Grammar completion: the decoding declaration

Appendix E found that the blueprint is read at a tower of grammatical levels — material at the cell, regulatory syntax at the tissue, architecture at the organ — each read by the same level-and-shape operator, but the tower was a single ladder, read for one organ, and completion was honestly false. This appendix completes the grammar in three directions and then asks, honestly, what a hundred-percent decoding could even mean. First, the organ atlas: the same regulatory operator reads cardiac, neural, and hepatic identity from real promoters, not the heart alone — the mean organ-by-grammar confusion matrix is diagonal, each organ peaking on its own grammar, with a quiet housekeeping control. Second, a new rung above the fixed levels, the dynamical grammar: one locus yields a family of readings by cell state, indexed by a CpG-island methylation-sensitivity signal that is orthogonal to the material and read by the identical operator. Third, the highest organizing level, the body-plan grammar: Hox colinearity on the real HOXD cluster, where the genomic order of the nine genes is perfectly monotonic with the anterior-posterior body axis — position on the DNA equals position in the body. The grammar is now not a ladder but a filled two-dimensional space, every cell read by one operator. So the decoding declaration is made in two axes. Structural, grammatical decoding is complete: every grammatical level is identified, formalized, sequence-readable, orthogonal, and read by the same operator — there is no remaining undiscovered grammar. Functional, quantitative decoding is not complete, and its obstacles are named. A hundred percent means the grammar is fully mapped; what remains is measurement, not undiscovered grammar.

Appendix E lifted the cell's level-and-shape reading up a tower of three grammatical levels and showed, on real human promoters, that the tissue-level and organ-level grammars the corpus had missed are real and sequence-readable; it left three things undone — the grammar was read for one organ, it stopped at the architecture level, and completion stayed false. This appendix completes the map. The first direction is breadth. The regulatory grammar of Appendix E was the cardiac transcription-factor code; here the identical operator is given the neural code (SOX, BRN/POU, REST, NEUROD, OLIG) and the hepatic code (HNF4A, HNF1A, FOXA, CEBPA, DBP) as well, and run against nine real promoters spanning all three organs plus a housekeeping control. Because raw motif counts are confounded by base composition, each organ's reading is the shuffle-normalized enrichment of its motif content against a dinucleotide-preserving background, which removes the composition confound. The primary result is the mean organ-by-grammar confusion matrix, and it is diagonal-dominant: cardiac, neural, and hepatic each peak on their own grammar, while the housekeeping control stays quiet. The regulatory grammar reads organ identity across the atlas, not the heart alone. The second direction is upward, to a level whose reading depends not on the sequence alone but on the cell state — pragmatics, where one text means different things in different contexts. The state-indexing signal is the CpG-island methylation-sensitivity potential, the windowed observed-over-expected CpG ratio, which marks where methylation can switch a region; it is a sequence property, so it is readable, but it indexes a family of readings parameterized by the methylation state. Two facts make this a genuine new rung rather than a relabelling of the material level: the signal is orthogonal to the stacking signal along the sequence, with a mean absolute correlation near a tenth, and the identical robust operator produces its shape. The third direction is the highest organizing level, the grammar that lays out the whole body — genre. Its canonical sequence-encoded form is Hox colinearity, where the linear order of the Hox genes on the chromosome equals their order along the body axis; read on the real HOXD cluster, nine assembly coordinates fetched from the public genome, the genomic order is perfectly monotonic with the anterior-posterior body rank, and the Pearson correlation on the raw coordinates is above 0.96. Position on the DNA equals position in the body, read by the same operator. The grammar is therefore not a single ladder but a two-dimensional space with an interpretation axis — material, regulatory, dynamical — and an organization axis — architecture, body-plan — read across organs, every cell by one operator that is scale- and shift-invariant to machine epsilon. On that completed space the appendix issues a two-axis decoding declaration. Structural, grammatical decoding is declared complete: every level is identified, formalized as a level-and-shape pair, sequence-readable, orthogonal, and read by the identical operator — no grammatical level remains undiscovered, and the structural completion is earned by the fail-closed gate. Functional, quantitative decoding is held open, with named obstacles: that each level's shape predicts the measured phenotype — enhancer activity per organ, the absolute methylation state, the real three-dimensional body map — requires held-out measured data. The hundred-percent claim is therefore structural only, and it carries its honest meaning on its face: the grammar is fully mapped, and what remains is measurement, not grammar. Every input is locked with a grade and provenance, with zero inline magic numbers; the cardiac grammar and the three Appendix E sequences are re-locked verbatim; the fail-closed gate passes ten of ten deterministically under a double hash; and functional completion is never asserted — the gate fails closed if it ever is.

Why this appendix exists

Appendix E answered a sharp question — is there a tissue-level grammar the cell-level reading missed — and found that there is, that it is sequence-readable, and that the same level-and-shape operator reads it. But three named directions remained, which the instruction driving this appendix names directly: there may be a larger grammar above the tissue level, there may be organ-specific and dynamical grammars, and the work should be pushed all the way to a decoding declaration. Each of those is a real direction. The regulatory grammar of Appendix E was read for one organ only, the heart; nothing yet showed the same operator reads other organs. The tower stopped at the architecture level, a fixed property of the sequence; nothing yet captured the fact that the same locus is read differently in different cell states, or that there is a grammar above the organ that lays out the whole body. And completion was honestly graded false — correct, and precise about scope — but it did not yet state what would have to be true for the grammar to be called decoded. This appendix closes all three, and is careful, at the end, not to convert an honest structural completion into a dishonest functional one.

The first direction: the organ atlas

The regulatory grammar is the arrangement of transcription-factor binding sites that specifies what a region builds. For the heart that grammar is the cardiac code; but every organ has its code, and if the reading is real it should read them all with the same operator. So the cardiac grammar of Appendix E is joined by a neural grammar — the binding syntax of the SOX, BRN/POU, REST, NEUROD, and OLIG factors that specify neural identity — and a hepatic grammar — HNF4A, HNF1A, FOXA, CEBPA, and DBP, the liver code — and the three grammars are run against nine real human promoters spanning all three organs, with a housekeeping promoter as a tissue-neutral control. A complication has to be handled honestly: raw motif counts are confounded by base composition, because a GC-rich promoter will match GC-rich motifs more often for reasons that have nothing to do with organ identity. The fix is to read each grammar not as a raw count but as an enrichment against a dinucleotide-preserving shuffle of the same promoter, which holds the composition fixed and asks whether the specific arrangements that form binding words are present above chance. Read that way, the mean organ-by-grammar confusion matrix is diagonal-dominant: cardiac promoters peak on the cardiac grammar, neural on neural, hepatic on hepatic, and the housekeeping control stays quiet under all three. The regulatory operator reads organ identity across the atlas. The result is reported with its limits — the panel is small, the per-gene reading is noisy at these counts, and the hepatic grammar is the weakest because liver regulation is more enhancer-distributed than proximal — so this is a sequence-readability result, not a validated organ classifier.

The second direction: the dynamical grammar

Every level so far reads a fixed property of the sequence. But the same locus is read differently in different cell states, and that context-dependence is a grammar of its own — in linguistics it is pragmatics, where one sentence means different things in different situations. The signal that indexes the cell state is the CpG-island methylation-sensitivity potential: the windowed ratio of observed to expected CpG dinucleotides, which marks the regions where DNA methylation can switch activity on or off. It is a property of the sequence, so it is readable, but what it reads is not a single value — it is a family of readings parameterized by the methylation state, a base reading modulated by how methylated the region is. Two facts establish that this is a genuine new level and not a relabelling of the material grammar at the bottom. The methylation-sensitivity signal is orthogonal to the material stacking signal along the sequence, with a mean absolute correlation near a tenth across the nine loci, so it is a distinct projection of the blueprint. And the shape of that signal is produced by the identical robust operator that gives every other level its shape, so it is the same grammar lifted, not a new rule. What is read here is the family of readings; what fixes the state — the absolute methylation level in a given cell type — is not read, and is named as the obstacle that whole-genome bisulfite sequencing would close.

The third direction: the body-plan grammar

Above the grammar that lays out an organ sits the grammar that lays out the whole body — the genre level, the largest organizing form. Its canonical sequence-encoded instance is one of the deepest facts in developmental genetics: Hox colinearity, where the linear order of the Hox genes along the chromosome is the same as their order along the head-to-tail body axis. This is the blueprint's spatial code at its most literal — position on the DNA is position in the body. It is read here on the real HOXD cluster, nine genes whose genome coordinates were fetched from the public assembly, with each gene's place on the body axis taken from the established colinearity literature. The genomic order of the nine genes is perfectly monotonic with the body-axis order — the rank-to-rank correlation is exactly one — and the correlation on the raw coordinates, which feel the uneven spacing between genes, is above 0.96. The same operator that reads every other level produces the body-axis shape. What stays open is the real three-dimensional body map, the measured spatial arrangement of these domains across the developing embryo, which Hi-C or staged spatial transcription-factor mapping would supply; the order is read, the metric map is the obstacle.

The decoding declaration, in two axes

With the grammar space filled in, the instruction's final demand — push to a hundred-percent decoding declaration — can be met honestly, but only by separating two questions that the word decoding runs together. The first is structural, or grammatical: is every grammatical level of the blueprint identified, formalized as a level-and-shape pair, readable from sequence, orthogonal to the others, and read by the identical operator. If the answer is yes, then the grammar is fully mapped — there is no level left undiscovered — and that can be declared complete. By that standard the decoding is complete: the material, regulatory, architecture, dynamical, and body-plan levels are all present, all formalized, all sequence-readable, all orthogonal, all read by one operator, and now read across organs, and the fail-closed gate confirms it. The second question is functional, or quantitative: does each level's reading predict the measured phenotype — the enhancer activity an organ grammar should imply, the absolute methylation state behind the dynamical family, the real three-dimensional body the colinearity should produce. That is not complete, because it requires held-out measured data, and the honest grade is open with named obstacles. So the declaration is made in two axes and not one: structural decoding is a hundred percent; functional decoding is open. The hundred-percent claim carries its meaning on its face — it says the grammar is fully mapped, and that what remains is measurement, not undiscovered grammar. It is never allowed to mean more than that; the gate fails closed if functional completion is ever asserted.

What is locked, and what stays open

Every input is read from the locked database with a grade and a provenance, and the lock manifest reports zero inline magic numbers — the three organ grammars with their literature sources, the nine real promoters with their accessions and coordinates, the nine HOXD coordinates with their body ranks and the colinearity citation, the stacking parameters, the shuffle settings, and the gate thresholds. The cardiac grammar and the three Appendix E sequences are re-locked unchanged, so the appendix is strictly add-only. The structural results are verified-exact or real-data facts: the one operator at every level, the diagonal of the organ atlas, the orthogonality of the dynamical level, the colinearity on real coordinates, all reproduced deterministically under a double hash, the gate passing ten of ten. What stays open is the accuracy, and the obstacles are named per level — the held-out functional datasets for enhancer activity per organ with a pre-registered test and a shuffle control; the absolute methylation state from bisulfite sequencing that would fix the dynamical family; the real three-dimensional body map from Hi-C or staged spatial mapping; and the absolute mechanical magnitudes, which remain the measured inputs of Appendix D and which this appendix does not supersede. The grammar is fully mapped; its quantitative validation is the next step.