docs(issues): [#2264] [#2278] record the discussion outcomes from PR #2428

docs/discussions/2003-overhaul-guardrails-and-automation/20261003-goals-and-boundaries/README.md

@@ -235,8 +235,12 @@ examines what outlives them. In brief:
 Cameron (`da2ce7`), owner of EPIC #2003, answered the six open questions in
 [review 5400754664](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5400754664)
 (round 1). The answers are transcribed here at his request, for him to confirm against the review
-text. Each is marked there as a decision for the EPIC owner to record; none has been recorded in
-EPIC #2003, #2264, or #2278 yet.
+text. The parts for EPIC #2264 are recorded under its
+[Decisions Recorded on This EPIC](../../../issues/open/2264-2003-refactor-semantic-link-conventions/EPIC.md#decisions-recorded-on-this-epic)
+section, the parts for EPIC #2278 under its
+[Decision Record](../../../issues/open/2278-2003-strengthen-pr-review-author-self-audit/EPIC.md#decision-record)
+section, and the parts for EPIC #2003 in that EPIC's specification by its overhaul pull
+request #2366.
 
 1. **The four-aspect frame** is right for sorting proposals. The gates are the layer every
    aspect's contract is expressed through, not a fifth aspect. What the frame does not yet name is

docs/discussions/2003-overhaul-guardrails-and-automation/20261003-semantic-linking-knowledge-graph/README.md

@@ -230,8 +230,10 @@ These are drafts for review.
 Cameron (`da2ce7`), owner of EPIC #2003 and #2264, answered the four open questions in
 [review 5400754664](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5400754664)
 (round 1). The answers are transcribed here at his request, for him to confirm against the review
-text. Each is marked there as a decision for the EPIC owner to record; none has been recorded in
-EPIC #2264 yet.
+text. The four answers are recorded in EPIC #2264's specification under its
+[Decisions Recorded on This EPIC](../../../issues/open/2264-2003-refactor-semantic-link-conventions/EPIC.md#decisions-recorded-on-this-epic)
+section, with the Progress Log entry of 2026-10-04 and the order 5 and 7 cells carrying the
+scope, as item 4 asks.
 
 1. **Yes, start "Normalize the semantic-link model" with the inventory,** following #2264's own
    precedent of inventorying frontmatter before fixing its model. The inventory is a tracked

docs/issues/open/2264-2003-refactor-semantic-link-conventions/EPIC.md

@@ -6,7 +6,7 @@ epic: 2003
 github-issue: 2264
 spec-path: docs/issues/open/2264-2003-refactor-semantic-link-conventions/EPIC.md
 epic-owner: da2ce7
-last-updated-utc: "2026-10-03 20:43"
+last-updated-utc: "2026-10-09 10:41"
 semantic-links:
   skill-links:
     - create-issue
@@ -343,9 +343,9 @@ Status values: `TODO`, `IN_PROGRESS`, `BLOCKED`, `DONE`.
 | 2.2 | Frontmatter | [#2281 - Add frontmatter validator command and pre-commit rollout](../../closed/2281-2264-frontmatter-validator-command/ISSUE.md) | Read-only command for explicit paths, `--staged`, and whole-tree validation with NDJSON diagnostics, location-dependent severity, repository-aware checks for strict v1 specs, and a pre-commit step. | DONE |
 | 3 | Frontmatter | Extend strict profiles and author guidance | Add approved profiles for ADRs, skills, agents, evidence records, and other known classes; keep prose and editor/agent discovery derived from the canonical Rust model. Protect `docs/templates/` against drift from the v1 contract, which #2281 excludes from validation. | TODO |
 | 4 | Conventions | Split convention ownership and migration policy | Move mixed normative content to clearly owned convention documents without a repository-wide document migration. | TODO |
-| 5 | Semantic links | Normalize the semantic-link model | Canonical relation and target types, graph semantics, accepted/rejected fixtures, and compatibility rules represented by the shared Rust model. | TODO |
+| 5 | Semantic links | Normalize the semantic-link model | Canonical relation and target types, graph semantics, accepted/rejected fixtures, and compatibility rules represented by the shared Rust model. Starts with a tracked inventory script and its output pinned to a named `develop` commit; targets without a path take one `kind:value` form, and the `issue` sweep recounts its scope at its starting head (decided 2026-10-03; see Decisions Recorded on This EPIC). | TODO |
 | 6 | Semantic links | Design and implement semantic-link validation | Validate target syntax, type, existence where applicable, uniqueness, and invalid combinations through a replaceable check. | TODO |
-| 7 | Path references | Decide path-reference scope and syntax | Decide whether machine-checkable prose paths are semantic links, a separate typed reference, or intentionally outside validation. | TODO |
+| 7 | Path references | Decide path-reference scope and syntax | Decide whether machine-checkable prose paths are semantic links, a separate typed reference, or intentionally outside validation. Bounded by the 2026-10-03 decisions: targets without a path take one `kind:value` form, links are validated by the purpose of their field, and the graph reads tracked files only (see Decisions Recorded on This EPIC). | TODO |
 | 8 | Path references | Specify and implement path-reference validation | If the path-reference scope issue approves validation, define historical handling, exclusions, diagnostics, and staged enforcement. | TODO |
 | 9 | External links | Classify link importance and checker limitations | Use the #2185 handoff to decide whether link importance belongs in typed metadata, prose policy, or checker configuration. | TODO |
 | 10 | Integration | Reconcile validators with the #2003 architecture decision | Preserve domain semantics while migrating checks to the selected execution, output, policy, and caching architecture when required. | TODO |
@@ -443,6 +443,14 @@ the residual report and the #2185
 baseline as its evidence and may conclude that the answer is prose policy, a small typed
 annotation, or a different checker configuration; the EPIC does not presume which.
 
+## Decisions Recorded on This EPIC
+
+- **Inventory first (2026-10-03, [review 5400754664](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5400754664), [Outcome](../../../discussions/2003-overhaul-guardrails-and-automation/20261003-semantic-linking-knowledge-graph/README.md#outcome)).** Order 5, "Normalize the semantic-link model", starts with an inventory, following this EPIC's precedent of inventorying frontmatter before fixing its model. The inventory is a tracked script plus its output pinned to a named `develop` commit; it classifies the corpus by source document class and target kind, including the free-form values and the 48 dropped `issue` entries as candidate target kinds.
+- **One `kind:value` form for targets without a path (2026-10-03, [review 5400754664](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5400754664), [Outcome](../../../discussions/2003-overhaul-guardrails-and-automation/20261003-semantic-linking-knowledge-graph/README.md#outcome)).** In direction, every target without a path takes one `kind:value` form. A plain scalar such as `issue:2264` loads as written and matches this EPIC's candidate syntax, and the migration is mechanical even for the 48 damaged entries, whose numbers survive as YAML comments. The sweep's specification recounts its scope at the head it starts from, because the register records 53 entries in 34 files where the discussion and its review find 48 in 30. This changes the frozen v1 reference convention, which this EPIC reserves to its semantic-link subissues.
+- **Tracked files only (2026-10-03, [review 5400754664](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5400754664), [Outcome](../../../discussions/2003-overhaul-guardrails-and-automation/20261003-semantic-linking-knowledge-graph/README.md#outcome)).** The graph that gates and agents rely on reads tracked files only. GitHub-hosted records may enter as a separately derived data set stamped with its capture time; if the friction register's links are to feed the graph, a tracked, dated export serves it, as `docs/external-snapshots/` does for external sources.
+- **Links validated by the purpose of their field (2026-10-03, [review 5400754664](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5400754664), [Outcome](../../../discussions/2003-overhaul-guardrails-and-automation/20261003-goals-and-boundaries/README.md#outcome)).** ADR links are context links, validated leniently, and the convention's existing rule is enforced: issue specifications are referenced by number, which would have kept three of the five broken ADR links valid. Historical records (ADRs, PR-review audits, closed specifications) carry context links checked for syntax only. The change-impact field definition in `docs/skills/semantic-skill-link-convention.md` is split from a context field, or qualified, as part of this EPIC's convention work.
+- **Where these decisions are recorded (2026-10-03, [review 5400754664](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5400754664), [semantic-linking Outcome](../../../discussions/2003-overhaul-guardrails-and-automation/20261003-semantic-linking-knowledge-graph/README.md#outcome), [goals-and-boundaries Outcome](../../../discussions/2003-overhaul-guardrails-and-automation/20261003-goals-and-boundaries/README.md#outcome)).** In this specification: this section, the Progress Log entry of 2026-10-04, and the Expected output cells of orders 5 and 7, which carry the scope changes. The four-aspect frame and the discussions convention are recorded on #2003 by its overhaul pull request #2366, and the audit-record contract on #2278; the discussions are not cited in place of these records.
+
 ## Open Questions
 
 - Which existing frontmatter variations are intentional compatibility cases, and which are errors
@@ -454,11 +462,11 @@ annotation, or a different checker configuration; the EPIC does not presume whic
   repository-aware validation?
 - Which generated schema format best serves editors and AI agents without becoming a second source
   of truth?
-- Should semantic-link target types be URI-like strings, YAML maps, or both?
+- Should semantic-link target types be URI-like strings, YAML maps, or both? *Answered in direction (2026-10-03):* one `kind:value` string form for every target without a path; see Decisions Recorded on This EPIC.
 - Should a file path be represented as `path:...`, a typed YAML object, or a plain string in selected
   fields?
-- Which references should be considered knowledge-graph edges, and which are only validation aids?
-- Should historical closed issues and PR review records be exempt from new reference syntax?
+- Which references should be considered knowledge-graph edges, and which are only validation aids? *Answered in part (2026-10-03):* the graph that gates and agents rely on reads tracked files only, and GitHub-hosted records enter only as a dated derived export; see Decisions Recorded on This EPIC.
+- Should historical closed issues and PR review records be exempt from new reference syntax? *Answered in part (2026-10-03):* historical records carry context links checked for syntax only, and issue specifications are referenced by number; see Decisions Recorded on This EPIC.
 - How should comments in Rust, shell, TOML, YAML, and Dockerfiles represent typed links without
   harming the host language parser?
 - Which OKF principles should be copied directly, adapted, or rejected for a repository whose
@@ -626,6 +634,7 @@ For each completed subissue, the default completion policy is:
 - 2026-09-30 10:22 UTC - GitHub Copilot - Added the unordered Draft Subissue Proposals list with
   the maintainer-requested draft for multiple related PRs in issue frontmatter; its order in the
   table is left to the EPIC's planner.
+- 2026-10-04 19:18 UTC - da2ce7 - Recorded the semantic-linking discussion's outcome and the goals-and-boundaries parts that name this EPIC (PR #2428, review 5400754664) under Decisions Recorded on This EPIC; marked the Open Questions they answer and noted the scope of orders 5 and 7 in the subissue table.
 
 ## Risks and Trade-offs
 

docs/issues/open/2278-2003-strengthen-pr-review-author-self-audit/EPIC.md

@@ -6,7 +6,7 @@ epic: 2003
 github-issue: 2278
 spec-path: docs/issues/open/2278-2003-strengthen-pr-review-author-self-audit/EPIC.md
 epic-owner: da2ce7
-last-updated-utc: "2026-10-08 17:28"
+last-updated-utc: "2026-10-09 10:41"
 semantic-links:
   skill-links:
     - create-issue
@@ -97,6 +97,13 @@ Decisions the matrix fixes and that subissues must respect:
 - the audit validator validates the audit body only; frontmatter and `review-finding:` target
   existence belong to EPIC #2264.
 
+Decisions recorded from the PR #2428 discussions (2026-10-04; [goals-and-boundaries Outcome](../../../discussions/2003-overhaul-guardrails-and-automation/20261003-goals-and-boundaries/README.md#outcome), [review 5400754664](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5400754664) of 2026-10-03):
+
+- the audit record is contract: a gate checks it from tracked bytes and GitHub state, whatever tool produced it;
+- copies of GitHub state: the record's identifiers stay, and every state it copies is stamped with its capture time or checked; that is the premise of the validator rows, and the record's thread states, which copy review-thread resolution, are already checked against thread captures;
+- `epic-owner` copies the GitHub assignee: PR #2428 repaired it on this EPIC and #2264, and it drifts again unless a check is added or the field defers to the assignee;
+- the specifications-and-rationale outcome sends no work to this EPIC: marking the issue template's optional and conditional sections is #2003's planning-template follow-up ([Outcome](../../../discussions/2003-overhaul-guardrails-and-automation/20261003-specifications-and-rationale/README.md#outcome), [review 5402713128](https://github.com/torrust/torrust-tracker/pull/2428#pullrequestreview-5402713128), round 3).
+
 ## Scope
 
 ### In Scope
@@ -269,6 +276,7 @@ requests where possible; a fixture is used only when no real review produced the
 - 2026-09-26 09:46 UTC - GitHub Copilot - PR #2339 merged as `a20e8f3a` and closed #2333; order 4 is `DONE`, its specification is archived under `docs/issues/closed/`, and AC2 is `DONE`, completing Phase 1. Three review findings on PR #2339 were still unprocessed at merge; with maintainer approval (<https://github.com/torrust/torrust-tracker/pull/2339#issuecomment-5844522320>) they are handled in a follow-up PR from `develop`, which also extends `process-pr-review`'s post-merge rule to findings unprocessed at merge. Orders 5 and 6 are unblocked.
 - 2026-09-26 13:33 UTC - GitHub Copilot - Maintainer set the order after Phase 1: order 6 next, then order 5 once PR #2344 merges and #2347 triage settles which audit-contract findings belong to it, because orders 5, #2344, and #2347 all edit `process-pr-review`. Maintainer approved the order 6 specification with the `no-stdout-result` output migration; created and linked subissue #2349 and moved its specification to `docs/issues/open/2349-2278-contract-checker-evidence-boundary/`. Spec-only PR pending. For order 7, #2266 recorded its placement decision (`contrib/dev-tools/checks/`, `cargo run --package`, `no-stdout-result`), so it is unblocked; the maintainer approved three decisions for its specification: keep `no-stdout-result` and record dropping the Python prototype's stdout JSON summary as a deliberate parity deviation; do not wait for #2281, adopting its NDJSON diagnostic catalog if it has merged when order 7 starts; and no pre-commit integration, because the validator needs `gh` and network access.
 - 2026-09-28 10:08 UTC - GitHub Copilot - #2347 triage settled which audit-contract findings belong where. None goes to order 5. `review-finding:pr-2313-f4`, `-f5`, `-f6`, `-f7`, and `-f9` go to the new order 11, which the maintainer approved; created and linked subissue #2362. `review-finding:pr-2300-f10` goes to order 8 (T3 approval <https://github.com/torrust/torrust-tracker/issues/2347#issuecomment-5865646588>).
+- 2026-10-04 19:18 UTC - da2ce7 - Recorded the audit-record decisions of the PR #2428 goals-and-boundaries discussion under Decision Record (review 5400754664): the record as contract, stamped-or-checked copies of GitHub state, and `epic-owner`; the specifications discussion (review 5402713128) sends no work to this EPIC.
 - 2026-10-08 16:50 UTC - da2ce7 - Recorded the six proposals of the PR #2484 review retrospective in the matrix: items 1 (the manual sweep) and 2 adopted for order 5, item 1's tool deferred to order 8, item 3 deferred to order 10, and items 4 to 6 left to EPIC #2003's register. No fixed decision of the Decision Record changes.
 - 2026-10-08 17:28 UTC - da2ce7 - Answered review 5460432014 on PR #2494: the Decision Record, AC1 and References now cover the PR #2484 retrospective, and the matrix's pre-push sweep fails only on a current reference to a deleted or renamed path, allowing references marked historical.