JCAS¶
Judgement-bound Context Assembly System.
Status: Canonical lexical document. Version 2.6, 2 October 2026. The name: JCAS, for judgement-bound context assembly system. No declaration: this concept carries no badge. The reason is given below.
Context assembly is bounded by judgement.
That is the whole claim, and the canonical name is the same sentence compressed rather than a label attached to it. Judgement-bound is bounded by judgement. Context assembly is what is bounded. System is the thing in which this holds. Nothing in the name is decoration, and a reader who has the sentence can derive the name, which is the test a term should be able to pass and most cannot.
The preposition does the work. Judgement is not a limit on how much may be assembled. The assembling starts from a judgement the maker has already made, when they know what the text can become, and it stops being provisional only when the maker judges what was assembled. Without a judgement behind it, an assembly has no direction to follow. Without one coming, it does not end. It just stops.
This document fixes a term and the arrangement it names. The term came together through several refinements. I set each of the following aside, for the reason given:
- context provider — names a person by the one part of their contribution that can be handed over, and is exhausted by its own supply: the more complete the context, the less needed the provider
- human in the loop — makes a person a component in a control system, where being in a loop means being informed rather than deciding
- judgement-aware — describes a temperament, not a condition; a system can be aware of judgement indefinitely and never be judged
- held assembly — plain and short, but it names the pause, and the pause is not the thing, and the pause is not what is being described
- judgement assembly system, and its acronym JAS — I proposed this as the prose short form and dropped it on 8 August: dropping bound and context leaves a phrase naming a system that assembles judgement, which is the claim exactly reversed
- judgement-bound context assembling system — the gerund kept the process visibly open, at the cost of a phrase nobody would say twice; system already carries the ongoingness that assembling was there to supply
What survives is what I kept.
I fixed this on 8 August 2026 and reversed myself twice inside the same day, once on the gerund and once on the short name, and both reversals are in the list above. I am recording the date because a term settled in a morning has not been tested by anything. I do not know whether bounded holds. It is possible that what people actually reach for is held, which I rejected on grounds I still think are right, and if they do reach for it then I will have been wrong about the word and right about the reason, which is the least useful way to be half correct.
The judgement-bound context assembly system is an SDC concept. It belongs to the Split-Domain Cognition canon and sits alongside the canon's other foundational marks. It is not the property of any project that builds on SDC; it travels with the discipline wherever the discipline travels.
The name¶
JCAS. That is the name, and it is the only one.
The letters stand for judgement-bound context assembly system, and the expansion is what the name contains rather than a second name competing with it. Write JCAS. Say JCAS. The expansion belongs at a first mention and in a definition, in the way that any set of letters has to be able to open once.
I ran a three-form system for a week — the full phrase where the concept was argued, a bounded assembly in prose, the letters on a form — and collapsed it on 15 August 2026. Three names for one thing is three chances for a reader to think there are three things, and the prose form was quietly becoming the real name while the document insisted the full phrase was canonical.
What the collapse costs, stated rather than hidden. The full phrase is derivable from the claim: a reader holding context assembly is bounded by judgement can reconstruct it, which is the test I said a term should pass and most cannot. JCAS does not pass that test. Four letters carry no argument, and anybody meeting them cold learns nothing until the expansion is supplied. I am accepting that cost for what a single short name buys — one thing with one name, sayable in a sentence, and no drift between registers.
What may be dropped is still decided by what inverts, and this is why the letters are the four they are. Judgement assembly system, and its JAS, read as a system that assembles judgement: the claim exactly reversed, since judgement is the one thing such a system does not assemble and is instead held to. The C is not decoration. It is what stops the inversion.
The negative is un-bounded, and it takes the hyphen.
Without the hyphen the word is ordinary English for limitless, which is not the failure being named and pulls a reader the wrong way. With it, the word can only be read as the negation of bounded: no judgement behind the assembling and none coming after it, so it simply goes out. I went round this twice — first unbound, then unbounded — and both times I was right about the problem and wrong about the fix. The difficulty was never the prefix. It was that the unhyphenated form is already a word with a job.
Etymology¶
Context — from Latin contextus, past participle of contexere, to weave together: com-, together, with texere, to weave. The etymology carries the argument here. Context is not a possession that somebody holds and hands over. It is something woven, which is to say assembled, which is to say the product of an operation. It is not a tribute paid to one. Every formulation that treats context as a thing supplied to a system has the direction of the weaving backwards.
Assembly — from Old French assembler, ultimately from Latin ad- with simul, together. The noun names both the act and its result, and on its own the second sense would close what this concept needs open. System resolves it: a context assembly system is plainly a system that assembles context, not a quantity of context that has been assembled.
Bound — from Old English bindan, to tie, to place under obligation. I chose it against aware. The word also gave us bounden, which survives only in bounden duty and in nothing else anybody says, and which I looked up for no reason connected to this document. An obligation is owed to somebody and can be outstanding; an awareness is a disposition and can be held indefinitely at no cost to anyone.
Semantics¶
Context assembly is bounded by judgement. Stated at length: a JCAS — a judgement-bound context assembly system — is an arrangement in which the maker judges first, by writing and correcting until they know what the work can become, and others assemble only inside that judgement. The others may be people, or small models that answer questions people ask of the material, by pointing at the passage that holds the answer or by deciding from a list written in advance; none of them writes. Small names the size of the window, not the door. A model of any size, on any machine, is small here when code holds its answers to a few tokens, all of them numbers: the number of a passage in the material, the number of an item on a list it was given, or none. The assembly is built in layers, each kept distinguishable and held until it has been judged, and the work is not finished until the maker has judged the result and holds it.
That is the whole claim, and the name belongs inside it. JCAS is not an abbreviation introduced later for convenience. It is what the thing is called once anybody has one.
The definition once carried three more words — under standards set in advance — and they are gone. Why they are gone is The loop, below. The loop and Working through it are the two parts of this document that came out of use rather than out of thinking, and the second corrects the first.
Small models may assemble, and no model writes. This document has drawn that line in three places. From 13 September 2026 it said the assembling system was AI, and the model wrote the layers. On 26 September it removed models altogether and made the layers views that one maker holds. Since 2 October a model may assemble again, in a narrower part: it answers questions people ask, by pointing at the material or deciding from a list the maker wrote, and the words stay with people. Whether a JCAS is bound still depends on how the work is set up. The account is in With small models, below.
Gathering without judgement is old. A research department can gather for years without anyone judging what it gathers, and better gathering does not help. The term draws the line SDC draws: between the judgement of what the work can become and the assembling that carries it further, whoever does the assembling.
The binding is a capability, not a posture. A system that cannot hold its assembly open pending judgement cannot be bound, however carefully the people around it intend to judge. The withholding has to be built. A check that runs after the assembled material has reached its reader prevents nothing, because prevention means withholding, and withholding is paid for in latency, in convenience, and in the pleasure of an immediate answer. Systems that deliver continuously have nowhere for the assembly to wait. They are not badly arranged. They are incapable, and the incapacity is invisible because nothing about the output announces it.
Where the binding is absent, what accumulates is judgement debt: material assembled, presented as finished, and never read by anyone who would carry the reading. The debt is not paid down by the assembling getting better. It is paid only by somebody judging, and it accrues in the meantime whether or not anybody is counting.
The loop¶
This section records how the loop was found on 8 August 2026, with one instrument on one piece of fiction. Its order was too fast, and Working through it, below, corrects it. The system in it was a model that wrote; since 2 October 2026 a model in a JCAS writes nothing (With small models, below). The account of how it was found is left as it was.
I built an instrument that enforced standards set in advance, used it on my own work, and the requirement did not survive the first use.
The instrument asked for a bounds file: what would count as good and what would count as done, written before anything was assembled. I could fill it in. What I filled in was not my standard. It was a standard I would have been willing to defend, which is a different object and a more flattering one.
A system then assembled eleven segments of a piece of fiction. The result was competent, complete and lifeless. Lifeless was not a criterion I had withheld. It was a word I did not have until the thing existed.
I rewrote three of the eleven. I was not stating a standard; I was fixing three passages that were wrong. But those three edits were the first place my standard had ever been visible, to me or to anything else. Every assertion wants its cause attached and the cause is usually unflattering. The narrator may reason in public and lose. Somebody in every arrangement has to be relieved by it. A short sentence goes where the machine wants a long one.
None of that could have gone in the bounds file. Not through carelessness. I did not know it. I found out what I wanted by seeing what I did not want, and correcting it three times.
A standard is demonstrated, not specified. It exists in acts before it exists in sentences, and for most makers it never becomes a sentence at all. Asking somebody to write their standard down before they have anything to apply it to asks for the hardest part of the work first, in the abstract, with no instance in front of them. Most people cannot. The ones who produce a confident answer have usually described a standard they admire rather than the one they use.
So the arrangement is a loop. A system assembles at span. The maker edits a part of it — a small part, and not as a specification. The system reads the standard off those edits and extends it to the rest. The maker edits again, and where the extension was wrong the correction sharpens what the standard was. This continues until the maker says it works, and works is their word, and there is no test for it.
Three things follow.
The first material's job is not to be good. It is to be specific enough to be wrong. I first wrote this about a model's first assembly. In practice the instance that judgement needs comes from the maker's own writing, which is already an assembly of their thoughts. A blank page cannot be corrected. A vague draft is worse than a lifeless one, because there is nothing in it to disagree with.
Scaling and layering are work at span, and the only part that is. Taking a standard that exists in three instances and carrying it to the other eight places it applies, or building further layers under it, is work at scale. It is not judgement. It is what judgement, once shown, can be extended by.
The span means something different now. It is not a cooling-off period on a finished thing. It is the interval between an edit and the next layer, so that a standard is not read off a mood.
The process in the canon's own terms¶
The maker writes first. Writing is already assembling: the maker gathers their own thoughts and puts them in an order. The principle counts a poem being written among the acts of mind that are rightly unified, so nothing needs splitting yet. No standard is written down in advance.
The maker judges until they know what the text can become. The judgement is an act on the text, not a statement written beforehand. It is visible in the corrections, and it is what everything after it is bound to.
Then assembly goes on inside the judgement, and the job is layering. Other people may do it as well as the maker, and so may small models that answer questions people ask, by pointing at the material or deciding from lists the maker wrote; none of them writes any of it. Each new layer extends what the judgement found to more of the material, or adds a layer under it. The direction was set by the judgement before it.
Each layer is held. Nothing leaves.
The maker edits part of it, and the edit is the judgement. Not a verdict delivered about the assembly — an act performed on it. In the first design the judgement was a mark, keep or cut, and the mark was what got recorded. In the loop the mark is derivable and the rewrite is what carries the standard, because a person correcting a passage says more about what they want than the same person grading it.
The layers stay distinguishable. Each layer stays apart from the others as the work grows, so every part can still be pointed at and whoever made it named. JCAS is this layering, held inside one judgement.
The binding runs forward. This is the part I had backwards. I had judgement binding the assembly it was about, which is why the instrument withheld release until every segment carried a mark. What a judgement actually binds is the next assembly. The edit of three segments is what the remaining eight are then assembled under. The thing being judged is already made; the only thing a judgement can bind is what comes after it.
Assembly at scale is qualification, not judgement. In SDC's stage vocabulary, what a new layer does between edits is read evidence into structured claims — take a standard that exists in three instances and find where else it applies, or layer under it. It makes no calls. A small model that decides from a list written in advance is doing exactly this. The moment a layer makes one, the loop has broken and the maker is approving rather than judging.
Judgement debt accrues by layering twice. If a second layer is built before the maker has edited the first, it compounds on a standard nobody confirmed. The debt is not in the words. It is in the fact that a standard was inferred, applied at length, and never checked.
Release is when the maker says it works. There is no other test, and no number.
And one risk I have no guard for. The loop can capture the person inside it. If the extension gets good enough I stop writing and start approving, and approval is not authorship. The loop ends when I say it works, and somebody who has stopped writing tends to say it works early.
Working through it¶
Added 13 September 2026. It has a model writing the layers. Without a model, below, withdrew models on 26 September 2026, and With small models let them back on 2 October to point and decide, never to write.
Until this date JCAS was theory. I had fixed the term, found the loop with one instrument on one piece of fiction, and reasoned out the rest. Then I began working through it on my own writing, and the order turned out to be wrong.
The loop as I first wrote it is too fast. It has a model assemble first, before anything has been judged. That is not how I work. When I write, I assemble my thoughts, and that first assembling is mine.
Assembly by a model comes after I have judged, when I know what the text can become. By then the direction exists, and assembly has a different job: scaling the text and layering it.
So JCAS is about layering. Some parts of the text are edited by the model and some by me, and the arrangement keeps the two apart as the text is built, so that each part can be pointed at. Every layer the model adds is held until I have judged it, as before. What changed is where the model comes in.
This corrects the loop on one point and leaves the rest standing. A standard is still demonstrated rather than specified, the edit is still the judgement, and the binding still runs forward. What does not survive is the claim that an assembly begins un-bounded, necessarily.
Without a model¶
Added 26 September 2026. Withdrawn on the model point on 2 October 2026 (With small models, below); the account is left as it was.
No model is part of JCAS. The layering I described on 13 September was between parts a model edited and parts I edited. That was the wrong line. The layers that matter are views I hold of the same material: I write from one view, read from another, and correct from a third, and each is mine. A model's layer is not a view anybody holds, so it has no place in the arrangement.
The rest stands. A standard is demonstrated rather than specified, the edit is the judgement, the binding runs forward, and each layer is held until it has been judged. What changes is who assembles: the maker does, every time, from a different view they hold.
With small models¶
Added 2 October 2026.
Without a model made JCAS too small. It kept the whole arrangement inside one person, writing, reading and correcting from different views, and left nobody else a place to carry a judgement further, which is most of what the term is for.
I judge first, by writing and correcting until I know what the work can become. Then others assemble, and only inside that judgement. They can be other people. They can be small models, which answer the questions people ask of the material, each by pointing at the passage that holds the answer or by deciding from a list written in advance.
No model in a JCAS writes. On 13 September I had a model writing the layers, and that is the arrangement the loop's last risk describes: the model writes, I approve, and approval is not authorship. A model that points at a passage or picks from a list leaves the writing with people, and the risk has nothing to work on.
A model of any size, on any machine, can take this part. What makes it small is its window: code keeps each answer to a few tokens, all of them numbers, the number of a passage or of an item on a list it was given, or none, and counts anything else as no answer. It answers in no words at all.
The judgement does not travel with the work. Whoever assembles chooses within what it leaves open and uses what it makes, and nobody assembling draws the scope. Somebody who cannot see what the work can become cannot be handed the part that decides it, however well they assemble.
Everything else in this document stands as written. What has changed is who may assemble, and the one thing none of them may do.
Scope¶
Everything above states the concept at the scope of a single assembly: one body of material, assembled, held, judged, released. That is where I fixed it and it is where it is easiest to check.
It holds at the scope of a practice too, and at that scope exactly one thing behaves differently. The terminus.
At the scope of a work the terminus is real. Release is when the maker says it works, the assembly leaves, that piece is over. At the scope of a practice there is no release. The standard a judgement sets goes back into use, use produces the instances that require the next judgement, and the binding renews instead of ending. A practice is not a long work. It is a sequence of bindings with nothing at the end of it.
Both scopes run the same mechanism, and the question that tests them keeps its shape:
Of an assembly — is a judgement coming, and will somebody hold the result?
Of a practice — is a next binding coming, or was the last judgement the last one?
The second question earns its place because a practice can pass the first one indefinitely and still have stopped. Every individual piece gets judged. Nothing rebinds. A standard set years ago governs everything made since and has not been touched, and each assembly is bound in good order the whole time. That is not judgement debt, because nothing is unpaid. It is a different condition and it has no name here, which is a gap rather than a decision.
Two relations, one word¶
Setting the scopes side by side shows bound doing two jobs in this document. I would rather separate them than leave a reader to do it silently, or to conclude that the definition and the loop contradict each other.
The release relation. The assembly is held, and nothing leaves until a judgement has been performed on it. This is what makes an assembly unfinished, and it is what the definition states.
The standard relation. The standard that judgement sets governs the next assembly. This is what runs forward, and it is what the loop found.
Both hold and they are not the same relation. An assembly is held for a judgement about itself; what that judgement then binds is what comes after it. I had the two tangled, and the tangle is why the Semantics section and the loop can be read as disagreeing.
Re-orientation¶
The loop says a judgement binds the next assembly, and it says this about the standard: the edits to three segments are what the remaining eight get assembled under.
A judgement can also change what is being assembled at all. Not the criterion, the object. Having seen the thing in front of me I am frequently no longer making what I was making, and the next assembly is of something else.
That is a stronger claim than the loop's, and it is checkable the same way. Of any two successive assemblies, ask whether the second was the same object under a revised standard or a different object altogether. Both are bindings. Only the second re-orients.
Definition, and what it is for¶
A JCAS runs perfectly well without the standard ever becoming a sentence. The loop says so plainly: the mark is derivable, the rewrite is what carries the standard, and for most makers it never becomes a sentence at all.
That is sufficient for one maker. It is not sufficient for two.
A standard carried only in rewrites can be extended by a system and cannot be handed to a person, because there is nothing to hand over. Definition is the act that converts a demonstrated standard into a transmissible one, and it stays optional exactly as long as nobody else needs it.
So the concept has a step it did not need while it described one person working alone. The test for it is the plainest in this document: can the maker say what they were correcting for? If they can, the standard has been defined. If they cannot, it is still operating, it is still theirs, and it cannot be given away.
The risk, and where its guard turns out to be¶
The loop closes on a risk I said I had no guard for. The loop can capture the person inside it: the extension gets good enough, I stop writing and start approving, and somebody who has stopped writing tends to say it works early.
At the scope of a work there is nothing to stop that, and the reason is structural rather than careless. The loop's terminus is the maker saying it works, so the failure and the success condition are the same sentence.
At the scope of a practice that terminus is not there. The standard goes back into use and a next binding is coming, so no point in the sequence offers approving a place to rest. Approval is a terminal posture. A practice that refuses its terminus has nowhere to put one.
This is not a guard in the sense of a mechanism that fires and fails loudly. It is the absence of the thing the failure needs.
Why there is no declaration, and the one certificate that is possible¶
A JCAS carries no badge, and I would rather state why than leave it looking like an omission.
A binding is a claim about a judgement that has not happened yet. From outside, an arrangement in which judgement is coming and one in which it never will are indistinguishable: both show assembled material and a stated intention. Any badge here would certify exactly the thing that cannot be seen from where the reader stands. So the concept has a canonical document and no self-declaration, and the asymmetry is a finding rather than unfinished work.
The finding has a scope, and naming it opens exactly one door. What cannot be seen from outside is a present binding. What can be seen, by somebody who was inside the loop while it ran, is a past one: whether the standard was demonstrated, whether it was defined, whether the binding renewed. That is not an inference from the artefact. It is testimony from a position the badge argument assumes nobody occupies.
So a certificate is possible under three conditions, and it is a different object from a badge.
It is issued from inside. Only by somebody who was in the loop with the practitioner for its whole span. A reader of the finished work cannot issue one, and neither can the practitioner.
It certifies a past engagement, never a present state. It says that on these dates, in this work, the binding was made and here is the degree to which the term was implemented. It does not say the practice is bounded now, because that is the thing the original finding says cannot be seen — and nothing about holding a certificate makes it visible.
It is never self-declared. The absence of a self-declaration is the finding and it is untouched.
A certificate of a past engagement and a badge asserting a present arrangement are different claims, and only the second one was ever refused.
Relation to existing terms in the SDC lexicon¶
SDC names the architecture. Variants name the domain-specific applications. The forms of follow name the modes by which an external project declares its relation to the canon.
JCAS names something none of those reach: the arrangement under which the work is done. It is a standing condition rather than a property of an object.
The split ratio bears on this concept at one point only. It reads a finished artefact for conflation against held judgement, and it can only be claimed from recorded provenance. The layers a JCAS keeps apart are that record. No ratio attaches to a JCAS itself. I once had an instrument log an authorship-share number under the ratio's name — what proportion of the calls were a machine's. That was a mistake of mine rather than a fault in either term, and the number is gone: keeping layers apart is not counting them, and the models a JCAS allows write nothing, so there is no share of the writing to count. The question a loop raises has no number in it at all: where was the standard set, and can you point at it.
Linguistics of use¶
JCAS takes the indefinite article when naming an instance — a JCAS — and pluralises regularly as JCASes. It is said as four letters, not as a word.
This is a JCAS; the essay was built in three layers, and each was judged before the next.
The plain identification, used where the concept is being fixed or distinguished from a neighbour.
It is a JCAS on paper, but nothing has been judged since March.
The ordinary working sentence, and the one the concept exists to make sayable. The arrangement can be nominally correct and materially empty, and the vocabulary should let a practitioner say so.
There is no binding here at all. It goes straight out.
The failure named plainly, without technical vocabulary.
That is judgement debt, and it is a fortnight old.
What accrues where the binding is absent. The debt takes a size and a duration, which is most of what makes it usable.
The neighbour this term argues with¶
Terms in this canon position themselves against the nearest serious account rather than against a straw one, and the nearest here is artificial general intelligence.
AGI names a hypothesised threshold of general capability. It is vague about what a system does, and it is a claim about machines alone. JCAS names one specific capability tied to an obligation, and it holds wherever work is assembled, by people or by small models. Of any arrangement of work, anybody can ask whether the assembly is bound to a judgement somebody will make and hold, and get an answer.
AGI asks how capable a system is and treats a threshold as the event worth waiting for. I think generality is the wrong axis. What I want to know about any work in front of me is whether its assembling is bound, and that question has an answer today, and the answer usually costs somebody time they did not want to spend.
Author. Prayas Abhinav.