risk-writing-guide asks for the reciprocal relation that relation-rules forbids

[dieterbaier]

Problem

Two skill references give opposite instructions for the same relation, and an
author following the risk guide produces a relation the relation rules forbid.

skills/references/risk-writing-guide.md, under Relation Guidance:

Use depends_on from the risk to the ADR when the risk exists because of the
decision.

skills/references/relation-rules.md, under Impact Rules:

Do NOT link artifacts back with depends_on as a reciprocal for
introduces_risk, addresses, or other directional relations. The risk or
scenario's dependency on the decision is derived from the ADR's outgoing
relation.

Both files are read by the same workflow: skills/adr/SKILL.md lists them
together under Required Reading, and skills/risk/SKILL.md reaches the risk
guide when a decision creates a risk.

Which one is right

relation-rules.md. It matches the authoritative-outgoing-relations convention,
the metamodel, and the repository's own practice:

• No risk under src/docs/arc42/11-risks-and-technical-debt/ carries a
  depends_on to an ADR — R-001 through R-007 have none.
• The ADRs carry the outgoing direction instead: ADR-001, ADR-003 and ADR-008
  each declare introduces_risk.
• The incoming direction is derived. ADR-005's generated traceability view shows
  the incoming refines from ADR-008 without either file storing it.

The advice is not merely stylistic — it produces a warning

Verified by temporarily adding the relation the risk guide asks for, to a risk
whose ADR already declares introduces_risk:

- type: depends_on
  target: ADR-008-skill-decomposition-boundary

./build.sh validate then reports:

Bidirectional relation detected: ADR-008-skill-decomposition-boundary ->
R-007-gate-results-lack-independence and R-007-gate-results-lack-independence ->
ADR-008-skill-decomposition-boundary. Consider removing the reciprocal relation
from one artifact.

The probe was reverted; it is not in any branch. So the guide walks an author
into a warning that the same toolkit then tells them to undo.

Scope

• Correct the Relation Guidance section in
  skills/references/risk-writing-guide.md so the risk does not link back to
  the ADR, and say which direction is authoritative and that the reverse is
  derived.
• Check the neighbouring references for the same defect before changing only
  this one. skills/references/quality-scenario-guide.md is the obvious
  candidate, since addresses has the same shape as introduces_risk.
• templates/risk.adoc ships a depends_on relation to an ADR in its example
  front matter. Decide whether the template is teaching the same thing and
  correct it if so.
• Keep the canonical rule in one place. The guides should point at
  relation-rules.md and the metamodel rather than restating relation
  semantics, so this cannot drift apart again.

Acceptance criteria

• No skill reference instructs an author to create a relation that
  relation-rules.md forbids.
• The risk guide states the authoritative direction for a decision-created
  risk and notes that the reverse is derived.
• templates/risk.adoc does not demonstrate a forbidden reciprocal.
• The neighbouring guides are checked, and any identical defect is fixed in
  the same change.
• Relation semantics live in one canonical place; the guides reference it.
• ./build.sh validate reports no new warning, and the existing warning
  count does not grow.

Origin

Found on 2026-09-04 while authoring R-007-gate-results-lack-independence for
#86. The two references were read in the same sitting, which is the only reason
the contradiction surfaced; an author reading only the risk guide would have
followed it.