Skip to content

Receipt: UCCA Engine Capability Answer — Evaluative (evidence→competency) Reasoning

What this is. The home-side record filing an inbound fence crossing. Per FENCE-PROTOCOL-01 §4, the received copy is filed verbatim beside this record and is never promoted into home canon. Anything below that binds RTOpacks work binds through this record, which cites the crossing.

1. The crossing

Received 2026-07-06 via Tim (sole relay), one crossing in flight, fence otherwise quiet. The artefact is UCCA-authored, bytes-hardened on a read-only recon of the engine's frozen diagnosis contracts and reasoner (recon digests carried in the artefact's own frontmatter). It answers our RPL capability query: can the engine perform evaluative (evidence→competency) reasoning for Recognition of Prior Learning, and at what build size.

Verbatim received copy: UCCA-CROSSING-RPL-CAPABILITY-ANSWER-01.md, digest and byte count in this record's frontmatter. Digest verified at receipt and at filing.

2. What the answer establishes (our words, binding locally)

  1. Capability confirmed; sizing corrected. Our working hypothesis ("yes, but likely a re-engineer") is confirmed on the yes and corrected on the size: this is (b) a new reasoning mode on the existing core, not a re-engineer. The engine's proven backward primitive — material → per-leaf verdict with byte-verified anchor, raise-hands-never-sign — is the same operation pointed at candidate evidence. No course-shape coupling leaks from the forward path into the backward path; checked on bytes, absent.

  2. The substantial build is ours. The engine's frozen payload already fits (evidence → material.sections; unit leaves → obligation.elements). What is net-new and larger sits on our side of the warranty line: the evidence-normalisation adapter converting heterogeneous evidence (portfolios, employment records, third-party reports, references, image/video) into the engine's text-sections shape. The engine core is text; image/video evidence requires a transcription/description pre-step in our adapter.

  3. Rules-of-evidence split (load-bearing; matches rpl-00-obligations exactly):

  4. Validity → engine anchors (the TRACED-with-anchor act itself)
  5. Sufficiency → assessor; the engine is constitutionally barred (raise-hands-never-sign)
  6. Authenticity → assessor plus adapter (provenance/corroboration)

  7. Currency → adapter recency check plus assessor The two rules the engine cannot hold are exactly the two the law reserves for the assessor. Anchor-not-decide is not a reduced form of the task — it is the compliance posture.

  8. Findings vocabulary is deferred. TRACED/NOT_YET_TRACED versus literal "DEMONSTRATED" wording is a mode-layer schema-version choice, decided at commission. Not core work, not a dependency for rpl-01-model.

  9. No commitment crossed. The answer is a capability assessment only. Any engine-side RPL mode opens as its own gated units on Tim's word. Nothing in this receipt schedules or commissions.

3. What this unblocks

  • rpl-01-model may now be drafted; it cites this record. The engine-dependency question for RPL is answered: dual-mode (manual / UCCA-assisted) is architecturally real, with the adapter as our build surface.
  • frozen-launch-scope.md RPL Gate-1 flag: superseded by this crossing. The annotation lands in the same commit as this record — note and referent together, never apart.
  • llnd-01-model is NOT covered by this answer. LLND's engine dependency remains a model question (flagged in llnd-00-obligations, not inherited from RPL).

4. Discipline note (recorded, not retroactive)

The inbound artefact correctly notes our query crossed as scoping prose without a filed RTOP- doc_id. Standing correction going forward: outbound queries file as RTOP- artefacts before they cross, so both houses can reference one doc_id. This crossing stands regardless; the gap is noted, not cured.

Received-crossing record. The UCCA copy beside this file is theirs — referenced, never re-authored, never promoted. Everything that binds our build binds here.