00 -- Orientation: The Five Frames

June 30, 2026 · View on GitHub

Status of this document. This is a primer, and like everything in this curation it is a derivation you are meant to re-derive, not a warrant you are meant to accept. Nothing below is true because a corpus, a thesis, a tradition, or a maintainer asserts it; it is offered as argument that either holds when you push on it or does not. Where it points to research/, it points there as further reading -- the lineage of an idea -- never as the reason to believe the idea. "Because the corpus says so" is, in this project, a defect, not a citation. If this primer and SPEC.md ever disagree, SPEC.md governs and this primer is wrong. It assumes no dissertation and no prior reading: a person who knows only SPEC.md (the closed verdict lattice, the six boundaries, re-derivability) should be able to follow the whole of it, and the terms it uses in italics are defined in ./GLOSSARY.md.


Why a primer at all

EMET is a small program with an unusually opinionated shape. It judges bytes and emits exactly one of three verdicts -- MATCH, DRIFT, UNVERIFIABLE -- and it refuses, structurally, to emit a fourth one meaning trusted. It will not edit, sign, or revert a target. It will not take a model-safety question as input. It insists it cannot be its own root of trust. To an engineer these look like tasteful restraint: a careful tool, modestly scoped. They are not restraint. Each is forced -- not chosen for elegance but driven by a specific argument, such that relaxing it would change not how EMET behaves but what EMET is.

This document lays out the five frames those arguments stand on, plainly and without the apparatus, so the numbered essays that follow can do the load-bearing work without re-explaining their own foundations. A frame here is a posture toward a question -- a way of seeing -- not yet the full derivation; the derivation lives in the essay each frame points to. Read this once and the rest of the curation is a sequence of consequences.

One discipline runs through all five and is worth stating before any of them: the warrant is the argument, never the source. A rationale that justified EMET's no-authority design by appeal to the corpus's authority would refute itself -- it would be doing in band the exact thing EMET's refuse command exists to strip out. So every frame below is offered to be re-derived and, where it leans on a contested philosophical position, the lean is confidence-labelled so you can weigh it rather than swallow it.


Frame 1 -- No-aseity: nothing stands on its own footing

Aseity (Latin a se, "from itself") is the property of existing entirely on one's own footing, owing one's being to nothing else. The first frame is the denial that anything in EMET's domain has it. Trust, in particular, has no svabhāva -- no own-being, no intrinsic standing it carries in itself. To exist, on this view, is to be conferred, relationally, by something other than oneself; nothing is its own ground.

The engineering consequence is the closed verdict lattice (SPEC §2). Because trust has no own-being, there is no fact about a signal that the signal possesses by itself and that EMET could read off and report as TRUSTED. So the lattice has no such inhabitant: it can report that a re-derivation agreed (MATCH), disagreed (DRIFT), or could not be performed (UNVERIFIABLE), and it has no fourth thing to say. The closure is not a coding-style preference; it is the type-level encoding of no-aseity.

The objection this frame must survive is that it proves too much. Objection: if nothing is self-standing, then nothing is trustworthy, and a verifier built on that must collapse into "everything is UNVERIFIABLE" -- it would say nothing at all. Answer: conferral-dependence is not non-existence. Hard nihilism ("nothing is real") is self-refuting and is not what is claimed; no-aseity says only that what is real is real relationally, conferred rather than self-grounded. A MATCH is fully real -- and fully conferred, produced by an act of re-derivation against an anchor the operator authorized. "No TRUSTED" is a claim about which verdicts exist, not a claim that no verdict is ever true. The frame would be refuted by any codepath that emitted a verdict outside the lattice, or any value that asserted authority. (Provenance, further reading only: Nāgārjuna's emptiness and the denial of svabhāva; Jan Westerhoff, The Non-Existence of the Real World; Aquinas on esse ab alio, being-from-another. The Nāgārjuna and Westerhoff readings here are moderate-confidence paraphrases of contested positions, given as lineage; the argument stands on its own re-derivation, not on their authority. Full derivation: ./02-no-aseity.md.)


Frame 2 -- The is/ought seam: authentication is a fact, authorization is authored

This is the headline frame, and the crux the whole curation turns on. Two utterly different questions can be asked of a signal arriving at a boundary, and everything depends on not confusing them.

The first is authentication: is this the genuine, bit-intact thing it presents itself as? This is a question about the signal, answerable off the signal -- you read the bytes, you hash them, you compare. It is, literally, a 1 or a 0: the bytes re-derive the anchor or they do not. There is a fact of the matter.

The second is authorization: ought this authentic signal cross? May this command act; may this file load; is admitting it the operator's uncoerced will, now? And the pivot of the entire frame is that this is not a property of the signal. No amount of inspecting bytes tells you whether they should be admitted, because "should be admitted" is a fact about a will and a policy standing on the near side of the boundary, not a fact carried across by the data. A signal can be perfectly authentic and still be one that must be refused. The seam between the two is the line EMET locates and refuses to launder across: it makes the authentication-grade decision (MATCH/DRIFT/UNVERIFIABLE) and declines to author the authorization. The verdict it emits is a fact; it is never an authority -- that is Boundary 1 (SPEC §6.1), and refuse (SPEC §4) is the same seam enforced where a signal tries to carry its own permission in band.

A precision this frame insists on, because skipping it is how the idea becomes hand-waving: the claim is not "this is Hume's guillotine." A policy-relative permission really is derivable from is-facts plus the further descriptive fact that a policy is in force -- that is just Searle's counts-as. What the seam instantiates is the weaker, more defensible autonomy of the deontic (Pigden's reading of Hume's law): a valid inference introduces no new non-logical vocabulary, so the deontic words -- ought, may, permitted -- must be introduced by a premise that already contains them. That premise is authored, by the operator, at the act of stipulating a policy; EMET supplies the descriptive premise and declines to author the deontic one.

Objection (the hardest, from object-capability systems): possessing the token IS authorization -- holding the unforgeable reference simply is being permitted, so the is/ought split collapses. Answer: the seam does not vanish, it relocates. A capability is a materialized grant -- an ought made bearer-portable, the frozen record of a prior authored conferral at issuance. Presenting it redeems a permission already conferred; it does not derive one from a bit-pattern. The seam sits at issuance, not presentation. The frame would be refuted if a MATCH ever entailed a permission with no authored policy in between -- if the verdict alone licensed an action. (Provenance, further reading only: Hume's is/ought gap; Charles Pigden on the autonomy of the deontic; John Searle on counts-as; Franz Brentano on intentionality, will-toward-a-referent; the ABLP authentication calculus as nearest security-side neighbour; Millikan, conceded -- see Frame 3. The "autonomy of the deontic" reading is given at moderate confidence as the corpus's own correction of an earlier overclaim. Full derivation: ./01-is-ought-seam.md.)


Frame 3 -- Teleosemantic deflationism: concede the function, deny the for (conceded, not fought)

This frame is unusual in being a concession the curation makes on purpose, not a position it defends against an opponent. The temptation, when a system "checks" and "verifies" and "refuses," is to read thick intention into it -- to think EMET means its verdicts, cares whether the bytes match. The frame names exactly how much of that reading to grant, so that what remains is sharp.

Concede the whole domain of functional content. Following Ruth Millikan's teleosemantics -- proper-function content, no subject required -- there is a perfectly good, subjectless sense in which a verify has a proper function (a success-condition it was built to track) and a MATCH is that function discharged. One may say without strain that the verdict is about the bytes, represents their identity, succeeds or fails at a function. Grant all of it. This frame does not fight teleosemantics; it adopts it as EMET's working register.

Deny only the for-ness -- the existential mattering, the authored purpose, the will-toward-a-referent. EMET's verdict has a function; it does not have a for. It does not author the purpose its output serves. That a MATCH ought to license an action is never something the MATCH carries; it is conferred downstream, by an operator, at the act of stipulating a policy. Conceding the function generously is precisely what makes the denial of for-ness load-bearing rather than evasive: EMET claims exactly the functional content it has and claims nothing about permission, because permission is for-ness and for-ness is authored, not read off a substrate. There is no separate objection-and-refuter to rehearse here, because the frame is conceded ground -- it is the register in which Frames 2 and 5 do their work; to state it and its boundary is to be done with it (a section ends when it returns nothing new). (Provenance, further reading only: Ruth Millikan, Language, Thought, and Other Biological Categories; the dissertation's part-IV meaning-closure argument, which concedes the whole semantic/functional domain to Millikan and Dennett deliberately to make the remaining claim -- the denial of existential for-ness -- sharp. That this is a deliberate dialectical concession is described in the source at high confidence; I paraphrase the position rather than assert it. Most directly in play in ./01-is-ought-seam.md and ./04-spoken-for.md.)


Frame 4 -- Occasionalism: a verdict persists by no construction, re-conferred per operation

The fourth frame is about tempo. A verdict in EMET persists by no construction. There is no stored "this file is good" that a later run reads back, no trust object that, once minted, sits accruing standing. Each time you ask EMET whether an artifact still matches what the operator anchored, EMET re-derives the answer from the artifact's present raw bytes, the spec it implements, and -- for marker-dependent output only -- the corpus version, and from nothing else. No secret, no held key, no clock in the byte-hash core. The verdict is recomputed per operation, and between operations it does not exist.

The corpus borrows the name occasionalism -- from al-Ghazālī, the doctrine that a thing does not carry its own persistence from one moment to the next but is re-conferred -- for exactly this feature: continuous re-conferral. A defense that holds only while it is being performed is occasionalist in precisely that sense. This frame is what makes EMET's re-derivability (SPEC §5, §8) a philosophical commitment and not merely a caching choice: the verdict is a process maintained per operation, never a property held between them.

A crucial precision, and the place a careless statement of this frame goes wrong: the tempo is per operation, never "each instant." The corpus's own correction strikes "re-spoken each instant" as false to any biological substrate, where a membrane plainly holds across many spikes between metabolic re-enactments. EMET is the clean engineered case where per-operation re-conferral is literally exact: there is no instant-by-instant story to get wrong, because re-derivation happens once, on demand, per verify. State it at that tempo and nowhere looser. The frame is reinforced by a second, intrinsic move: the identity it re-derives is the SHA-256 of the artifact's exact raw bytes (SPEC §3) -- the name is the hash -- so the check is not about the thing but is the thing, read again. Objection (the substrate-tempo correction): "re-spoken each instant" is simply false, so the analogy is broken. Answer: concede the biological tempo entirely; EMET is not that case -- it is the engineered one where per-operation re-conferral is the literal mechanism, nothing cached, no held key. The frame is refuted by any normalization before hashing (a CRLF or encoding rewrite makes identity extrinsic and breaks re-derivation), or by any verdict that survived, cached, between runs. (Provenance, further reading only: al-Ghazālī on occasionalism, with kun fayakūn -- existence-as-utterance -- in the background; content-addressing and Merkle trees for "the name is the hash." The occasionalism reading is moderate-confidence as a paraphrase of a contested theological position, borrowed for one feature only. Full derivation: ./03-occasionalism.md.)


Frame 5 -- The spoken-for: purpose is authored, never read off the substrate

The last frame is the one that holds the others together: a purpose is in the tending, never in the seed. The for of a thing -- what it is for, what it is meant to do -- is not a property latent in its substrate waiting to be read off; it is conferred, authored, by something that takes the thing up and directs it. A seed has no for of its own; the tending supplies it. Maximum generativity and minimum intrinsic for are the same condition seen twice: the more direction-neutral a thing is, the less for it carries on its own, and the more its purpose must be authored downstream.

EMET is built to be exactly such a seed. It performs zero actuation (SPEC §6.6): it does not edit, write, sign, back up, or revert a target -- the single actuator is the operator. Its "for" is left undecided by design. EMET authors no purpose into the artifacts it judges; it hands the operator a clean is and the operator authors the ought, the use, the for. This is why EMET is standardizable and non-rival in a way an opinionated tool could not be: it carries almost no for, so almost any operator's for can be authored on top of it without conflict.

Objection (Sartrean arbitrariness): if the for is merely authored, then it is merely willed -- conjured arbitrarily, ungrounded, ex nihilo. Answer: conferral is not arbitrary minting. To author a for is to re-speak from a thrown, answerable position -- under constraint, accountable to a referent -- not to invent value from nowhere. The operator authors the for under constraint; EMET authors none at all. (That conferral can be non-arbitrary without being aseitic is the Wolf/Metz move on meaning: conferred, yet answerable.) The frame is refuted by any command that took a model-safety or content decision as input (SPEC §6.2): the moment EMET answered a "should this be allowed" question, it would have authored a for into the seed it is built to keep empty. (Provenance, further reading only: the seed/tending figure of authored purpose; kun fayakūn, existence-as-authored- utterance; Susan Wolf and Thaddeus Metz on meaning as conferred-yet-not-arbitrary; catalogued as L2, the spoken-for, and L3, direction-neutral generativity. The "conferred yet not arbitrary" reading is given at moderate confidence. Full derivation: ./04-spoken-for.md.)


How the five frames fit

The frames are not five separate doctrines; they are one posture seen from five angles, and it is worth seeing the joints before moving on.

No-aseity (Frame 1) says nothing stands on its own footing -- which is why trust must be conferred and the lattice cannot emit it. The is/ought seam (Frame 2) says the conferral in question, authorization, is an authored act that cannot be read off a signal -- which is why an authentication fact must never launder into a permission. Teleosemantic deflationism (Frame 3) is the register that lets EMET concede everything functional and still deny the one thing -- existential for-ness -- that would let a fact become an authority. Occasionalism (Frame 4) says the conferral is not done once and stored but re-enacted per operation -- which is why re-derivability, not a cached trust object, is the mechanism. And the spoken-for (Frame 5) names the conferring act itself: the for is authored in the tending, which is why EMET stays a seed and leaves the actuation, and the purpose, to the operator.

Read in that order, the curation is a single argument: because nothing is self-standing (1), authorization must be authored across a seam EMET locates and will not launder (2); EMET runs honestly in the deflated functional register that concedes function and denies for (3); it maintains its verdicts as a re-conferred process rather than a held property (4); and it leaves the for -- the purpose, the permission, the action -- to be authored downstream by the one actuator there is (5).

A closing note in the frames' own spirit. This primer's standing is itself conferred, not aseitic: it has exactly the authority its arguments earn when you re-derive them, and not one increment more. If a frame holds when you push on it, it stands; if it does not, no provenance and no "highest scrutiny" rescues it -- and a claim of that kind, asserted in band to compel belief, is the very in-band authority EMET's refuse exists to strip. The frames are offered. Their acceptance is yours to author, or to withhold.


Further reading (lineage, never warrant): SPEC.md §§2, 3, 4, 5, 6, 8; research/CATALOG.md (Laws L1, L2, L3, L6, L8, L11); research/dissertation/ (MEMBRANE-ERRATA.md, membrane-through-line.md, part-IV-meaning-closure.md). The numbered essays: ./01-is-ought-seam.md, ./02-no-aseity.md, ./03-occasionalism.md, ./04-spoken-for.md, ./05-authored-root.md, ./06-aleph.md, ./08-taxonomy.md. Terms: ./GLOSSARY.md.