the original recognition has drifted away from reality.
Normally, reconstruction provides the way back to reality. It lets us reopen the recognition, check what made this thing into, say, “a court” in the first place, and test whether that recognition still holds up. But if reconstruction is blocked, the mismatch with reality can persist: the system keeps reasoning correctly, but from something it has wrongly recognised.
And through that tiny gap, all kinds of administrative nasties can enter our lives:
counterfeits can inherit the authority of the genuine article,
invalid acts can acquire the status of valid ones,
assumptions can harden into facts,
labels can substitute for evidence, and
decisions built on those recognitions can accumulate until nobody wants to reopen the original mistake.
The danger is that once the counterfeit has acquired the same recognised “it-ness” as the genuine article, ordinary downstream reasoning may have no way to tell them apart.
The problem may not be bad reasoning at all.
The reasoning can be impeccable.
The thing being reasoned about may have been wrongly instantiated before the reasoning even began.
I have used AI to create the following summary.
I hope a few of you get the same “jaw on the floor” moment — and find yourselves wondering how something this simple could have been hiding in plain sight.
My bots tell me that all the individual ingredients have precedents. What we haven’t yet found is anyone combining them into quite this architecture.
That doesn’t prove it is absolutely new. But it does make the question rather interesting.
Enjoy!
Introduction
R–R–R explains how we decide what something is, build other assumptions and decisions on top of it, and can eventually become so invested in that original decision that it becomes increasingly difficult to correct — even when reality says it is not what it claims to be.
R-R-R could be called a theory of “it-ness”:
How does something become an “it” that we can reason about in the first place?
Philosophy asks if it exists and how we know things about it. R–R–R asks about the machinery in between: how something first acquires the status of an “it” about which those questions can even be asked.
My central claim is architectural rather than terminological.
None of the main ingredients — recognition, provenance, reconstruction, path dependence, feedback, predictive updating, institutional closure, trust chains, or corrigibility — is claimed as unprecedented in isolation.
The potentially new object is their recursive arrangement.
Three ideas should be kept distinct:
Ontic caching is the mechanism: recognition compresses enough provenance and structure for something to become “operationally available as X” — in plain English, how something becomes an “it”. In everyday language: we remember what something is without continually remembering why we are entitled to call it that.
The ontological cleaver is the method: move upstream of X, separate the recognised state of the “it” from the operation that made it an “it”, and reconstruct X back to its provenance and warrant — how we check what made it an “it”.
Corrigibility architecture is the broader object of study: the conditions under which Reality retains, or loses, an effective route back into recognised states — how an “it” can remain what we think it is, or cease being treated as such when Reality no longer supports it.
Together, they expose a possible missing operational layer between epistemology and ontology:
the machinery by which something becomes an “it”,
how its “it-ness” can be checked, and
how Reality can correct “it” when the recognition is wrong.
1. Most reasoning begins too late
Consider various professions and their core activity:
A therapist reasons about a belief.
A lawyer reasons about a court order.
A theologian reasons about Scripture.
A scientist reasons about an observation.
But before any of those arguments can begin, a prior operation has already occurred.
Something encountered in Reality has been recognised:
as a belief;
as a court order;
as Scripture;
as an observation.
Ordinary reasoning therefore looks roughly like:
X → reasoning about X → downstream conclusions
R–R–R inserts the missing setup operation:
Reality → Recognition → X → downstream reasoning
and then supplies the reverse operation:
X → Reconstruction → provenance / grounds → Reality
This separates what ordinarily appears as one thing into several distinct elements:
the recognised state;
the recognition operation that instantiated it;
the provenance that warrants it;
the structures subsequently built upon it;
the independent constraints capable of correcting it.
That is the ontological cleaver.
2. The missing layer between epistemology and ontology
Epistemology asks questions such as:
What do we know?
What warrants belief?
Ontology asks:
What exists?
What kinds of things are there?
R–R–R inserts an operational question between them:
What made this available to the system as this kind of thing in the first place?
Before we can ask whether propositions about X are justified, X must already have become available as X:
Before we reason about “the court”, something has been recognised as a court.
Before we reason about “danger”, something has been recognised as danger.
Before we reason from “Scripture”, some body of texts and authority relations has acquired that recognised status.
Recognition therefore does more than generate another belief. A belief is something we can reason about. Recognition is part of the machinery that determines what there is, for us or for a system, to reason about in the first place.
Recognition institutes an operative ontology for subsequent reasoning.
Reconstruction then asks whether that instantiation remains warranted.
The loop is therefore:
Reality → Recognition → operative ontology → reasoning/action → Reconstruction → Reality → possible re-recognition
The object of study is the corrigibility of instantiated ontology.
3. Recognition is ontic caching
Why is this operation normally invisible?
Because recognition exists largely to make reconstruction unnecessary.
Human beings cannot continually reconstruct the provenance of everything they encounter.
I see a chair.
I do not reconstruct chairness.
I recognise my friend.
I do not authenticate their identity from birth records.
I spend money.
I do not reconstruct the monetary system.
I receive a court order.
I do not reconstruct constitutional and statutory authority every time I read one.
Cognition and civilisation depend upon an enormous optimisation:
Recognise now; reconstruct when necessary.
Recognition therefore functions like a cache.
Complex Reality, prior experience, provenance and validation collapse into a cheap usable token:
money, judge, mother, danger, diagnosis, Scripture, safe
Call this ontic caching:
Recognition compresses enough provenance and structure for something to become operationally available as X.
That compression is not a flaw.
It is what makes reasoning tractable at scale.
4. Successful recognition hides the recogniser
A successful cache does not continually announce itself as a cache.
Likewise, successful recognition stops appearing as recognition.
We do not normally experience:
“something currently recognised by this system as danger.”
We experience: danger.
We do not usually say:
“the institutional act currently recognised as a judicial order.”
We say: the order.
The recognition operation disappears behind the recognised state.
Hence:
A discipline’s ontology begins where its routine reconstruction stops.
Psychotherapy needs categories such as: belief, trauma, attachment, resistance.
Law needs: court, order, jurisdiction, judgment, precedent.
Science needs: observation, measurement, evidence.
Theology needs: Scripture, revelation, doctrine.
A discipline cannot reconstruct these from first principles whenever it uses them.
It must cache.
Professional sophistication therefore develops largely downstream of recognised objects whose setup operations have become invisible.
That is efficient.
It also creates a blind spot.
5. The ontological cleaver
R–R–R deliberately breaks apart this abstraction.
Ordinary reasoning asks:
What follows from X?
The ontological cleaver asks first:
What made this X?
Then:
What warrants continuing to recognise it as X?
Then:
What has become dependent upon X?
And finally:
Does X still survive reconstruction under independent constraint from Reality?
This is analogous to breaking abstraction during software debugging to peer inside the underlying logic.
For example, a programmer may reason perfectly correctly about:
customer.balance
But if the underlying object has been mistyped, misidentified or incorrectly populated, reasoning about the value cannot fix the underlying problem.
At some point the debugger must ask:
How did this object acquire its type and state?
R–R–R applies the same manoeuvre epistemically:
Do not debug the conclusions yet.
Debug the ontology that made the conclusions possible.
6. The psychotherapy example
Consider:
“I know I’m safe, but I feel unsafe.”
If “safe/unsafe” is treated as a single primitive psychological state, this looks contradictory.
R–R–R separates:
Propositional recognition → safe
from:
Operative recognition → danger
The second need not be an explicit proposition. It may be bodily, affective, procedural, implicit or relational.
The person can therefore sincerely know:
“I am safe.”
while an older recognition process continues returning: danger.
The apparent contradiction is partly dissolved by better decomposition.
This also reframes the familiar problem of insight without change.
Instead of:
Why does the person know the correction yet resist it?
we can ask:
Why has recognising the correction not caused execution of the reconstruction required to change the operative recognition?
Thus:
Recognise(correction) ≠ Execute(reconstruction).
Once those operations are separated, the deeper question becomes almost the reverse:
Why would we ever expect recognising the correction automatically to execute the reconstruction required by it?
In other words, knowing that the old “it” is wrong does not, by itself, rebuild everything that was constructed on the assumption that it was right.
Giving up smoking is more than simply not putting burning sticks in your mouth.
7. Reconstruction is cache validation
Recognition gives speed by hiding provenance.
Reconstruction reverses that optimisation.
It asks:
What produced this recognised state?
What warranted it?
Does that warrant still hold?
What has subsequently been built upon it?
Does the recognised state survive present Reality?
Reconstruction therefore performs something like cache validation.
Sometimes the cached recognition remains valid.
Sometimes it requires updating.
Sometimes the underlying conditions have changed.
Sometimes the original recognition was mistaken.
Sometimes it was entirely correct when formed but is now obsolete.
The crucial property is therefore not permanent correctness.
It is corrigibility.
8. Reality supplies the independent constraint
Reality has a specific operational role in R–R–R.
Reality is what recognition must ultimately answer to. Reconstruction is the route by which Reality can correct recognition. If that route becomes blocked, the system can drift away from Reality.
Reality is therefore not merely whatever the recognition system currently calls “real”.
Otherwise the framework becomes circular.
The recognition system can classify events, generate explanations and achieve internal coherence.
But it cannot legitimately guarantee successful reconstruction by redefining away every contrary constraint.
The healthy corrective loop is:
Reality → Recognition → action/reasoning → consequences → Reconstruction → Reality → possible re-recognition
A healthy system need not always be correct.
It must remain open to this loop.
This yields an important distinction:
correct / incorrect is primarily a property of a current state;
corrigible / incorrigible is a property of an architecture.
A system can presently be wrong yet healthy if Reality retains an effective route by which it can eventually be corrected.
A system can presently be right yet epistemically dangerous if that route has closed.
9. Dependencies make recognised states load-bearing
Once X has been recognised, other operations begin using X.
So:
Recognition X → Dependency 1 → Dependency 2 → Dependency 3 → …
A child may recognise:
conflict = danger
Then perhaps:
danger → vigilance
vigilance → appeasement
appeasement → attachment strategy
attachment strategy → relational expectations
relational expectations → identity
The original recognition has become infrastructure.
The same occurs institutionally:
A recognised legal state generates decisions, expectations, procedures and reliance.
A scientific recognition generates theories, experiments, papers and funding programmes.
A theological recognition generates doctrines, institutions and moral commitments.
The recognised state becomes load-bearing.
10. Restart cost
Once dependencies accumulate, reopening X no longer means merely changing one proposition.
It may require recomputing much of what has been built upon X.
That is restart cost.
Recognition saves computation by collapsing provenance.
Dependencies accumulate on the compressed state.
Reconstruction reopens the compression.
Therefore:
The more successfully a recognised state has functioned as infrastructure, the more expensive its reconstruction can become.
This distinguishes structural entrenchment from mere strength of belief.
Two people may explicitly assign equal confidence to the same proposition while facing radically different consequences if it changes:
One recognition may be nearly disposable.
Another may support identity, attachment, livelihood, legitimacy, doctrine or survival strategy.
The second is load-bearing.
11. Authority coupling multiplies the cost
Dependencies alone do not explain the full effect.
Recognitions can become coupled to meanings carrying exceptional authority:
vigilance = survival
compliance = attachment
self-sacrifice = moral goodness
existing decision = institutional legitimacy
recognised status = protection of vulnerable people
interpretation = fidelity to God
Now reconstruction threatens more than factual correction.
It appears also to threaten:
survival;
attachment;
identity;
morality;
legitimacy;
sacred duty.
Authority coupling therefore changes the economics of correction.
The question:
Could X be wrong?
inherits another question:
What higher value appears threatened if X is reopened?
That can make a recognition extraordinarily expensive to reconstruct even when contrary evidence is strong.
12. Recognition curvature
Ordinary path dependence says:
Reversal becomes expensive because much has been built upon the existing path.
R–R–R identifies a narrower recursive effect.
At sufficient restart cost, the consequences of reopening X begin influencing how information capable of reopening X is processed.
That is recognition curvature.
The system no longer merely faces a costly update.
The cost of the update begins affecting the update mechanism itself.
Contradictory evidence may be:
ignored;
reframed;
downgraded;
procedurally excluded;
experienced as threat;
absorbed through auxiliary explanations;
or redirected against the person presenting it.
The surface mechanism varies by domain.
The architecture is:
Recognition → dependencies → restart cost → impaired Reconstruction → preserved Recognition
Path dependence makes reversal expensive.
So:
Recognition curvature makes the expense of reversal influence the processing of reversal-relevant evidence.
This is where restart cost enters epistemology.
13. The hardest errors may contain correct reasoning
Suppose Recognition X is wrong.
Everything downstream may nevertheless be competently reasoned.
X → valid inference 1
valid inference 1 → valid inference 2
valid inference 2 → valid inference 3
A large coherent structure can emerge.
Nothing about the validity of those downstream operations repairs X.
Indeed, every additional valid inference may create another dependency and make reconstruction more expensive.
False conclusions are visible to ordinary logical hygiene. We can inspect the inference, find the error, and correct it.
False recognitions are harder to see because they sit upstream of the reasoning. Once the recognition has disappeared into the ontology, everything downstream can be internally sound.
Hence:
The hardest errors may be false or obsolete recognitions upon which too many correct downstream operations have subsequently been performed.
This goes beyond the familiar observation that a valid argument can begin from a false premise.
The error may occur before propositional reasoning properly begins.
The object itself may have been wrongly instantiated.
The reasoning can be sound.
The operative ontology can still be wrong.
Correct reasoning cannot rescue a wrongly instantiated object.
Worse, it can deepen the upstream error by building more things that depend upon it.
14. Why mature disciplines can become trapped
This provides a possible explanation for some stubborn professional problems.
A discipline may be trying to solve a problem using objects instantiated downstream of the operation that generated it.
Psychotherapy debates:
insight, resistance, belief, schema, attachment.
But the decisive distinction may sit upstream:
what is being recognised, by which operative recogniser, and why can the reconstruction required by the recognised correction not execute?
Law debates the properties of an order.
But a prior issue may concern how the relevant act acquired the recognised status “order” and whether that status reconstructs to sufficient authority.
Theology debates conclusions from Scripture.
But some disputes may actually concern the root recognition through which texts or institutions acquired authoritative status.
This gives a deliberately provocative proposition:
A discipline’s hardest problems may sometimes be generated by the ontology in which the discipline is trying to solve them.
The ontological cleaver moves one level upstream.
15. Why the framework travels
Why should the same framework illuminate psychotherapy, jurisprudence, theology, science and institutional behaviour?
Not because their subject matter is fundamentally alike.
Because their information architecture partly is.
Each must:
compress complex Reality into actionable recognitions;
operate recursively upon those recognised states;
accumulate dependencies upon them;
avoid continually recomputing provenance;
and nevertheless:
retain a route by which Reality can force reconstruction and re-recognition.
In this limited sense they are recursively committed recognition systems.
Their content differs.
Their vulnerability is structurally similar.
16. The closest antecedents
The individual parts of this architecture have substantial intellectual ancestry:
Peirce supplies fallibilism, interpretants and recursively continued inquiry.
Sellars attacks the idea that cognition begins from an innocent, uninterpreted Given.
Lakatos shows how a hard core can be protected by adjustments elsewhere in a system.
Luhmann describes recursively self-reproducing systems that process disturbances through their own internal distinctions.
Predictive processing models hierarchical expectations whose precision affects responsiveness to prediction error.
Path-dependence theory explains why accumulated commitments make reversal increasingly expensive.
Trust architectures such as PKI make explicit the importance of provenance, validation chains and class-wide consequences when a trust anchor fails.
R–R–R does not need to deny any of these.
Its residual claim is narrower:
Recognised states become operative ontologies upon which dependencies accumulate; dependency and authority coupling raise the cost of reconstruction; and sufficiently high reconstruction cost can begin altering the treatment of the very evidence capable of forcing reconstruction.
The novelty claim therefore lies in the recursive architecture connecting:
ontic caching;
dependency accumulation;
authority coupling;
restart cost;
reconstruction;
and corrigibility to Reality.
That claim remains open to hostile reduction.
It should.
17. The possible missing “ology”
The architecture sits awkwardly between several established domains:
It is epistemological because it concerns warrant, evidence and correction.
It is ontological because recognition institutes the objects upon which subsequent reasoning operates.
It is cybernetic because its central issue is feedback and recursive correction.
It is systemic because dependency, coupling and restart cost are relational properties.
The nearest sober descriptions may therefore be:
epistemic systems theory
or:
ontological cybernetics.
But the more specific R–R–R object is:
the reconstructability and corrigibility of instantiated ontology under independent constraint.
A possible name is therefore:
reconstructive ontology.
Another is:
corrigibility architecture.
The computational formulation is:
ontic caching.
These names refer to different aspects of the same structure:
Ontic caching describes how recognised states are instituted.
The ontological cleaver describes how they are reopened.
Corrigibility architecture describes whether Reality retains an effective correction pathway.
18. The ontological cleaver in one pass
The whole method can be expressed as a sequence of questions:
What made this X?
What warrants continuing to recognise it as X?
What has subsequently become dependent upon X?
What higher-authority meanings have become coupled to X?
What would reconstruction now cost?
Is that cost affecting the treatment of reconstruction-relevant evidence?
Can Reality still force re-recognition?
That sequence separates:
state from operator;
object from typing operation;
recognition from warrant;
coherence from reconstructive validity;
current correctness from corrigibility;
cached ontology from provenance;
correction recognised from correction executed.
A problem can then be reassembled at the level at which its causal structure actually resides.
19. The architecture in six propositions
The argument can finally be compressed into six claims.
1. Recognition is necessary ontic compression.
Complex Reality and provenance are collapsed into actionable recognised states.
2. Successful recognition disappears into ontology.
The system stops seeing “recognised as X” and simply sees X.
3. Reasoning and dependencies accumulate upon X.
The recognised state becomes progressively load-bearing.
4. Reconstruction reverses the compression.
It reopens provenance, warrant and dependency.
5. Reconstruction cost can become recursive.
At sufficient restart cost, the system begins processing reconstruction-relevant evidence in ways that preserve the recognition.
6. Incorrigibility occurs when Reality loses an effective route back into recognition.
The system can then preserve its recognised version of the world even as the world itself increasingly contradicts it.
• • •
That produces the shortest complete formulation of the framework:
Recognition caches ontology.
Reconstruction audits the cache.
Reality prevents the cache becoming sovereign.
Or, philosophically:
Recognition institutes operative ontology.
Reconstruction audits it.
Reality arbitrates.
The largest implication follows:
Some supposedly hard problems may be hard because reasoning begins after the decisive operation has already disappeared from view.
The ontological cleaver moves one step upstream.
It asks what instantiated the object around which the problem has been built.
Sometimes the problem survives intact.
Sometimes its location becomes clearer.
And sometimes the apparent problem turns out to have been generated by the cached ontology in which everyone was trying to solve it.