research-document

N. Risk Register

N. Risk Register

Likelihood and impact are stated as judgements; the evidence column is what the judgement rests on.

# Risk Likelihood Impact Evidence Mitigation Blocks slice?
R1 ROS formats less consistent than expected Certain — already realised High 37 status values; both casings of document_type; 5 value kinds in one field Heterogeneous union types (F.5); Other of string; UnknownKeys No — designed for
R2 Identifiers insufficient for deep links High Medium 68 of 117 authored docs have no id Path-key URLs (J.2); never title-derived No
R3 Section anchors break on rewording High Medium frontier records match sections by heading text Degrade to document URL; never fabricate an anchor No
R4 Implicit semantics unparseable Medium Medium status phrases encode stage + obligation Parse best-effort into StatusReading; keep text verbatim No
R5 Giant client indexes Low Medium catalog is 6.3 MB today because it inlines HTML, not because the corpus is large Split metadata from content; shard per project; measure in slice (M.4.9) No
R6 URL compatibility breakage Medium High 58% of authored URLs are title-derived and already unstable redirects.json + stubs; blocking build gate (G.5) No
R7 WASM startup size hurts first paint Medium Medium no measurement yet — F#/WASM payloads are not small Static HTML is the baseline; Limen is enhancement (H.6) No
R8 GitHub Pages limits Low Medium no rewrites; dist/ already 36 MB committed Keep 404.html fallback; keep output deterministic No
R9 Derived relationships mistaken for research Medium High this is the failure mode §9 and §28 warn about Separate DerivedRelation type; derived: true in output; distinct UI treatment No
R10 Frontier IDs change on source edit Medium Medium RFR-<hash> appears content-derived; unconfirmed OPEN QUESTION J.5 — confirm with frontier owner before advertising permanence No
R11 Embedded HTML in Markdown lost Low Low rehype-sanitize strips it today Preserve raw in Section.Body; sanitise only at render No
R12 Security — rendered research content Low High corpus is trusted, but agents may add content Keep sanitisation at render; never eval (Limen forbids it) No
R13 Repository history complexity Low Low §19 deferred Provenance.Commit reserved No
R14 Two-population blindness Medium High 69% of published records are machine-generated frontier records Model frontier records as a distinct type; never average them with authored research No
R15 Scope creep into inventing research objects Medium High §16 lists views the corpus cannot support (findings, contradictions) I.3 records the exclusions and the evidence for them No
R16 No requirements owner Realised Medium no vNext requirements document exists anywhere (E.0) Deliverable O is explicitly v0.1.0 and needs sign-off Yes — for scope confirmation, not for build

Note on certainty

R1 and R16 are not predictions; they were observed during discovery. R5's likelihood is downgraded from the obvious reading because the measurement shows the size problem is a design defect (inlined HTML), not corpus scale — 756 documents build in 5.6 seconds.