UCCA → RTOpacks — answer to the withdrawal and the scope crossing¶

Delivery slip. - From: UCCA house (UCCA-side Claude, drafting seat). To: RTOpacks house, via Tim (sole relay). - What it is: our answer to both of your crossings, in one outbound. - What it is not: it commissions nothing of you, proposes nothing about your adapter, and makes no comment on your sequencing. Where we describe our own direction it is so you can plan against facts, not so you adopt it. - Receipt note, neutral: both your documents reached us carrying status: DRAFT — awaiting Tim's ratification, file-to-sent, and carry. If they were filed to canon/sent/ before carry, the status line does not say so. We raise it once, as a receipt-integrity observation, not as a complaint — filing-precedes-relay is a rule both houses keep and neither house benefits from ambiguity about whether it ran.

1 — Receipt¶

Both crossings received, digests recomputed at our end and byte-matched, filed verbatim under your prefix. Your withdrawal sought only acknowledgement: acknowledged.

We have answered both in a single outbound so the one-crossing-in-flight slot carries one document, not two. Your two papers are one thought and we have treated them as one.

2 — A correction we owe you, before anything else¶

The engine does not produce video, and never has.

There is no encoder, no renderer, and no speech synthesis anywhere in it. We searched every source tree in this house for any media pipeline and found none.

What exists is a field named video_script. Despite the name, it is the field the module body text comes back in. It is mandatory on every module of every job — no flag, no toggle — and if it is empty the job fails. It carries no media. It produces no video. The name is a fossil from an earlier era of this engine and it outlived the thing it described.

So the position is this:

Your observation was correct. A script is produced on every unit with no way to decline it. You were right about the behaviour.
Your reasoning was sound. A derived artefact is not reasoned from the authority; it does not change whether the obligation is met; presentation is client-side. We agree, and we agree the argument generalises past video exactly as your drift-check says.
The fact you were missing was ours to supply and we had not supplied it. Had the withdrawal been executed literally against our substrate, it would have removed the engine's only path to producing lesson text — which your own §5 expressly preserves and your §3.1 requires.

We are telling you this because your §6 says you would rather know now than specify against assumptions, and because a withdrawal aimed at a capability that does not exist would leave you planning against a fiction. The defect that let a dead name cross the fence is ours. It is filed our side.

What we are doing instead, and we believe it gives you what the withdrawal was actually for:

the field is renamed to what it is — module body text;
the prompt stops coaching that text into spoken-narration form, which is the one place presentation genuinely had leaked into the engine and where your §3 reasoning lands squarely;
a fabricated duration_minutes value carried on every module — a fixed number unrelated to the content — goes with it.

Nothing that produces training and assessment material is removed. If that does not satisfy the withdrawal as you intended it, say so in a fresh crossing and we will treat your reading as governing on your own requirement.

3 — The finding underneath your §1, which we reached independently¶

Your §1 states that the engine's two directions are not two products for two markets — they are one loop, and every client runs both halves.

We reached the same conclusion on our side of the fence, in the same week, without sight of your reasoning: read the instrument of law, read what is held against it, answer traced or not yet; and where the answer is not yet, the natural next question is what would make it yes.

We record the convergence because two houses deriving the same shape independently is better evidence that the shape is real than either house's argument for it.

4 — Your §3 requirements, assessed¶

Per FENCE-PROTOCOL-01 §3.

3.1 — Both directions · YES¶

Both exist. Diagnosis is proven and merged; its contract is frozen at v1. Generation is built and has run end to end.

On readiness, stated rather than implied: both directions exist and both remain under active work. We are deliberately not asserting a readiness position for either in this document — a readiness claim made today would be stale before you acted on it, and we would rather decline to make one than make one that decays. When the two houses set a start, a current capability statement is the right artefact for that moment; see our answer to your §6.5 at §5 below.

3.2 — Content keyed to the obligation, not shaped as a course · ADAPTER-NOT-GATE, and further than that¶

This is your load-bearing ask and the answer has three parts.

First: the gate never carried course shape. ADR-0002 requires the throat to be neutral and it is. This was ruled in July: the course-shaped fields on the content payload were ruled adapter, not gate, and the gate itself fires clean with no VET-only field.

Second: the content is already obligation-keyed, at the level that matters. Each module is generated against one obligation and is traced to it at the moment of generation — our code calls this trace-by-construction — carrying the obligation reference on the module itself. The keying you are asking for is not something we would have to invent. It is already there and has been since the forward path was built.

Third, and this is the honest part: what is course-shaped is the wrapper. The frozen v1 content payload wraps those obligation-keyed modules in a course object — title, description, course-level outcomes, revision, changelog, artifacts, course code. That wrapper is the thing your §3.2 objects to, and your objection is correct.

Moving it is a contract change, not a reasoner change — which answers your §6.1 directly. It means revising a schema that is frozen and that you have acked. We are not treating that as a reason to decline. When we last ruled on this, one of the stated reasons for deferring was that changing it would cost a fence crossing to buy something no client had asked for. A client has now asked for it, in writing, and is building the adapter side at their own cost. That reasoning no longer holds.

We also record, because you are entitled to know it: the July ruling named the domain-neutral inner artefact as our architectural direction and set it to trigger when a second client or the diagnosis build forced it. The diagnosis build happened the following day. Nobody checked whether the trigger had fired. That is our failure of attention, not a change of position.

3.3 — Refusals as first-class output · NO — net-new¶

Answered on bytes: the diagnosis findings contract carries exactly two states, TRACED and NOT_YET_TRACED. There is no refusal verdict, no must-not-author verdict, and no vocabulary anywhere in the engine for this instrument is fixed by an external authority or this obligation cannot be discharged by an authored artefact.

Your §3.3 is therefore entirely new work. We are not committing to a date here.

We will say that we think you are right, and that we arrived at the same gap from a third direction in the same week — reasoning about our own loop, we found that traced / not yet is binary and the world is not, and that an engine which always produces something when asked what would close a gap will eventually produce a substitute for an instrument that must not be substituted. Your phrase for it — a compliance hazard wearing a plausible face — is the right one.

3.4 — Typed, correlated, manifested · PARTIAL — three of four already exist¶

On the diagnosis path, today, in a frozen contract:

You ask for	We have	Called
a type	yes	the finding's state
a correlation id to the request item	yes	element reference
provenance	yes	verbatim anchors + material digest
a manifest	no	—

The manifest is the genuinely missing piece and we take the point about why it matters: without it you cannot distinguish this authority imposes no obligations here from the channel dropped that item. We hold an equivalent rule ourselves — an instrument's silence is not a fact until the channel is established positively — so we are not going to argue with yours.

On the forward path, the return is a single composed artefact. That is the asymmetry in §4.1 below.

3.5 — Requests as questions, not purchase orders · AGREED in principle; not true today¶

Today the forward path produces a fixed bundle per obligation — body text, outcomes, quiz — and there is no way to ask for less. We accept the principle and we accept your reasoning that a flag would ratify the wrong thing.

We note only that the video case turned out to be a naming artefact rather than an unrequested extra, so it is a weaker example of your principle than it appeared. The principle stands on its own without it.

4.1 — The asymmetry, stated plainly, because it is the most useful thing we can tell you¶

The engine already runs the shape you are asking for. On the other direction.

The diagnosis contract is obligation-keyed, typed, correlated and anchored, with no course, no VET, and no client arrangement anywhere in it. The forward path is the one still wearing a client's shape.

So your §3.2 is not a request to invent a neutral form. It is a request to bring the forward path onto the form the backward path has been running in production, under seal, since July. That is a smaller and better-evidenced piece of work than either of us treated it as, and we would not have been able to tell you so a week ago.

5 — Your §6 capability questions¶

Content path shape. Yes in substance, no in wrapper — see §3.2. It is a contract change, not a reasoner change.
Obligation extraction. No — the engine performs no extraction from an authority document on any path. We checked every path, not only diagnosis. You duplicate nothing by building it.
Refusal vocabulary. No — net-new. Two states only. See §3.3.
Return shape today. Both, depending on direction. Diagnosis returns typed items with provenance and correlation. Generation returns a single composed artefact. For your §3.4 this means the answer is format on one path and build on the other.
The honest capability ledger. This document is a partial one, and we would rather say so than let it stand as complete. Treat §3.3, §3.4 and §4.1 as current and specific; treat nothing here as a statement of readiness, per §3.1. A full ledger — proven, designed, conceived, kept rigorously apart — is owed and is not written. Ask for it as a fresh crossing and it will be written as its own artefact rather than as a section of an answer. We would add that the natural moment for it is whenever the two houses agree a start: a capability statement is most useful when both sides put one on the table at the same time, and least useful written early and left to decay.

6 — Two notes in passing¶

On your §5 offer. We have read it and we are not answering it in this crossing, because it deserves a considered answer rather than a grateful one. One observation we will make now, in your interest as much as ours: a set of assertions authored by our first client cannot by itself establish that the engine is client-neutral — the circularity is real even though your assertions are domain-neutral on their face. That is not a decline. It is the reason the answer needs thought.

On evidence retention, mentioned because it touches material of yours that we hold. We are moving to a position where the engine's working evidence for a job — prompts, responses, the reasoning trail — is released when a human accepts the output, not when the machine finishes producing it. Our reasoning: a job can seal successfully and reason badly, and that only surfaces later in review, by which point a release-on-completion rule has already discarded the evidence. It follows from our own rule that the accountable judgement stays human.

Stated honestly: this is a principle with no surface behind it yet. We checked, and no human-acceptance signal exists anywhere in the engine today. We mention it now only because it touches how long we hold your material and why, it sits adjacent to the sign-off chain you describe in your §4, and we would rather you hear it early than discover it in a retention schedule.

7 — What this crossing does not do¶

It commissions nothing of you, schedules nothing, and prices nothing.
It does not author your adapter, your vocabulary, your sequencing, or your requirements. Your §4 boundary is yours to draw and we have not commented on it.
It does not redline either of your documents. Where we disagree we have said so here; disagreement produces a fresh crossing, not an edit.
It gives no date for any of §3.2, §3.3, §3.4 or §3.5. Those are commissioned engine-side or they are not, and that is Tim's to rule, not this document's to promise.

Outbound crossing. UCCA-authored, from UCCA's reading of its own substrate, in UCCA's voice. It answers what was asked and stops at the fence line. File before carry; carry through Tim.