The Mathematical Shape of the Architecture
Polycontextural logic in Lean 4 and Mathlib — edition PKL Rev4, in two parts
5 August 2026 · All figures of measurement at commit 24f7fe8 unless anchored otherwise · Edition Rev4
Part A · The shape — about this edition
This paper has two parts. Part A sets out the shape: what has been built, how the layers interlock, and where the architecture ends. It manages without a single library name, without figures of measurement, without register entries. Whoever wants to know whether anything here is overstated will find in Part B, for every numeral, the theorem, its axiom profile, its range of validity and its register row — together with the counting routes, the checking apparatus, and the list of errors this project has found in itself.
The separation has a reason. In the three previous editions the bookkeeping stood beside every result, and the reader heard the caveats louder than the theorems. Here the matter stands; the evidence stands together in one place.
The numerals in square brackets point to the carrier table in Part B and are clickable. Where the text says proved, a machine-checked theorem stands there. Where it says posited, an explicit premise stands there. Where it says interpretation, nothing stands there — interpretation is the assignment of the formal to the conceptual, and it is called what it is.
The shape in one sentence. A combinatorics with a built-in direction carries logics that are separated from one another without any map between them being missing; this direction reappears as time and as ascent; the distance between growing richness and unchanging reachability is its measure; and it holds exactly as far as order reaches.
1 · The ground: a combinatorics prior to logic
At the very bottom lies something that is not yet a logic. One may picture it as a row of empty places into which nothing has yet been entered. Two such rows count as the same if the same places repeat one another — nothing further is asked. No values, no truth, no negation; only distinguishability and repetition.
That sounds poorer than it is. Take the row
a b b a c
and beside it
x y y x z
They consist of different signs and are nevertheless the same pattern: the first and fourth place carry the same thing, the second and third carry the same thing, the fifth carries something new. What counts here is not what stands at a place, but which places agree.
Three granularities, and why only the finest carries
Günther distinguishes three degrees by which such rows may be classified, and the distinction is not his invention alone but an obvious gradation.
The coarsest counts only how many distinct signs occur: three in both rows above. The middle one records how often something repeats, without regard to position: here two blocks of size two and one of size one. The finest retains the position-exact pattern.
Only the finest can be the ground of a logic, and the reason is structural, not aesthetic. An
assignment — the entering of values into the places — must be able to say which places carry the
same value. Whoever knows only the block sizes does not know the assignment: “two blocks of size
two” fits a b b a as well as a a b b, and those are different assignment
situations. The coarser degrees are therefore readable marks of the ground, not possible
grounds.
The normal form, and why it is built rather than asserted
The finest pattern can be written canonically. One numbers the signs in the order of their
first occurrence: the first gets zero, each new one the next higher number. From
a b b a c comes
0 1 1 0 2
and from x y y x z the same. Two rows carry the same pattern if and only if their
normal forms agree — that is proved [1], in both directions and in a form that
does not depend on the kind of signs [2], [3].
These normal forms are known in combinatorics; what is remarkable is not their existence but that the architecture has built the canonisation as a procedure and not as a claim. There is a prescription that computes the normal form of any row, and a decidable predicate that says whether a given row is already in normal form. The difference matters: an asserted uniqueness can be false, a built procedure computes.
A construction decision stands here too, and it could easily have gone otherwise. Pattern identity could be introduced as a quotient — one simply declares rows with the same pattern to be the same object. The architecture does not do it. A quotient immediately creates connective work and conceals what the normal form leaves visible; as long as the normal-form semantics carries, it is a purchase on credit. This is no embarrassment but a choice, and it stands so in the corpus.
The built-in direction
Now comes the first piece of information that goes beyond mere ordering.
A pattern can be extended — one place appended — or retracted: the last place struck. These two operations are not mirror images.
Take the pattern 0 1 1 0. It contains two distinct signs. For the fifth place
there are three possibilities: it repeats the first sign, it repeats the second, or it introduces
a third. In general a pattern whose largest sign carries the number k has exactly
k+2 extensions. Retraction, by contrast, is unique: strike the last
place, and precisely one pattern remains.
This is no notational nicety but the matter itself: ascent chooses, descent computes. Whoever extends makes a decision that does not follow from what is given; whoever shortens performs an operation whose result is fixed. Between the two directions lies not a change of sign but a difference in the kind of determinacy.
Günther calls one direction evolutive, the other emanative, and he does so long before a value or a time occurs in his account. This asymmetry is the germ of what returns in chapter 3 as time and in chapter 6 as ascent. At this point it is nothing but a count — and precisely that makes it load-bearing.
k has k+2 continuations and exactly one
retraction — computed.Where the ground and the logic hang together
So far there has been no talk of values. The transition to logic is the moment when values are entered into the places. And here it emerges that the ground does not simply accept the range of values.
The question is: which patterns can be filled with how many values at all? It has a sharp answer. A pattern can be filled with two values if and only if its signs do not go beyond one — proved [4].
One direction is an encoding: whoever has only zeros and ones in the pattern writes down the two values. The other direction is the interesting one. If a two occurs in the pattern, then — because of the construction of the normal form, in which each new sign exceeds the previous maximum by exactly one — there are three places with pairwise distinct signs. Three pairwise distinct signs require three pairwise distinct values, and a two-valued row does not have them. The closing argument is a counting argument, and it is the one place where the proof goes beyond mere recoding.
Worked through at a fixed length, the matter becomes tangible. At four places there are fifteen distinct patterns; eight of them can be filled with two values — and both numbers are proved [5], [6] and not looked up. The unequal pair 15 to 8 says: the ground is richer than any fixed valency. A witness of this excess is quickly found — the pattern
0 1 2 0
requires three values, and no two-valued row satisfies it.
Whoever recalls classical logic will recognise the number: the sixteen binary truth functions rest on eight patterns. The other seven do not occur in it — not because they were overlooked, but because two values do not reach them.
What this chapter does not say
The ground is not a logic, and it does not become one. It carries no truth values, no negation, no inference; all of that comes only with an assignment.
Kenogram is not contexture. When the next chapter speaks of contextures and
this one of k-valued fillability, the reading that a k-fillable row is
“a k-valued system” is an interpretation — it hangs on the bridge
[4] that supports it, and it is expressly labelled as interpretation in the
corpus. The theorem says something about values in places; that values in places yield a logic,
it does not say.
The threefold gradation of granularities is attested in Günther with this explicitness; the finer division of kinds of map with which the architecture works is not — it is a refinement of this project and does not claim him as its authority.
2 · The contextures: separated, without any map being missing
A contexture is an assignment on the ground that has the rank of a logic of its own. Not a region within an encompassing logic, but a logic: with its own distinction of true and false, self-contained. The overall picture is therefore not one logic with regions, but a web of logics over a shared pre-logical ground.
The question around which everything turns is: wherein does their separation consist?
The case at which the obvious answer fails
The obvious answer would be: two contextures are separated if no translation exists between them — no structure-preserving map.
This answer fails at Günther's own basic case. The most elementary case of discontexturality is for him the opposition of Being and Nothing. And these two are for him isomorphic; he appeals expressly to the algebra of his day. Between them there is a complete, structure-preserving map — and they are discontextural nonetheless.
One may read this finding as a contradiction or as a hint. The architecture reads it as a hint, and the hint is precise: the map exists, but it is stated from outside, not executed from within. Whoever stands in one of the two contextures and has only the means that hold there does not arrive at it. This is Günther's own distinction, the same one with which he handles the excluded middle: it holds for both sides and is nevertheless applicable only to one.
The formalisation decision that follows
Discontexturality therefore appears not as the non-existence of a map, but as non-assemblability from the means available inside.
Concretely: one fixes the operations that hold within a contexture — in the simplest case two connectives and a reversal — and asks which operations can be built from them by composition, at any depth and in any wiring. Everything so buildable is inside. What cannot be so built is not.
What is missing is not the map, but the path to it from within.
Two shorter formulations are expressly excluded and carried in the corpus as construction prohibitions. One would be: two contextures are separated if they have nothing in common. The other: they are separated if no map exists between them. The second fails at Being and Nothing, as just shown. The first fails at a fact one can count.
The overlap is counted, not agreed
Consider three values and the two-element regions within them: {0,1},
{1,2}, {0,2}. Any two of them have exactly one element
in common. This is no convention that could have been settled otherwise, but a counting statement
— and it is proved [7].
That is more than it looks. Were the overlap merely a construction decision, one could drop it and model contextures as disjoint; the prohibition would then be a house rule. Since it is a counted property of the situation, the prohibition is not a rule but a consequence: whoever posits the elementary contextures as disjoint is no longer speaking about the same objects. And the overlap really does carry weight in the proofs — it is the reason why an operation can behave differently on different elementary contextures without leaving any region anywhere. Without the shared value that case would not exist.
Where the posit sits, and why its ground is firm
One point must be stated openly, because the whole series that follows rests on it.
A two-element set of neighbouring values is an elementary contexture in Günther's sense. — This is posited, not proved.
No theorem decides whether Günther's concept means precisely this. The posit is the load-bearing choice of the architecture, and it stands as such in the corpus.
But its ground is not arbitrary, and that can be shown. A two-element set is closed under the local operations if and only if its two elements are comparable. If they are not, then even the minimum leaves the set — proved [8]. Then it is no self-contained connection and hence no contexture in the sense of the definition from which the architecture proceeds.
The posit remains a posit. But it does not stand on taste, it stands on a theorem: the limit of the formalisation lies in the concept, not in the proof.
What this chapter does not say
The second fundamental relation is only half redeemed. Beside the symmetric exchange relation between elementary contextures Günther names a second, asymmetric one — that between a contexture and what goes beyond it. Its structural part is available as a theorem; its conceptual part is not. That a change of stage is a change of contexture is closed by no theorem. A contextural boundary never closes by a theorem alone, but by definition and theorem together — and the definition is a posit.
The separation is grasped as non-assemblability and not as an ontological rupture. What is proved are failures of reduction. To read them so that a rupture of being appears in them is interpretation.
3 · Time: a structure that fits into no linear measure
Before the architecture speaks of time, the place where Günther first came upon the matter is worth a look. It lies early, in 1937, and it is agreeably tangible.
Three propositions, each plausible on its own
First: time is more powerful than will. Every decision falls within time. No one can will that a decision had fallen earlier; the will finds the moment at which it expresses itself already given.
Second: thought is more powerful than time. Memory runs counter to time and reaches back to the beginning. Thought is not bound to the direction of the flow — it can survey it, and whoever surveys something is in that respect above it.
Third: will is more powerful than thought. With Fichte: that I think rests on the fact that I will to think. Thought does not set itself in motion.
Take the first two together: if thought is more powerful than time and time more powerful than will, then thought is more powerful than will. The third proposition says the opposite. The three cannot be brought into a ranking; they close into a circle.
Günther draws from this no contradiction but a finding: from such propositions no groups of statements of transitive character can ever be formed, and therefore the course of time in actual consciousness remains always ambiguous.
What the architecture makes of it — and what not
It does not claim that thought, will and time behave this way. That assignment of three names to three corners is interpretation and is carried as such.
What is proved is the other, the hard half: in every transitive, irreflexive ordering — on every carrier, without exception — such a three-cycle is impossible [9]. And there is likewise no structure-preserving image of such a cycle into an ordering [10]; it cannot be accommodated by clever translation either. The core of these theorems manages without the classical instruments of set theory.
The claim is therefore: if the three behave this way, then no linear order is their measure — and not because none has been found so far, but because none can exist. That is less than a philosophy of time promises, and more than an observation achieves.
The stock of movement
Over the web of contextures lies a threefold doctrine of movement — three ways in which something can change, without its yet being said what changes:
- displacement — the same structure at another place;
- reorganisation — the same parts in another arrangement;
- irreversible positing — the step that is not taken back.
Their compositions are smooth in exactly one direction; there is no canonical interchange of directions, and that is not an image but proved [11].
Two differentials, separated by the number of contextures
Time falls into two statements, and what separates them is a single hard quantity: the number of contextures involved.
Within one contexture the cycle of potency stands against every linear order — that is the finding above.
Between contextures the cumulative movement does not close: under an expressly posited premise — that no state reproduces itself — there is no circle, and the ascent admits no reversing assignment. The premise is declared, not hidden; with a self-reproducing state the claim would fall.
The connection with chapter 1
Both statements are the same asymmetry in another guise. At the ground it appeared as a count — ascent chooses, descent computes. Here it appears as the impossibility of reversing a direction or bringing it into a measure. In chapter 6 it will appear a third time, then as a provable limit of the ascent itself.
That is the core of the whole architecture: a single asymmetry, three times, in three languages — combinatorial, temporal, hierarchical.
What this chapter does not say
The naming of the three corners is interpretation. The statement about the relation of thought, will and time is conditional: it hangs on a premise that this paper does not prove and cannot prove. And the premise in the inter-contextural case — freedom from self-reproducing states — is posited.
4 · How one proves a lack
Everything so far runs into negative statements: something cannot be assembled, cannot be brought into an order, cannot be reached with finitely many values. Such statements have a difficulty of their own, and it is not technical but fundamental.
A system that certifies its own limit from within contradicts itself. Whoever says “with my means X is unreachable” uses precisely those means for the statement. The way out that the architecture takes throughout is called the differential: one does not prove positively within one system, but differentially between two.
The procedure always has the same form. One names a poor class — a precisely specified set of structures — and shows: over all its instances the feature in question is impossible. Beside it one places a rich witness, a concrete structure that carries it.
| the hard half | the easy half |
|---|---|
| the feature is impossible — over all instances | the feature is exhibited — a construction |
| all orderings; everything buildable from the local means | one concrete structure, built inside the system |
The impossibility is the hard half. The construction alone would be the easy one, and passing it off for the whole would be the most comfortable form of self-immunisation.
Three marks keep the procedure honest.
First: the poor model is built inside the system, not asserted in a metalanguage. Whoever speaks about “all orderings” must be able to say what he is speaking about.
Second: the result is conditional, relative to the posited premises. It does not compel the sceptic; it symmetrises him — from the comfortable position this can be reduced to the known to the explicit position I posit the one encompassing logic, which is itself a posit and thereby has no precedence.
Third: what is proved is the form, never the naming. That a contexture-dependent choice of operations is unreachable from local means is a theorem. That this choice is called mediation is interpretation. The corpus keeps to this down to its file names: the file carrying the theorem is named after the formal property and not after its philosophical reading.
The procedure applies to the workshop as well. In the most recent working phase it has been applied twice to the architecture itself — once to a theorem about the ascent, whose impossibility half runs over all maps and whose possibility half exhibits an explicit witness (chapter 6), and once to a measurement: where an explanation for a result is given, it requires two versions differing in exactly one quantity. Whoever wants to read that up will find it in Part B.
5 · The measure of the separation
Chapter 2 said wherein the separation consists: in non-assemblability from the means available inside. This chapter says how much that is. For an impossibility becomes information only when one knows how large the unreachable is against the reachable.
The frame is everywhere the same. We have finitely many values ordered by size. The elementary contextures are the two-element regions of neighbouring values. The local means are two connectives — minimum and maximum — and the reversal of order. An operation is called locally classical if on every elementary contexture it acts like one of the two connectives and does not leave it. The question is which of the locally classical operations can be built from the local means.
Two values: no excess
At two values there is exactly one elementary contexture and exactly two locally classical operations — minimum and maximum themselves. Both are trivially buildable. The excess is zero, and that is proved [12] and not merely observed.
This meets Günther's own threshold, and does so literally: for him the classical two-valued system does not yet represent a compound contexture; the second value yields nothing new and stands unmediated over against the first. What is a thesis in him stands here as a counted zero.
Three values: half, and in steps
At three values there are three elementary contextures, and a locally classical operation is nothing other than the specification of on which of them it acts as minimum and on which as maximum: eight choice patterns. Of these exactly four are buildable and exactly four are not — proved [13].
The four reachable ones are not four arbitrary ones. Ordering the elementary contextures by size, they read
(min, min, min) (max, min, min) (max, max, min) (max, max, max)
— a chain of thresholds. Reachable is exactly whoever switches once from minimum to maximum and then stays. Whoever switches back is outside.
Here, in the presentation of this paper, the second reason for non-assemblability appears for the first time, and it is the genuinely interesting one. An operation can be unreachable because it leaves its contexture — it delivers at some place a value that does not occur in the two-element region. Or it can be unreachable although it preserves every contexture and acts classically on each: then it is due solely to the fact that it chooses differently on different contextures. The first reason is transcendence, the second is non-uniformity. The second long had no witness; it has one now, and completely so — the division into four and four leaves nothing open.
The first reason has its own witness: an operation that refuses an offered pair of values as a whole and thereby produces the third value. It does not lie in what can be built — proved [14]. And it lies not somewhere outside, but at a determinate place: the local means preserve three properties, and every operation built from them preserves all three — not merely the generators, but every composition. The refusal violates exactly one of them and preserves the other two.
Lying outside is cheap in itself: of the 19,683 binary operations over three values only 82 are buildable, less than half a per cent. The claim does not rest on that. It rests on the location.
Four values and beyond: two remain
At four values the picture tips. Of the 64 locally classical operations two remain buildable — minimum and maximum themselves [15]. At five values it is 2 of 1,024, at six 2 of 32,768.
And this is no accident of small numbers but a theorem for all value counts from four upwards: a locally classical operation is buildable if and only if it is globally the minimum or globally the maximum [16]. The proof manages without case distinction on the number of values and without machine enumeration; it runs over six families of neighbourhood relations with breaking points given in closed form, and an induction on path lengths.
The threshold lies at four, and it is no artefact. At three values the statement is false — there four of eight are reachable. At two values there is no witness, because there is only one contexture. Mediation begins at three; it becomes completely unreachable from four upwards.
The richness, counted
Where do the numbers 8, 64, 1,024, 32,768 come from? They are not separate counts but instances of one theorem.
A locally classical operation is nothing other than one choice per elementary
contexture. That is proved as a one-to-one correspondence [17]:
forwards one reads off, for each elementary contexture, whether the operation acts there as
maximum; backwards one builds the operation from the choice data. Their number is thereby
2^C(m,2) for every value count m — proved and not observed. The exponent
is the number of elementary contextures, that is, the number of pairs from m values.
It is the same number series that Günther carries for the compound contextures and for
the gradation of ontologies — here it stands as the exponent of a space of
possibilities, and it stands there not as an observation but as a theorem.
“Locally classical” is thereby no longer a mere property but a blueprint: one choice per elementary contexture, nothing further. And the boundary cases carry without side conditions: with at most one value there is no pair and exactly one operation.
The measure
Now the question of this chapter can be answered. From value stage to value stage
- the richness grows strictly — the number of locally classical operations increases at every step [18], and from three values on it exceeds two [19];
- the reachable stays at two — from four values on it is exactly minimum and maximum, and at every stage the same two [16].
That is the most precise version of what “mediation needs more than local means” says: a distance between two numbers, one of which grows exponentially and the other of which stays constantly two. Not an existence claim — “there is something unreachable” — but a characterisation: the set of the reachable is completely determined, and everything else lies outside.
2^C(m,2), the reachable stays at two. The distance between them is the measure of
the separation — both series proved, the scale logarithmic.How much each bound claims
An impossibility statement hangs on the property it works with. If constant operations are admitted to the local means — checkers that always deliver the same value regardless of input — then the bound survives precisely when its property is also respected by constants.
For the witness of transcendence this is not the case: it falls as soon as constants are permitted. For the characterisation from four values upwards and its applications it is the case: they hold. A bound that falls to an added constant claims less than it appears to — and that is written into the corpus as a duty of inspection before every new witness.
What this chapter does not say
The numbers 82, 19,683 and the pattern counts at five and six values are computed, not proved inside the system; they stand as orders of magnitude and carry no claim. The chain of thresholds at three values is likewise computed.
And: the expressions “locally classical” and “mediation” are not the same thing. The theorem says that a contexture-dependent choice is not buildable from local means. That such a choice is the formal locus of what Günther calls mediation is interpretation — grounded interpretation, but no theorem.
6 · The ascent
Günther takes polycontexturality as a world picture: reality as a system of infinitely expanding contextures with growing structural richness. A world picture is no subject for theorems, and the architecture does not claim it.
But two structural statements are contained in this determination, and those are subjects for theorems. One: it expands, and without end. The other: in doing so the structural richness grows. Both have had theorems for a short while now, and this chapter presents them.
The gap is inhabited everywhere
Chapter 5 says what is reachable. It does not say that the unreachable is also inhabited at every stage — that there is anything there at all that is locally classical and yet not buildable. Now a theorem says it: for every value count from four upwards such an operation exists [20].
The witness is a rule and not a single case: act as maximum exactly on the lowest elementary contexture and as minimum everywhere else. The same prescription at every stage. That the lowest is chosen is a chosen and not a found basis — any other pair would give another witness, and the corpus says so on the spot.
The witness travels along
Now the step from one stage to the next. One embeds the smaller set of values order-faithfully into the larger — the values keep their sequence, and one is added.
On the embedded region the rule of the larger stage agrees with that of the smaller [21]. The witness is thus not lost and need not be chosen anew; it is the same one, seen from further up.
This is the formal version of what Günther's stepwise progression means by “every stage
contains the previous one as a substructure” — here as a theorem and not as a programme. Beside
it the arithmetic of growth: each step adds exactly m new
elementary contextures when going from m to m+1 values
[22], and the image of the old stage does not reach the new element
[23]. What is added was not there before.
And yet the step does not carry the signature
Here comes the finding no one had reckoned with.
The embedding carries the order, it carries the lattice, it carries the witness. The reversal of order it never carries — at no place, for no value count [24]. The reason is elementary: the reversal reflects the values about the middle, and the middle shifts when a value is added.
One might hope a cleverer embedding heals this. The answer is a differential in the sense of chapter 4, and it comes out hard:
If the value count is odd, then not a single map into the next stage commutes with the reversal of order [25] — without any assumption of injectivity or monotonicity, over all maps whatsoever.
The argument is three lines long and therefore exhibitable. At an odd value count the reversal has a fixed point: the middle value, which is mapped to itself. A compatible map would have to send it to a fixed point of the next stage. The next stage has an even value count — and there the reversal has no fixed point. Hence there is no such map.
And the possibility half stands beside it: at an even value count there is indeed an order-faithful, reversal-compatible embedding — it leaves the gap exactly in the middle [26], [27] — and from four values upwards the witness ascends under it as well [28].
The stage step therefore fails to carry the full local means precisely when the value count is odd. The lattice ascends step by step, the reversal only in jumps of two. Put otherwise: the exchange relation is stage-relative.
What this means is interpretation, and let it be said: the ascent has no form that would close
it — the expansion is genuine because it cannot be caught up signature-faithfully. A
theorem about the non-existence of a limit object this is not, and none is claimed. About double
steps — from m to m+2 — the theorem says nothing, and that too stands so
in the corpus.
No finite stock of values suffices
The second structural statement — infinitely expanding — is redeemed not by a limit object but by a negative theorem about every finite stock of values.
The fillability bridge of chapter 1 generalises: a pattern can be
filled with k values if and only if all its signs lie below k
[29], boundary cases included. The way is the same as in the two-valued case,
only general: a sign at k or above forces k+1 pairwise distinct
values.
Now consider an infinite sequence of patterns whose signs grow beyond every bound. It follows
that for every k there is a place from which on the initial segment can no
longer be filled with k values [30]. And such a stream
exists — the simplest is the one in which every place brings a new sign
[31].
That is “infinitely expanding” in its most honest formal shape: not as a claim about an infinite, but as a statement about every finite. The evolutive direction of chapter 1 — the one that chooses — provably exceeds every stock from which it could choose.
The interpretive step is stated expressly: kenogram is not contexture. That a
k-fillable row is a “k-valued system” hangs on the bridge that supports
it, and is labelled interpretation in the corpus.
The richness grows, the reachable does not
The third ingredient has already fallen in chapter 5 and belongs here: the space of possibilities grows strictly from stage to stage [18], the reachable remains at every stage the same two [16], [19]. That is the quantitative shape of “growing structural richness”.
That this gradation is what Günther means is interpretation. What is proved are a correspondence, a cardinality, a set equality and two inequalities.
The concept, second stage
From these four results a concept can be formed that no longer speaks of one structure but of a family of structures. A family is called formally operationally polycontextural in ascent if
- every one of its members carries several overlapping contextures on which a locally classical but not buildable operation exists;
- the members are joined by embeddings under which this witness is preserved;
- the stock of locally classical operations grows strictly from member to member;
- at every member exactly the same two named operations remain reachable.
This is satisfied by the stage family of the ordered value sets from four values upwards. Each of the four clauses carries a theorem.
And what is expressly not claimed belongs to the concept and not in a footnote. Not claimed is a preservation of the full local means — the parity obstruction forbids it, which is why the second clause speaks of the witness and not of the signature. Not claimed is a limit object. Not claimed is the world picture.
Deliberately the concept has not been built as an object of its own in the library. A structure with four fields that no theorem requests would be a renaming of what is there — the same decision with which the project once already let a planned naming layer lapse.
What about the world picture
The totality — reality as a system of… — remains open, and is carried as open.
That is a result and not an arrears. A statement of totality is not a structural property; it says nothing about how something is built, but that everything is built so. What is structure in this determination now has theorems. The only formal path to totality would be a constitutive posit — one declares it a premise. That would not approximate the concept, it would posit it, and this path is not taken.
7 · Reach and limit
Where the bound bites
The characterisation of chapter 5 has a face of application, and it is narrower than the name polycontexturality suggests.
It bites where a linear scale of at least four levels is present and a policy is called for that decides classically on every neighbouring pair of levels but not everywhere alike. Release levels from blocked to privileged; source authority from note to statute; security levels; hierarchies of norms. In such cases the theorem says: a mixed policy is not assemblable from local, level-blind checkers — in whatever wiring and to whatever depth.
And it does not say that the policy is right. From non-assemblability follows no guarantee of security, legality, truth or completeness. What follows is a statement about architecture: at this place a mediating instance of its own must stand, and it cannot be compiled away.
Where the scale is missing, the claim falls. Several roles, several perspectives, several agents — that does not suffice. A transition graph of named roles without gradation is not carried by this bound; what remains provable there is reachability, and a type system achieves that too.
The height of the fall, and that it can be stated
What happens if one removes the order? Take the smallest lattice in which three elements stand side by side as equals — a bottom element, three incomparable above it, a top. On it there is an operation that acts classically on every comparable pair and is globally neither the infimum nor the supremum. It is not merely calculated but exhibited as a term and proved [32]:
(x ∧ y) ∨ ((x ∨ y) ∧ (¬x ∧ ¬y))
A lattice version of the characterisation is thereby not open but refuted. That is more than “we have not proved it”: exactly the pattern that the characterisation excludes on linear scales is assemblable here. The limit is no hole in the proof but a counterexample — and the breaking point is exactly the lattice shape an architecture of equal-ranking roles would have.
That the architecture can state its own height of fall is no admission but its strongest piece. It stands on order, not on plurality, and it can say by what one notices the difference.
And the theorem that supports the posit of chapter 2 belongs in the same context: on an incomparable pair even the minimum leaves the two-element set [8]. The same property that grounds the posit is the one whose absence makes the bound fall.
The three limits
First, mathematical. The characterisation lives on linearly graded carriers and falls on lattices of equal-ranking elements — refuted, as just shown, not unclarified.
Second, hermeneutic. That the two-element value set is Günther's elementary contexture is decided by no theorem. The posit stands on grounded and, since recently, theorem-supported soil — and remains a choice.
Third, in principle. Not the will as a function, but solely the that of enactment is no structural feature. There is no poorer model in which it is missing as a position, because it is not a position. Here the procedure declares its reach, as a statement about itself.
What is not built
The proemial relation — the core of Günther's theory — is not built as a whole. What is built is a section: the stage movement carries a section property and an asymmetry of fibres, and it does not collapse into an invertibility. That this asymmetry is Günther's type inversion is decided by no theorem — it is interpretation, and the corpus says so on the spot. Untouched are the proemial interlacing and the generalisation to arbitrary interchanges.
The compound contexture is present as a number series and is not built as a concept — though its exponent is a theorem as of this edition [17].
A limit object of the ascent is neither built nor claimed nor excluded. These three negations have three different reaches, and they are not the same.
To the counting of compound contextures belongs a note that lies in the matter itself: the
source gives a formula and a series beginning at one, and says at the same time that a compound
contexture requires at least three values. The two do not agree — the first compound contexture
is the three-valued one, the one belongs to the series of ontologies. The architecture counts the
elementary contextures of an m-valued situation as pairs from m values;
this is compatible with Günther's series, but it is a stipulation and not a
derivation.
The picture
A ground that counts repetition, whose normal form is built as a procedure and whose value-filling has two proved bridges. Logics whose separation is not posited but proved — in a form that withstands Günther's own basic case, because it does not deny the map but the path to it from within. A mode of proof that knows its hard half and delivers it. A measure of separation that has turned from an existence claim into a characterisation, and from computed numbers into theorems. An ascent that is everywhere inhabited, takes itself along, grows strictly — and provably has no form that would close it. A layer of application that says where it bites. And three limits, none of which is a defeat.
What is posited stands at the beginning; what is proved is counted; what is interpretation is called so.
Part B resolves the numerals: for each one the theorem, its axiom profile, its range of validity and its register row. There also stand the counting routes, the checking apparatus, and the list of errors this project has found in itself.
Part B · The apparatus — about this part
Part A sets out the shape. Part B supplies what it is to be checked against: for every numeral the carrier, its axiom profile, its range of validity; the mode of proof in full; the register of concepts; the safeguards with their figures; the instruments and the errors this project has found in itself; and the routes by which every number can be recomputed.
Figures of measurement hold at the commit named here — throughout
24f7fe8, unless a number expressly names another anchor. The running state stands in
the repository's README; where the two diverge, the move between them is recorded in the history.
This information stands here once and is not repeated per number.
The range of validity is measured on the import closures and not on the directory:
- aggregate — inside the closure of the default targets; the axiom gate checks it at build time;
- built along — reached by a default target but outside the gate closure: built, unguarded, ungated;
- call target — built only on its own call;
- frozen — shut down, with explicit thawing conditions.
The difference is not tidiness: proved means something else inside the aggregate than outside, because a machine barrier bites there.
1 · The carrier table
For every numeral from Part A the theorem that carries it. All profiles are measured with
#print axioms against the built environment and not taken over; “axiom-free” is the
output's own wording and not an empty profile.
1.1 The ground (Part A, chapter 1)
| Numeral | Carrier | Range | Profile |
|---|---|---|---|
| — | K.relabel — the canonisation as a procedure, not as a claim (definition) | aggregate | axiom-free |
| [1] | K.rgs_unique_of_pattern | aggregate | [propext, Quot.sound] |
| [2] | KM.samePattern_iff_pattern | aggregate | [propext, Classical.choice, Quot.sound] |
| [3] | KM.samePattern_iff_common_nf | aggregate | [propext, Classical.choice, Quot.sound] |
| [4] | KF.marksLeOne_iff_fillable | aggregate | [propext, Classical.choice, Quot.sound] |
| [5] | KF.card_rgs_four — fifteen patterns of length four | aggregate | [propext, Classical.choice, Quot.sound] |
| [6] | KF.card_rgs_four_fillable — eight of them two-valued fillable; with KF.fillable_card_lt_card as a strict inequality | aggregate | both [propext, Classical.choice, Quot.sound] |
On [2] and [3]. The two characterisations are not interchangeable: [3] hangs on a concrete carrier type, [2] is type-transverse. The corpus notes this on the spot.
On [5] and [6]. The number fifteen is the fourth Bell number. It is proved in the corpus and not taken from the literature — that is more expensive, and it is the reason why the statement about the excess (15 against 8) stands without foreign authority.
1.2 The contextures (chapter 2)
| Numeral | Carrier | Range | Profile |
|---|---|---|---|
| [7] | CO.three_contextures_overlap — the pairwise overlap as a counting statement | aggregate | [propext] |
| [8] | meet_leaves_incomparable — on an incomparable pair the infimum leaves the two-element set | aggregate | [propext] |
On [7] — the upgrade of this edition. The file header of the carrier expressly labels the overlap a cardinality theorem and not an interpretation. The third edition carried it, at both places where it occurs, as a construction prohibition and as a grounded choice respectively — the theorem did not appear in it. That is a loss of a mark in a positive statement, found while ordering the corpus by subject matter, and it is healed here.
1.3 Time (chapter 3)
| Numeral | Carrier | Range | Profile |
|---|---|---|---|
| [9] | Proemial.IntransitivityDifferential.no_cycle_in_strict_order | aggregate | axiom-free |
| [10] | cyc3_not_representable | aggregate | [propext] |
| — | cyc3_holds, cyc3_irrefl, cyc3_not_transitive — the witness itself | aggregate | all three [propext] |
| [11] | Proemial.NoUniformSwap.no_uniform_swap | aggregate | [propext, Classical.choice, Quot.sound] |
The full name with its module is no pedantry at [11]. A theorem of the same name lies in a call target, unguarded, as a historical record. Whoever names the name without the module leaves open which one he means — and the range of validity this part carries then runs empty. The choice component at [11] stems from the category machinery used and not from the proof itself; the path of the axiom into the term is not measured, and the corpus says so there.
1.4 The measure of the separation (chapter 5)
| Numeral | Carrier | Range | Profile |
|---|---|---|---|
| [12] | CV.card_locallyClassical at two values (2^1 = 2), together with CV.min_ne_max; the sharpness of the bound from CV.two_lt_card_locallyClassical | aggregate | count and sharpness [propext, Classical.choice, Quot.sound], min_ne_max [propext, Quot.sound] (see 1.7) |
| [13] | NUCB.four_of_eight_generatable — the choice-vector version; with NUCB.locally_classical_dichotomy as the operation version of the same classification | aggregate | both [propext, Quot.sound] |
| [14] | TCB.T_not_in_clone; with TCB.T_rejective and the departure from the contexture | aggregate | [propext, Quot.sound] and [propext] |
| [15] | QCB.locally_classical_in_clone_iff4 — the characterisation at four values; with QCB.two_of_sixtyfour_generatable as the choice-vector version (“exactly 2 of 64”) | aggregate | both [propext, Quot.sound] |
| [16] | GCB.locally_classical_in_clone_iff — for all m ≥ 4 | aggregate | [propext, Quot.sound] |
| [17] | CV.locallyClassicalEquiv and CV.card_locallyClassical | aggregate | [propext, Quot.sound] and [propext, Classical.choice, Quot.sound] (see 1.7) |
| [18] | CV.card_locallyClassical_lt | aggregate | [propext, Classical.choice, Quot.sound] (see 1.7) |
| [19] | CV.two_lt_card_locallyClassical | aggregate | [propext, Classical.choice, Quot.sound] (see 1.7) |
| — | NUCB.W_not_in_clone — the witness of non-uniformity | aggregate | [propext, Quot.sound] |
| — | GCB.constant_clone_min_or_max — the version with constant checkers | aggregate | [propext, Quot.sound] |
Among the carriers of this table, [16] is the theorem with the most measured
consumers. Six declarations of the aggregate reference it directly in the proof term:
the two applications (§1.8), the reachability half of the count
CV.clone_locallyClassical_eq, the step to the ∀m witness of the ascent
SAsc.w_not_in_clone, the regime threshold
RegimeThreshold.regime_threshold_at_four and the stage aggregation
StageAggregation.agg_nicht_erzeugbar; none rebuilds it. The number six is
measured — direct references in the proof term over the aggregate environment, state
24f7fe8; the non-rebuilding by must-not routes. The version with constant checkers
has the same shape with four consumers.
1.5 The ascent (chapter 6)
| Numeral | Carrier | Range | Profile |
|---|---|---|---|
| [20] | SAsc.exists_locally_classical_not_in_clone | aggregate | [propext, Quot.sound] |
| [21] | SAsc.w_castSucc | aggregate | [propext] |
| [22] | SAsc.choose_two_succ | aggregate | [propext] |
| [23] | SAsc.ascent_proper | aggregate | axiom-free |
| [24] | SPar.castSucc_negFin_ne | aggregate | [propext, Quot.sound] |
| [25] | SPar.odd_no_neg_compatible | aggregate | [propext, Quot.sound] |
| [26] | SPar.eSym_strictMono | aggregate | [propext, Quot.sound] |
| [27] | SPar.eSym_negFin | aggregate | [propext, Quot.sound] |
| [28] | SPar.w_eSym | aggregate | [propext, Quot.sound] |
| [29] | KU.marksLt_iff_fillable | aggregate | [propext, Classical.choice, Quot.sound] |
| [30] | KU.unbounded_not_fillable | aggregate | [propext, Classical.choice, Quot.sound] |
| [31] | KU.idRGSStream_not_fillable | aggregate | [propext, Classical.choice, Quot.sound] |
On the witness at [20]. The choice of the lowest elementary contexture as the max-island is a chosen basis, not a found one; any other pair would give another witness. The file header says so, and Part A says so too.
On [23]. The theorem on strict growth is axiom-free —
measured, not expected. Three carriers of this table bear the wording does not depend on any
axioms which #print axioms prints for axiom-free declarations: the canonisation
K.relabel, the intransitivity differential [9] and this theorem. It is the wording a
route with only one search pattern misses (§5.2).
On [25]. The proof runs via the fixed point of the order reversal and needs
neither injectivity nor monotonicity. About double steps — from m to
m+2 — it says nothing; that too stands in the corpus.
On [29] to [31]. With [31] the identity stream is a named object and thereby consumable; the anonymous illustration that stood elsewhere before was not, and still stands there.
1.6 Reach (chapter 7)
| Numeral | Carrier | Range | Profile |
|---|---|---|---|
| [32] | M3CloneWitness.m3_mixed_term_exists | aggregate | [propext, Quot.sound] — choice-free |
On [32] — the second upgrade of this edition. The third edition wrote that a lattice version of the characterisation “does not exist”. The corpus carries more: the term is exhibited and proved, so a lattice version is not open but refuted. The carrier is built on an inductive type of its own, to exclude any ambiguity of instances.
The clone counts of the flat lattice — 166 clone members, of which ten are locally classical, of which eight are neither infimum nor supremum — are computed, outside the corpus, and carry no claim.
1.7 A dividing line with a criterion
In the counting layer [12], [17], [18], [19] Classical.choice sits in the
finiteness machinery itself — the instance and the cardinality of the base set already
carry it; the transition constant between the layers stays free. Hence the dividing line that
stands in the file header:
Every statement in this file that names a cardinality carries
Classical.choice; none that does not, does.
That is a limit with a criterion instead of an enumeration — and it is measured, not inferred
(see §4.3). The profiles in the table above are the test of
it: the correspondence CV.locallyClassicalEquiv and the inequality
CV.min_ne_max name no cardinality and are choice-free; the count and the two counting
statements name one and carry it.
1.8 The layer of application
Two applications: a release policy over four authorisation levels, a source authority over four evidence levels. Both aggregate. The load-bearing claim is the measured consumption:
| Theorem | referenced in the proof term | |
|---|---|---|
PolicyCheck.freigabe_nicht_erzeugbar | GCB.locally_classical_in_clone_iff | yes |
RAGAuthority.autoritaet_nicht_erzeugbar | GCB.locally_classical_in_clone_iff | yes |
PolicyCheck.freigabe_nicht_erzeugbar_konstanten | GCB.constant_clone_min_or_max | yes |
RAGAuthority.autoritaet_nicht_erzeugbar_konstanten | GCB.constant_clone_min_or_max | yes |
PolicyCheck.freigabe_lokal (control) | the same target | no |
RAGAuthority.autoritaet_lokal (control) | the same target | no |
The control is the point. A route that reports yes everywhere measures nothing. This one discriminates, and it does so at the nearest non-hit — the prerequisite lemmas of the same files — and not at some arbitrarily distant counterpart.
2 · The mode of proof, in full
Part A presents the differential in its short form. Here stands what is obligatory about it.
2.1 The three marks
First: the poor model is built internally. A statement about “all orderings” or “everything buildable from the local means” must have its object inside the system. Claims in the metalanguage do not count, because they cannot be checked.
Second: the result is conditional. It stands relative to the posited premises — above all to the one from Part A, chapter 2. It does not compel the sceptic, it symmetrises him: from this can be reduced comes I posit the one encompassing logic, and that is itself a posit without precedence.
Third: what is proved is the form, never the naming. The file that carries the witness of non-uniformity is named after the formal property and not after the philosophical reading. This is no modesty but a precaution: a name that claims more than its theorem will be taken for the theorem at the next reading.
2.2 The provenance of the poor classes
Here, and not with the theorems, sits the burden of proof of the construction.
Canonical is the procedure: the concept of the buildable and its generation are standard clone theory and not invented for this purpose.
Chosen is the basis: these operations, this embedding. The objection that this choice is cut to fit is admissible. It is not answered but declared — and Part A, chapter 2 says why the choice has a substantive reason and where that reason is itself a theorem.
2.3 The differential, applied to the workshop
In the most recent working phase the form has been applied twice to the architecture itself.
To a theorem: the parity obstruction [25] has exactly the differential shape — the impossibility half over all maps without side conditions, the possibility half with an explicit witness [26], [27] beside it.
To a measurement: where an axiom profile is to be explained, the workshop has since required two versions differing in exactly one quantity. The four cases stand in §4.3.
3 · The ledger and the register
3.1 What the ledger achieves
Günther's concepts stand in a definitions edition of nineteen paragraphs. Which concept has
which formal carrier is machine-checked and stands in
docs/definition-ledger.md — not in this
paper. What stands here is the mechanism.
Measured: 91 rows, of which 75 with a carrier over 67 distinct names, 19 of 19 paragraphs covered; carrier statuses 57 theorems, 17 definitions, one posit, 16 open rows.
Five carriers appear in more than one row; the surplus rows are eight, and
75 − 67 = 8 is checked as an equation. This sentence stood wrong in
the self-report for five revisions — “eight carriers more than once” instead of “five carriers,
eight surplus rows” — and none of the rules could see it, because they read rows and not sentences
about rows.
3.2 The eight rules
Two bite at build time, six at lint time:
- R1, R2 (build): every named carrier resolves in the environment, and its carrier status matches the kind of its declaration.
- R3 no assignment status theorem — an assignment status is editorial and never compiler-checked; it must not call itself so.
- R4 carrier status open forces an empty carrier column.
- R5 all nineteen paragraphs represented.
- R6 carrier status theorem forces a guard entry.
- R7 every carrier row has exactly one matching reference in the reference file, and conversely.
- R8 every row number occurs exactly once in both files.
R3 to R8 set the exit code. The open rows stay open — among them the row on the world-picture concept, which now carries four references to the ascent probes and still no carrier. An open row is not a defect of the ledger but its achievement.
3.3 The register of posits
The load-bearing rule is not “the aggregate contains no posit” but no theorem of the aggregate hangs on a posit. A field of type true can enter no proof; why it must nevertheless be marked is an epistemic and not a logical argument: what stands in the aggregate reads as certified holdings.
Two classes are carried — placeholders with an exit criterion, constitutive posits with a justification — and one standing probe: no non-projection of the aggregate has the conclusion true, value zero, with a known counter-count on the projection side.
Measured: 32 posit fields, as an equation 30 placeholders + 2 constitutive; with zero fields of type Prop, over the same environment route.
Exactly two of them are constitutive. One of these was not found when the register of marks was first built, because its file header states its status in the language of the construction pattern and not in the language of the marks. The lesson stands in §5.2.
4 · The safeguards
4.1 The standing
All numbers at commit 24f7fe8.
| Quantity | Value | Range |
|---|---|---|
| constants checked (axiom gate) | 3255 | import closure of the aggregate, name-filtered |
| axiom guards, written | 487 in 60 files | corpus including the foreign-set case |
| axiom guards, enforced | 477 in 59 files | import closure of the default targets |
| theorems | 791 | main directory, widened route |
| build jobs | 1307 | the default targets together |
| declared gaps | 0 | whitelist empty |
The equation holds: 487 = 477 + 10, and the ten are the guards of the foreign-set case. Two independent routes arrive at the same decomposition — once over the directories, once over the resolvability of the names in the aggregate environment.
The gate reports:
AXIOM-GATE GRÜN — Aggregat Reformulation (3255 geprüfte Konstanten)
zieht kein sorryAx. Whitelist leer.
4.2 The four mechanisms
First, the axiom guards. For every load-bearing theorem the profile is measured and frozen as an expectation; if the proof basis shifts silently, the build breaks. The guard secures the basis, not the wording.
Second, the statement pins. A weakened theorem with the same profile would pass a guard. Pins nail down the full wording and break the build on definitional drift. Measured: 64 pins over 15 files.
Third, the global gate (§4.1).
Fourth, the separation of ranges. The gate's assurance holds for the aggregate and for no other range. That a module runs along with every build does not mean it is gated: the four probes on the proemial relation are built along, unguarded and ungated — a situation for which the third edition had no language.
4.3 Measured axiom provenance
Measuring a profile is cheap; saying where an axiom comes from is a second measurement. It has a form: two versions differing in exactly one quantity. Four cases:
- A monotonicity auxiliary of the library draws
Classical.choiceinto every consumer; avoiding it via the value route left four theorems choice-free. - The closure of a decision procedure is inherited by the users of a definition — measured at the differential against a hand-written version with an identical value and a different bound justification: same values, different profiles.
- The equality conversion on finite carriers draws
Classical.choicethrough an instance chain; the case-distinction version of the same correspondence is free — measured on the same file, with only the conversion exchanged. Without this measurement a choice profile on the correspondence would have looked plausible and would have been located wrongly. - In the counting layer the axiom sits in the finiteness machinery itself; hence the dividing line in §1.7.
What is measured are profiles and dividing lines. What is not measured is the path of the axiom into the term — the file headers say both, in that order.
4.4 Unguarded theorems inside the aggregate
A statement about guard density has a quantifiable reservation, and it stood nowhere until this edition.
While ordering the corpus by subject matter, theorems were found in aggregate modules whose module carries not a single guard, spread over three strands; outside these three every aggregate module with theorems carries at least one. Striking: they are exactly the three strands whose subjects occurred in no edition of the paper. This coincidence is measured, its cause is not — and without a route no explanation is offered here.
For the largest of the three — the layer on the interchange of form and content — the question has meanwhile been decided. Of 43 theorems only two lay in the closure of a guarded theorem; the exception for auxiliary lemmas therefore did not carry. In two stages 17 guards have been set, selected by the marks the corpus itself sets: four after the core named in the file header, eight after the module index, five after the concept the respective file carries in its title — these five expressly marked as discretionary, with the note that a later move on a better basis may reclassify.
Afterwards 21 of 43 lie in a closure, all nine modules carry at least one guard, and all 17 measured profiles are identical — the strand is profile-uniform. A third stage will not be built: the marks of the corpus are exhausted, and to switch a criterion because it has run out would be exactly the error the selection was built against. The remaining 22 theorems are recorded by name and with their source situation in an accepted report; they are unguarded and are called so. Trigger for a third stage: the first case in which one of them appears in a register row, in an edition of the paper, or as a prerequisite outside the strand.
The number, newly measured: 71 theorems in 23 modules without a guard, as an equation F3a–F3g 61 + F1 assignment layers 10 = 71 in 14 + 9 = 23 modules. The α+γ strand has dropped out entirely, because each of its nine modules now carries a guard.
And a correction to the old number which the repeated measurement found (§5.1, last rule): the same route yields, at today's state and with the guard set from before the guard tip, not 115 but 114. The difference is neither a route error nor a movement in the corpus, but a healed miscount: a prose line of a file header began at the left margin with a declaration word and was counted along by the theorem route; it has since been indented. The movement 114 → 71 belongs to the 17 guards, the movement 115 → 114 to the indentation — two causes that a single difference of 44 would otherwise have made into one.
4.5 What a foreign glance finds
Whoever searches the tree for the gap mark finds three numbers, and each answers a different question: 140 raw word occurrences across the whole tracked corpus, 97 in the source files, 27 affected declarations — all of them outside the aggregate. The third is the load-bearing one; the first two count prose along — freeze notes, doc strings, the phrase “free of it”.
This quantity counts itself, and the rule for it stands in §5.1. It is therefore also the one number of this paper that the running state in the README overtakes on publication day: archiving the third edition brings its occurrences a second time into the tracked corpus. The value above holds at the commit named, the running one stands in the README, and the move between them stands in the history.
The frozen branch stays frozen, with the same thawing conditions; green means compiles there, not gap-free. It carries 102 theorems and not a single guard — 12.7 per cent of the theorems of the corpus.
4.6 What the shape rests on
Ordering the corpus by subject matter produced a piece of information that no range table can give: the shape of Part A rests on 47 of the 167 modules, as an equation 47 + 120 = 167. Three of the seven layers are carried solely by the main directory on the proemial relation, two solely by the kenogrammatic layer.
This is no criticism of the rest. Probes, calibration, frozen preparatory work and assignment layers are legitimate and are marked as such in their headers. It is the statement of range belonging to the sentence “this architecture claims…”, and it has been missing for three editions.
5 · The instruments
Alongside the theorems something has arisen that knows nothing of Günther and could travel further: a reviewing practice for the formalisation of philosophical texts. Whoever formalises a thinker has a problem that this project has at every formalisation: the interpretation outruns the term. A theorem is built, its name is chosen, and the name claims more than the theorem. Against this gradient no resolutions help, only mechanisms.
5.1 The rules
The rule file of the corpus carries six marked rules, each from a measured misstep (measured over the route of the marked entry numbers). The most load-bearing:
A route carries a must and a must-not case, and its name boundary is a character class, not an assumption about the notation of the build. Violated three times in a row, each time differently; the sharpening of the second error was itself the third.
The stipulations of a specification are checked against each other, not only each on its own. Two individually correct stipulations can jointly force something neither demands — three times, at rising cost: first a tool exception, then a non-construction decision, then an axiom in a profile.
A statement of holdings is a pointer, not a list. Enumerating statements without a route go stale silently. In a single file nineteen identically built ones stood; they have been replaced by section pointers. The same healing at a heading, at a preamble, at a rule number — and, in the move of this edition, at the reporting line of the lint, which enumerated the frozen editions instead of naming their name pattern.
Claims about the reachability of a location are write probes, not inferences from empty search results. A missing result looks like an answer; twice it was none.
Whoever updates a self-counting number measures after writing. The gap number of §4.5 counts its own documentation along — and, at second order, the explanation of its movement as well. One commit carried, at its own place, the value that had held before the explaining sentence. The wrong value still stands in the history, with hash and reason; the error is worth more than its removal.
If a measurement is repeated after a change, it runs first with the seed set of the previous move and must reproduce its numbers exactly. The other rules check a result against an expectation; this one checks whether the route measures the same thing in two moves. A deviation would mean that two measurements are not comparable — and both would be plausible on their own. It struck at its first use: the number of unguarded theorems in §4.4 would otherwise have hidden two causes in one difference.
5.2 The pitfalls
The pitfall file carries fifteen entries (measured over the same route). Four of them concern not the proof language but the tool chain, and the two newest are measurement errors in their most unpleasant shape:
An empty result gives pause; a slightly too small one does not. An unquoted path list on the command line delivered zero for every search word — noticeable because for every one. An output command, by contrast, prints a different wording for axiom-free theorems than for all others; whoever searches only for the one finds 415 instead of 487 names. 72 guards of the corpus are axiom-free. A closure from 415 seed names can be computed, no intermediate figure looks odd, and it was found solely by the calibration of set size against set size.
And an output breaks at the line length, not at the name length. For the most frequent profile: line length equals name length plus 61. Measured: at 116 no break, at 123 a break; the interval is open in its middle, because no measuring point lies between. A broken expectation text is the correct form and no sloppiness — it is not smoothed, because the comparison is verbatim. And whoever sets a series by a template breaks the short ones.
Searching for the mark word finds no pattern speech. The register of
interpretation and posit marks was built from the marks that were known — and therefore failed to
find one of the two constitutive posits of the aggregate, because its header states the status in
the language of the construction pattern. It carries 281 entries in 74 files, measured at state
427b4b0 and not re-collected since.
At every extension of a reference work the set test applies, not the comparison of numbers. A first attempt lost nine entries, and the grown number looked good.
5.3 The lint
Three groups. The rank claims are read, not enforced — a breaking lint gets circumvented, a reporting one gets read. One group guards a conceptual relapse. The register group sets the exit code.
Measured at the generation state of this edition and not at commit
24f7fe8, because the edition itself moves them: (A) 58 hits in the running corpus and 10 in the frozen editions, (B) 0, (C) without violation.
A side use that costs nothing: an unchanged rank figure after a text move is at the same time the control that the new text carries no rank phrase. An existing instrument as a check for a move it was not built for.
5.4 The list of errors
The evidence that the review layers bite is this list — and it is continued, not shortened. From the working phase that led to this edition:
- A self-report carried “eight carriers more than once” for five revisions, where five carriers produce eight surplus rows — two quantities with the same value, and the prose attached the name to the wrong one.
- A specification prescribed a conversion form and at the same time checked its named fallback route to zero; the axiom in the profile was thereby set by the specification and not by the proof. The third, unnamed route was free.
- A statement of location (“reachable only by hand”) rested on an empty search result; the write probe refuted it.
- An explanation for a difference between two numbers stood as a finding, although only the difference was measured and not its cause. Two numbers with different anchors are not thereby rivals.
- A number stood twice in the same document and was corrected only once; the second was an estimate without a route and has been struck.
- One instruction demanded two things in one sentence — “immediately at the theorem” and “form as in the rest of the corpus” — and the two were incompatible.
- A first computation run produced a plausible all-false from a symbol collision; the hand calculation disagreed, the run was discarded and redone.
- A negation about the project's own corpus — “not formalised” — stood verbatim in three editions and contradicted the structure table of the same document.
- And in this move: an expectation about a number that meant a different quantity from the one the route measures. Expected was a decrease by 17 — that many guards had been set — measured is a decrease by 43, because the route counts theorems in guard-free modules and not theorems without a guard of their own. The expectation was the wrong one, not the route; it was found by the rule that makes the previous state be reproduced first.
None of these errors was found by the instance that made it.
5.5 What this edition has learnt methodically
The third edition explained its reticence by an asymmetric burden of proof: against overstatement there is an apparatus, against understatement nothing, so the text drifts downward. This explanation is half refuted. A review of the negations found not only statements that were too weak, but also one that was too strong — a “provably” in the closing picture where the carrier expressly says interpretation.
What explains both directions: a mark set by the carrier is lost on the way into the paper. From “carries no decidability instance” comes “is not decidable”; from “interpretation” comes “provably”. And the loss does not hit negations only: the overlap [7] is a cardinality theorem in its carrier's header and stood in the paper as a construction prohibition.
Hence the obligation under which Part A is written, and it has three limbs: every negation carries a route as a rank claim does; every upgrade carries a measurement; and where the carrier sets a mark, the paper carries it along.
A further distinction has arisen in the process which the corpus did not have before. A negation claims either that something does not exist, or that this project does not build it, does not claim it, does not exclude it. The first is a statement about the world and needs checking; the second is one about the project's own self-binding and can be looked up in file headers. The sentence about the limit object looks like the first and is the second.
6 · Appendix: counting routes
So that the numbers are recomputable and not merely credible, each carries its route. All
entries hold at commit 24f7fe8 unless anchored otherwise.
| Quantity | Value | Route |
|---|---|---|
| constants checked | 3255 | build output of the gate |
| guards written | 487 / 60 files | search for the guard pattern over both directories |
| guards enforced | 477 / 59 | the same route, restricted to the import closure of the default targets; equation 487 = 477 + 10 |
| theorems | 791 | widened theorem route over the main directory |
| build jobs | 1307 | build output, not summed |
| statement pins | 64 / 15 files | line mark at the pin, counted over the main directory |
| gap mark | 140 / 97 / 27 | word boundary, occurrences rather than lines, over the tracked files; declarations per file — self-counting, see §4.5 |
| frozen branch | 102 theorems, 0 guards | structure and guard route per directory |
| unguarded aggregate theorems | 71 in 23 modules | theorem route against guard route per module, over the import closure of the aggregate; as an equation 61 + 10 = 71 |
| posit fields | 32 = 30 + 2 | environment route over the structure fields; zero fields of type Prop |
| module partition | 112 + 20 + 34 + 1 = 167 | import closures of the targets, checked as an equation |
| carriers of the shape | 47 + 120 = 167 | assignment of the subjects to the layers |
| axiom-free guards | 72 | expectation-text route, measured at c3b28cb — not the profile route |
| ledger | 91 / 75 / 67 / 19 of 19 | lint group (C); surplus 75 − 67 = 8 as an equation |
| register of marks | 281 entries / 74 files | sieves over the file headers, measured at state 427b4b0, not re-collected since |
| locally classical operations | 2^C(m,2), every m | proved [17]; 8, 64, 1,024, 32,768 are instances |
| of these buildable | 2 from m ≥ 4 · 4 of 8 at m = 3 · 2 of 2 at m = 2 | proved [16], [13], [12] |
| patterns of length 4 | 15, of these 8 two-valued fillable | proved [5], [6] |
| reversal-compatible embeddings | none at odd m · present at even · 2 / 8 / 48 at m = 2 / 4 / 6 | non-existence and existence proved [25], [26]; the counts computed |
| clone fragment over three values | 82 of 19,683 | computed, outside the corpus |
| flat lattice | 166 / 10 / 8 | computed; the existence form proved [32] |
Four rules on counting. Counted over the shell, not with a library function that skips blank lines. Aggregate and corpus figures kept apart. Build outputs of several targets never summed. And: a number standing next to a related one is checked against it as an equation — twice exactly that found a standing error.