docs(discussions): adopt the round template in the AI model provenance discussion - #2487

Records Copilot review 5455109937 (F1, F2) with their fixes, verification, and reply URLs.

…rust#2487 audit

Same defect that da2ce7 raised on the PR torrust#2485 audit (review-finding:pr-2485-f2): the Category item wrapped after documentation instead of after correctness as in the template's verbatim block.

…enance in commits

876dbf1 docs(pr-reviews): [#2003] record the PR #2486 reply URLs (Cameron Garnham)
c087841 docs(pr-reviews): [#2003] record the AI-assistance statement as the fix for R4 (Cameron Garnham)
82d4c63 docs(discussions): [#2003] state on each of my entries that AI assistance is not disclosed (Cameron Garnham)
7e70da7 docs(pr-reviews): [#2003] record the maintainer's review of PR #2486 (Cameron Garnham)
ee8be1d docs(discussions): [#2003] restate my automation-dependency reason and note the B4 revision (Cameron Garnham)
30b6014 docs(pr-reviews): [#2003] add the PR #2486 review audit (Cameron Garnham)
a4bbce2 docs(discussions): [#2003] state the pinning premise of my project-automation entry as a concept (Cameron Garnham)
9036a06 docs(discussions): [#2003] add my round on AI model provenance in commits (Cameron Garnham)

Pull request description:

  ## Summary

  This is my contribution round on the AI Model Provenance in Commits discussion: I add the position I first posted as comment 6035743101 on PR #2467 as attributed entries under eleven topics, re-framed from the number labels A1–A5, B1–B6 and C1–C7 into semantic labels that can be cited, accepted, amended or rejected one by one. Each entry now also ends with the AI-assistance statement in the maintainer's wording (R4).

  The maintainer invited the round on PR #2467 and on issue #2473, whose rounds convention merged in PR #2481. The branch sits on `develop` at `285938179`, the merge of the opening author's adoption round, so my entries sit under his topics.

  Related to #2003

  ## Changes

  Two files.

  `docs/discussions/2003-overhaul-guardrails-and-automation/20261007-ai-model-provenance-in-commits/README.md`:

  - Under each of the opening author's topics Q1–Q9, my entry replaces the `No positions yet.` placeholder under `#### Positions`, matched by topic number.
  - Two topics of my own follow his Q9: Q10 What a provenance record can truthfully contain, and Q11 The shape of the rule, each with my entry.
  - Each entry ends with the link to the position's first posting on PR #2467.
  - `tokenizer` added to the file's cspell ignore line; it occurs in the position text and has no other use in the repository.

  `docs/pr-reviews/pr-2486-review/PR-REVIEW.md`: my audit record of the review bot's one finding on this round (F1, the pinning premise of my `project-automation-is-a-dependency` entry) and its fix, which states that premise as a concept.

  | Topic | Entry labels (comment number) |
  | --- | --- |
  | Q1 | `granularity-by-role` (C1) |
  | Q2 | `models-whose-content-survives` (C2) |
  | Q3 | `identifier-from-the-harness` (C3) |
  | Q4 | `precedent-key-acceptable` (C3) |
  | Q5 | `harness-configuration-is-the-source-of-truth` (C4) |
  | Q6 | `human-edits-need-no-correction` (C4) |
  | Q7 | `format-checkable-completeness-not` (C5) |
  | Q8 | `no-provider-default-trailers-meanwhile` (C6) |
  | Q9 | `legal-input-parked` (C7) |
  | Q10 | `enumeration-has-no-boundary` (A1), `acceptance-is-the-choice` (A2), `personal-toolchains-are-private` (A3), `watermarks-mark-rendering-not-authorship` (A4), `regulation-targets-systems-acting-on-people` (A5) |
  | Q11 | `humans-author-and-certify` (B1), `tool-record-not-authorship` (B2), `scope-by-content-not-presence` (B3), `project-automation-is-a-dependency` (B4), `declared-not-detected` (B5), `provisional-and-regulation-independent` (B6) |

  ## Note for the maintainer

  Your adoption round moved the open questions into Topics Q1–Q9, so my round no longer carries topic texts of its own for them: each of my Q1–Q9 entries sits under your topic of the same number, in place of its placeholder, as the template intends for a topic's first position. Q10 and Q11 carry the premises and the proposed shape of the rule, which the nine questions do not cover.

  Nothing of your text is moved, copied or edited: your Context, your topic texts and the Outcome read as at `285938179`.

  Under the merge gate in `docs/discussions/AGENTS.md`, a contribution round merges on document quality (linters, factual claims about the repository, no edits to another participant's content or to the Outcome), not on agreement with its positions.

  ## Verification

  - Scope: against `develop` at `285938179`, the diff touches only the two files above. In the README, the only removed lines are the nine placeholders and the cspell ignore line, which returns with `tokenizer` added; a script that undoes the placements recovers your file byte for byte.
  - Entries: a script compared each of the eleven entries, from its heading through its "First posted as" line, with the branch before the rebase: all are byte-equal, and each `#####` heading sits under the `### Q<n>` of its number.
  - Verbatim text: before the rebase, a script extracted each labelled point from my comment, applied only the label substitutions (number labels to semantic labels in code form, "question N" to "QN") and compared the result with the README paragraph under the same label. Fifteen of the sixteen points rendered whole are identical; B4's opening sentence was revised afterwards for review finding F1 (the `project-automation-is-a-dependency` entry, authored 10:30 UTC on 2026-10-08, before the rebase onto the #2487 merge), so at the head that entry no longer matches the comment; for C3 and C4, split across Q3/Q4 and Q5/Q6, the only differences are the split titles and, for C3, its lead sentence repeated in both entries and its two-clause final sentence divided between them. Apart from the B4 revision, the entries are unchanged since.
  - Anchors and links: topic and entry anchors are unique; the eleven `#### Positions` headings repeat by the template's design and render as `positions` to `positions-10`. Every relative link and in-page anchor in the README resolves at the head; the comment URL on PR #2467 is external.
  - Gates at the head: pre-commit profile gate on the compute pod at this head, base `develop` `285938179`, exit 0 (76 s); `linter markdown`, `linter cspell`, `linter lychee` exit 0; `frontmatter-validator --all` with the same repository-wide counts as `develop` (19 errors, 4 warnings on unrelated specs, re-measured at `develop` in the same run); `validate-audit-record.py --pr-number 2486 --base 2859381` over the live review comments: 1 row, 7 log entries, 0 failures.

  ## After the maintainer's review

  Four commits follow the maintainer's review (5469325284); my audit record carries each finding.

  - `docs(discussions): [#2003] restate my automation-dependency reason and note the B4 revision`: for R1, the opening sentence of `project-automation-is-a-dependency` now gives the reason a project tracks what its automation uses rather than describing today's repository; for R2, the Q11 "First posted as" line notes that this entry revises `B4`'s opening sentence in this round (review findings F1 and R1). B4's opening sentence therefore now reads as the R1 text, not the comment's.
  - `docs(pr-reviews): [#2003] record the maintainer's review of PR #2486`: R1 and R2 fixed in the branch, R3 fixed in this description, R4 answered on its thread, R5 and R6 left for the reasons in their replies.
  - `docs(discussions): [#2003] state on each of my entries that AI assistance is not disclosed`: for R4, after the maintainer's follow-up on its thread, each of my eleven entries ends with the statement in his wording, "AI assistance not disclosed: the author's private tooling may include machine-learning tools."
  - `docs(pr-reviews): [#2003] record the AI-assistance statement as the fix for R4`: the R4 row and its detail block record the follow-up and the fix.

  ## Validation at the new head

  Gates at the heads: pre-commit profile gate on the compute pod at `c087841d2` (before the replies) and at `876dbf121` (with the reply URLs recorded), base `develop` `f960c78d7`, exit 0 (79 s and 77 s); `linter markdown`, `linter cspell`, `linter lychee` exit 0; `frontmatter-validator --all` with the same repository-wide counts as `develop` (19 errors, 4 warnings on unrelated specs, re-measured at `develop` in the same run); `validate-audit-record.py --pr-number 2486 --base f960c78` against the review comments captured after the replies were posted: 7 rows, 15 log entries, 0 failures (before the replies, the only failures were the five Reply URL fields then pending).

ACKs for top commit:
  josecelano:
    ACK 876dbf1

Tree-SHA512: 230344175395473e7093837ac5f123878ca42b4377ea9e4de88fa80137bfefbe798c4f165dd52227072eb5e105c31249880951db57d1530eee7bc7a5633f26ad