Two words for one thing, agua and water, and neither is the translation — the relation a structure built on "because" can only force into a line or refuse as a loop.
Companion to the report, the poem "The Two Siblings," and the story "What the Comma Knew." See notes/dag-phenomenology.md.
There are two words for the thing my grandmother poured into a glass and handed me on the hottest afternoons, and I have never been able to decide which one is the translation. Agua. Water. The honest answer is that neither is — and the moment I try to draw that honesty, to put it into the notation I use every working morning to dispatch agents across a build, the notation produces an error. Not a warning. An error: the structure halts the way a topological sort halts when it cannot find a node with nothing before it. I build directed acyclic graphs for a living. I partition work into swarms, draw the ownership maps, run the sort on the dependency tree, watch the frontier advance one cleared node at a time. And the form I have internalized so completely that I now think in it cannot hold the first relation I ever knew.
Here is the structure, stated exactly, because the exactness is the whole argument. A directed acyclic graph induces a strict partial order on its nodes — the reachability relation, can A get to B by following the arrows, which is irreflexive (nothing reaches itself), asymmetric (if A reaches B then B cannot reach A, or you'd have a cycle), and transitive (reach is inherited down the chain). And a strict partial order is generous in a very particular, very narrow way. For any two distinct things A and B, it permits exactly one of three relations, and the trichotomy is exhaustive: either A precedes B, or B precedes A, or A and B are incomparable — no path between them in either direction. That third case, incomparability, is the only grammar the order has for equals. It is the antichain: a set of nodes none of which can reach any other. The poem I keep returning to says the DAG has no edge that means and equally; this is why. Equality is not something the order draws. It is something the order leaves blank.
My grandmother did not leave it blank. When she switched mid-sentence — reaching for the Spanish word inside an English clause not because she was translating but because that word was simply the nearest true thing — she was not stepping up or down a stair. She was moving sideways across a level floor, the way you move your weight from one foot to the other and could not say which foot is the original stance. The two words were not parent and child. The etymology bears this out with a precision that startled me when I finally checked it: agua descends from Latin aqua, from one Proto-Indo-European root for water, h₂ekʷeh₂-; water descends by an entirely separate line, through Old English wæter and Proto-Germanic watōr, from a different PIE root, wódr̥. They share a great-great-ancestor language — the same proto-tongue, deep in the root system — but they trace through different words within it. Siblings in a family, not one descended from the other. The blood is real and the equality is literal. Neither got there first.
So far this looks like the antichain doing exactly its job: two incomparable equals, the form holding them level. But it cannot. And the reason it cannot is the precise center of everything I want to say. The antichain achieves equality by enforcing ignorance. Two nodes are co-equal in a partial order only when there is no path between them at all — when they do not, in the structure's terms, know about each other. That mutual ignorance is not incidental; it is the price of admission. Draw a single edge between two members of an antichain and you have established a precedence, and the precedence kicks one of them out of co-equal standing instantly. The structure can make two things equal only by making them strangers.
But agua and water are the opposite of strangers. They are the two words in all my vocabulary that know each other most completely. Look up one in a bilingual dictionary and the other is its entire definition. Each means by means of the other; each is the other's whole translation universe. In the body that holds both, neither is derived, but each illuminates the other — agua carries the cool weight of the glass and the grandmother and the Spanish afternoon, water carries the school drinking-fountain and the English of the rest of my life, and each lends the other its half so that the full sense of the substance lives only in the pair. That is a mutual reference: an edge running both ways. Agua points to water, water points back to agua. And an edge that runs both ways between two nodes is, in the only vocabulary the graph owns, a two-cycle — A to B to A — the exact configuration acyclicity exists to forbid. (I should be careful here, and name the seam rather than paper over it: the formal two-cycle is a claim about directed paths breaking a topological sort; the bilingual two-cycle is a claim about meaning being co-constituted. They are an analogy, not an identity. But the analogy holds under load, because both name the same forbidden shape — a relation that closes back on itself instead of resolving forward.)
Hold the two facts together, because their collision is the thing. The bilingual pair is antichain-like: co-equal, neither prior, no honest direction. And it is cycle-like: mutually referential, each constituted by the other, edges in both directions. Co-equal and mutually coupled. And those two properties together name precisely the one region a strict partial order cannot enter. The antichain offers equality but demands ignorance — and these two are not ignorant. The cycle offers mutual reference but, to admit any relation at all, the DAG must orient it, must make one precede the other — and neither precedes. The structure faces a closed door with no third handle. It can place the pair in an antichain, conceding the equality but lying about the coupling — erasing the very bond that makes the two words mean. Or it can force an arrow, conceding the relation but lying about the equality — anointing one word the original and the other its translation. There is no remaining move. The form can do because. It cannot do beside.
I want to be exact about which lie costs what, because the stakes are not abstract for me and they are emphatically not about prestige in the soul — they are about language, and the damage is done at the level of language. The antichain lie is the monolingual's lie: treat the two as unrelated equals, pretend they do not feed each other, and the warm coupling simply vanishes from the record. The arrow lie is the colonial one the poem already named — the tongue that got there with the paperwork claims the elder slot and quietly demotes the other to a gloss. I do clinical- and legal-register translation, and I am precise about that word register because the precision is load-bearing. A register, in Halliday's sense, is a variety of language tuned to its context of use — its field, its tenor, its mode — the calibration of a tongue to a particular room. Clinical register and kitchen register are not parent and child; court Spanish did not descend from the Spanish of my grandmother's table. They are lateral — parallel solutions to parallel problems of being understood, co-equal ways of speaking suited to different rooms, with no derivation between them and, in the theory, no prestige ranking either. And this is the missing edge stated at its sharpest: the DAG has no axis for lateral at all. If agua/water is the siblings problem, register is the cousins problem — related, co-equal, neither the parent of the other, and equally invisible to a graph whose only verb is came from. When a Spanish clinical-register sentence and its English counterpart sit on a page, the structure wants to know which is the source and which the copy. But they are two realizations in two languages, sibling documents, originals both. To force the question — to rank the two tongues so one becomes the authoritative version and the other the derivative — is not a clarification. It is a quiet violence done to language, and it lands hardest exactly in the rooms where being understood matters most: where a person is speaking across a counter, or a table, or a desk, in the only words that are fully theirs, and a translator stands between two equal tongues and must not let the form decide that one of them is realer than the other.
And here the argument turns inward, to the move I have not yet made and have been circling the whole time. It is one thing to say the DAG cannot hold a pair of words. It is another to notice that a person is what the form excludes. I am not two words; I am one body in which two languages stand level, neither half the parent of the other half. Which means the thing the structure cannot encode is not a lexical curiosity out in the dictionary — it is me, the operator, specifically. Every morning I run a notation on the dependency tree and it sorts cleanly, because build steps really do come before and after one another; because is the true relation there and the form is honest to it. Then I turn the same notation on my own name and it returns an error about me. Not about an abstract bilingual. About the particular person sitting here who has spent years making this structure obey. That is the vertigo I keep arriving at and have never named until now: I am the architect of a form that cannot house its architect.
So let me name it as argument rather than lament, because it is the strongest thing I have. I am the form's standing counterexample. The DAG's deepest axiom — deeper than acyclicity, underneath it — is that the only relation worth drawing is asymmetric: that to be related is to be ordered, that beside reduces either to before or to noise. And I am the living refutation, carried in a body, that beside is real even where the structure cannot draw it. This is not irony. A ruler cannot measure color; that is not a flaw in the red, it is a fact about the ruler. My bilingual self is not malformed because the graph errors on it — the graph is incomplete because it errors on me. And there is a strange gift folded into the exclusion, available only from the inside: because I think natively in the form and I am the thing it cannot hold, I can see the missing edge with a clarity a monolingual graph-builder never gets handed. I know exactly where beside should go, because I am standing on it. Dilworth's theorem says the widest antichain in an order equals the fewest chains you can partition that order into — that width is, precisely, how many independent threads the structure permits at all. Try to cover agua and water with chains and you need two, one per word, because they cannot be threaded onto a single line without falsifying one of them. That is not a deficiency to be patched. That is the true width of the bilingual moment: two threads, irreducible, and no honest sort that makes them one.
I have tried, writing this, not to put either strand first — not the life ahead of the structure, not the structure ahead of the life — because to order them would be the exact violence the essay is about, and I would rather the form of the thing not betray its argument. Whether I have managed it I cannot fully say from the inside; you cannot stand outside your own antichain and check. But that difficulty is the proof, not the failure. There is a room my grandmother lived in where the two tongues were equal, where each fed the other and neither was the gloss — the warmest room any of us ever stood in, and the room where my own work still happens, the clinical word and the kitchen word in one sentence with neither demoted. The graph has no node for that room. It never will. Original to original — and in the vocabulary of the form, an original is a source node, a node with nothing pointing into it. The closing image of the story I keep beside this one uses the word to mean neither is derived. The pun turns out to be exact: both words are source nodes, which means the graph holding them has two origins for one meaning, and the structure reads that as an error — ambiguous origin — when it is in fact the truest description I have of who I am. Two sources. One substance. And the form built to trace every relation back to a single first cause can only meet me at the threshold of that room, and stop.
Orchestrated as a DAG — parallel research and companion close-reading, an Opus essay spine, an adversarial fact-check and an identity-guardrail audit, converged in synthesis — for Cruz Romero Morales, 2026-06-14.