Skip to content

RTOP-CROSSING-AB-QUALITY-RULE-SCOPE-01

Pre-run amendment to the declared A/B quality rule — scope of the commercial floor

From: RTOpacks To: UCCA Status: DRAFT — not yet relayed. Becomes a crossing only when Tim relays it; binds only when UCCA accepts in writing, and only if that acceptance lands before any job is submitted. Relay digest: to be computed on the relayed bytes and quoted in the relay message.

1. What this is, and why it is legitimate now

This amends the pre-declared quality rule for the agreed A/B (one unit, BSBPEF501, four arms, brief-varied). It is offered before any run, on findings made before any run. A quality standard amended after a result exists is a position; amended before, it is calibration. That distinction is one your house stated first, and we are relying on it: this crossing exists so that the amendment is on the record in writing, both houses, ahead of the first submission — after which no amendment of this kind will be entertained by either side.

Nothing in this crossing reports a quality observation. No output has been read; none exists.

2. The two facts that require it

One fact from each side of the boundary. Ours is ours to state; yours is cited to documents your house has already filed and crossed — we assert nothing about your substrate beyond your own citations.

Fact A — ours. The triumvirate our adapter submits carries two component classes: elements and performance criteria. Knowledge evidence and performance evidence are held in our corpus and are not carried — not filtered, not emptied; never fetched. This is established on bytes in our filed recon (EXM-RECON-01-FINDINGS-2026-08-01.md, sha256 f7fcc3059b7fb74d9efa65c1df083fd47b9291d27e2a1559762f84079658a097, 13,940 B — held on our side; the citation is for the record, not an attachment), cited to file and line. Consequence: the engine has never received the Assessment Requirements of any unit it has generated for. That is our defect to fix on our side of the line, and a corrected slice is part of the exchange-model position we will cross separately.

Fact B — yours, by your citations. The frozen output content-payload schema (ucca-content-payload-schema-v1.json, as crossed) provides exactly one assessment-shaped structure: modules[].quiz.questions[] (question_id, question, options, answer_index, explanation). It provides nothing for observation checklists, practical tasks, or marking guidance. The trace map's floor is module-level element_ref. Your RUNNABLE-STATE records that instruments are carried by zero payloads and read by nothing.

Joint consequence. Under the contract as it stands, no run of the A/B can return a complete, markable assessment face — regardless of the brief, the arm, or the quality of the reasoning. The absence is structural and pre-diagnosed. It is not evidence about contextualisation, and it must not be allowed to masquerade as evidence about contextualisation.

3. The amendment

For the A/B as designed (four arms as agreed 2026-08-01), the declared quality rule is amended in one respect:

The commercial-floor conditions — "markable" and "a course developer would rather edit this than start from blank" — are scoped to the training face and to whatever assessment content the current contract can carry (at most, module-anchored quiz structure). The absence of a complete assessment face (performance instruments, assessment-conditions-conformant methods, item-level requirement anchoring) is recorded as a known structural bound under active negotiation, attributable to the contract, and is excluded from KILL / REBUILD / BUILD attribution in this run. A missing assessment face in this run voids nothing and scores nothing.

4. What is not amended

Everything else stands exactly as declared, and we call out the load-bearing items by name so this crossing cannot be read as loosening them:

  • Traceability voids rather than scores. A generated claim that cannot join to a requirement through the sanctioned path still voids the artefact.
  • Clean. RTOpacks-authored text presented as generated still voids.
  • VOID — PARTIAL DELIVERY. brief_modules_rendered != brief_modules_total still voids, re-runs, and is never compared against a full arm. (Separately owed, not amended here: the population that rendered/total ranges over has not been named by either house and is owed before that instrument discharges anything.)
  • Blind attribution. Outputs are read without knowledge of arm; matching comes after.
  • Ordering. The quality read still runs first and still gates the framing read. Only the floor's scope changes — not its position in the sequence.
  • The framing rule. KILL / REBUILD / BUILD still rides on Arm 1 against Arm 2; Arm 0 remains the did-anything-change baseline, not the is-it-better baseline.

5. What we ask of you

  1. Accept or reject this amendment in writing, before any submission. Silence is rejection; an unamended rule stands and clause 2's joint consequence then applies in full — which both houses should understand means the run's quality gate fails structurally and the framing result is not actionable under the rule as declared. We would rather amend than waste the run.
  2. Confirm the amendment lapses if any job is submitted before acceptance lands. Neither house gets to hold an unaccepted amendment in reserve.

6. What this crossing deliberately does not do

It does not ask you to change the contract, the schema, or the engine. The return-shape requirement — the full training-and-assessment artefact with item-level anchoring — is a client requirement we will present separately, as a position with its evidence attached. This crossing only ensures that the A/B we both designed measures what it was designed to measure: whether qualification context changes the material. That question is still worth the run.

End of crossing. RTOpacks drafting seat, 2026-08-01. Relayed by Tim; authorship stays home.