psyclaw 0.27.22 → 0.28.0

package/vendor/ars/academic-paper/references/disclosure_mode_protocol.md

@@ -0,0 +1,478 @@
+# Disclosure Mode Protocol
+
+**Status**: v3.2 + #596 venue-intake extension (#108 anchor-specific renderer unchanged; shared intake-integrity gate extended to both tracks)
+**Parent skill**: `academic-paper`
+**Mode name**: `disclosure`
+**Purpose**: Generate either (a) a venue-specific AI-usage disclosure bundle aligned to the policy snapshot recorded by ARS (v3.2 path, default), or (b) a policy-anchor-specific disclosure rendered from the 4-anchor matrix (PRISMA-trAIce / ICMJE / Nature / IEEE) when the author targets a policy anchor rather than a specific journal venue (#108 path). This mode renders AI-use disclosure; it does not certify that the complete submission package satisfies every reporting requirement of the venue.
+
+---
+
+## Two parallel tracks (#108 + v3.2)
+
+The `disclosure` mode dispatches on the author-supplied selector:
+
+| Selector | Track | Lookup source | Output shape |
+|---|---|---|---|
+| `--venue=<v>` (v3.2, default) | Venue track | `venue_disclosure_policies.md` database (v2, 15 policy targets: ICLR / NeurIPS / Nature / Science / ACL / EMNLP + the #596 medical-publishing set — ICMJE / NEJM / The Lancet / JAMA / BMJ / PLOS / Frontiers / 中华护理杂志社 Chinese Nursing Journals Publishing House / 国际眼科杂志 International Eye Science) | Applicability/status bundle: `REQUIRED`, `ACTION_ONLY`, `NOT_REQUIRED`, or `UNKNOWN`, plus an explicit typed halt status when rendering cannot proceed |
+| `--policy-anchor=<a>` (#108) | Anchor track | `policy_anchor_table.md` 4-anchor × 16-field matrix | 4-anchor-conditioned render per `policy_anchor_disclosure_protocol.md` |
+
+The two tracks are **selector-mutually-exclusive by default** — one selector picks one track. When the author supplies **both** selectors in the same invocation, the renderer evaluates compatibility per concern #7 rules: a consistent pair (Nature venue + nature anchor, both sourced from `shared/policy_data/nature_policy.md`) proceeds; any other pair is **rejected with an explicit error** listing the policy conflict. Silent precedence between selectors is forbidden. See [policy_anchor_disclosure_protocol.md §5](policy_anchor_disclosure_protocol.md) for the full conflict-resolution detail.
+
+If neither selector is supplied and the pipeline orchestrator does not infer one from upstream context, the mode prompts the user to specify which selector applies. The venue track remains the default for explicit journal submissions; the anchor track applies when targeting policy frameworks (e.g., compliance reporting to ICMJE-adopting journals collectively, or pre-submission alignment to IEEE author guidelines).
+
+**Conflict resolution (concern #7) — exhaustive cases:**
+- Supplied both, **consistent pair** (only currently defined case): `--venue=Nature` (any Nature Portfolio variant string) **and** `--policy-anchor=nature` → both target Nature substantive policy via the shared source pointer → **proceed**.
+- Supplied both, **any other combination** (e.g., `--venue=Nature` + `--policy-anchor=ieee`; `--venue=ICLR` + `--policy-anchor=icmje`; or a Nature-venue spelling that does not match the canonical set with a non-nature anchor) → **reject** with explicit error citing the policy conflict; require the user to drop one selector. Silent precedence is forbidden by §4.4 #7.
+- Supplied only one selector → run that track.
+- Supplied neither selector → prompt the user to specify.
+
+---
+
+## Why this mode exists
+
+`academic-paper` already ships two generic AI disclosure examples in `journal_submission_guide.md` ("Minimal Disclosure" and "Detailed Disclosure"). They are venue-agnostic educational examples, not a fallback inside this mode: they don't know that Nature requires disclosure in the Methods section specifically, that ICLR requires it in the paper body with acknowledgement that "LLMs were used as general-purpose writing tools", or that ACL requires the disclosure in the Acknowledgements section. A venue-track lookup failure therefore halts rather than silently using either example.
+
+The v3.2 venue track closes the venue-specific gap. The #108 anchor track closes the policy-framework-specific gap that emerges when authors target a policy anchor (PRISMA-trAIce SLR guideline, ICMJE recommendations, Nature Portfolio editorial policy, IEEE author guidelines) rather than a specific journal venue.
+
+---
+
+## Inputs
+
+1. **Paper draft**: current manuscript text (the mode needs to know what the AI actually did in order to describe it accurately).
+
+2. **Selector** (one of):
+   - **Target venue/policy target (`--venue=<v>`)**: journal, conference, publisher-wide policy, or umbrella recommendation label (v3.2 path; the option name remains `--venue` for compatibility). If the target is in the database (v2: ICLR, NeurIPS, Nature, Science, ACL, EMNLP, plus the medical-publishing targets ICMJE, NEJM, The Lancet, JAMA, BMJ, PLOS, Frontiers, Chinese Nursing Journals Publishing House 中华护理杂志社, International Eye Science 国际眼科杂志), use the cached policy. If not, refuse to guess. The user may paste the target's current official policy as evidence for manual review/curation, but pasted prose does not create an executable session policy and the venue-aligned renderer remains halted until aliases, predicates, required facts, and placement channels are curated together. Canonicalize only explicit aliases recorded below; do not silently map a publisher or journal family to one member journal.
+   - **Policy anchor (`--policy-anchor=<a>`)**: one of `prisma-trAIce, icmje, nature, ieee` (#108 path). Anchor lookup follows `policy_anchor_disclosure_protocol.md`.
+
+3. **Pipeline signal** (#108 anchor path only): `slr_lineage=true|false` set by the upstream pipeline orchestrator. Required for `--policy-anchor=prisma-trAIce` per §4.3 G2 invariant. Cold-start invocation requires explicit `mode=<value>` parameter; silent fallback to general track is forbidden.
+
+4. **What AI use occurred**: the mode reads the paper's commit history / pipeline log (if using the full `academic-pipeline`) and separately asks the author to report AI use outside ARS. Intake is the union of logged and author-reported use; a missing log entry never erases an external use. Require the author-attested `external_use_inventory_confirmed` state (`KNOWN(none)` or `KNOWN(list supplied)`); `UNKNOWN` halts before categorization is finalized even when pipeline logs exist. On the venue path this is `UNKNOWN/HALTED/UNRESOLVED_INPUT`; the anchor path emits its own intake-pending annotation. Collect the complete Phase 2 category table, including citation checking, editing, analysis, visual/artwork/media assistance, and the `OTHER / UNCLASSIFIED` catch-all; do not maintain a shorter intake enum. If the pipeline log is not available, ask the user to confirm every category as well as the external-use inventory.
+
+5. **Venue-required disclosure facts** (#596 venue path only): collect the fields selected by the venue matrix in Phase 2b. Each field has state `KNOWN`, `NOT_APPLICABLE`, or `UNKNOWN` and a value/evidence pointer when `KNOWN`. Pipeline logs and session metadata may prefill facts, but the renderer must show the prefilled value to the author for confirmation; it must not infer a missing fact from manuscript prose or invent a value. The anchor path keeps its separate field contract in `policy_anchor_disclosure_protocol.md`.
+
+### Venue selector aliases added by #596
+
+Match case-insensitively after trimming surrounding whitespace. The left-hand
+label is the canonical database heading; the remaining values are accepted
+aliases. Any value not listed here or in the pre-v2 selector set is unknown.
+
+| Canonical target | Accepted aliases |
+|---|---|
+| BMJ | `The BMJ` |
+| Chinese Nursing Journals Publishing House | `中华护理杂志社` |
+| Frontiers | `Frontiers journals` |
+| ICMJE | `International Committee of Medical Journal Editors` |
+| International Eye Science | `国际眼科杂志` |
+| JAMA | `Journal of the American Medical Association` |
+| NEJM | `The New England Journal of Medicine`, `New England Journal of Medicine` |
+| PLOS | `PLOS journals`, `PLOS ONE` |
+| The Lancet | `Lancet` |
+
+`JAMA Network` without the exact journal name and an arbitrary `Frontiers in
+...` title are not silently treated as JAMA or Frontiers. Ask for the exact
+target or current target policy. This prevents a family/publisher label from
+quietly inheriting one member journal's instructions.
+
+---
+
+## Venue-path result envelope
+
+Every invocation that selects the venue path records `disclosure_outcome` as
+`REQUIRED`, `ACTION_ONLY`, `NOT_REQUIRED`, or `UNKNOWN`. This is the
+venue-policy applicability result, not the exhaustive execution state.
+Separately record `execution_status` as `READY` or `HALTED` and, for a halt, one
+typed `halt_reason`: `UNRESOLVED_INPUT`, `PROHIBITED_USE`,
+`INCOMPATIBLE_FACT`, `CONTRACT_GAP`, `UNCURATED_POLICY`, or
+`POLICY_SCOPE_GAP`.
+
+An unresolved applicability fact yields `UNKNOWN/HALTED/UNRESOLVED_INPUT`. A
+known prohibited or incompatible fact retains the applicability result
+established so far and sets the corresponding halt reason; do not relabel known
+prohibited conduct as epistemically `UNKNOWN`. `REQUIRED` produces disclosure
+blocks only with `execution_status=READY`.
+
+This envelope does not replace the anchor track's independent output contract.
+Selector conflicts and a missing selector are dispatch errors before a track is
+selected, not venue outcomes.
+
+---
+
+## Process
+
+### Phase 1: Intake + lookup (selector-aware)
+
+**Step 1a — selector dispatch:**
+- Both `--venue=<v>` and `--policy-anchor=<a>` supplied → check policy compatibility per the Two-parallel-tracks section above. **Consistent pair (currently only any Nature Portfolio venue + `--policy-anchor=nature`, where "Nature Portfolio venue" includes canonical labels {"Nature", "Nature Portfolio", "Nature (Nature Publishing Group)", "Nature Publishing Group"} and the journal-family prefix `"Nature "` matching e.g. "Nature Medicine", "Nature Communications", "Nature Climate Change", etc.) → route the consistent pair to **step 1c (anchor path)** so the shared canonical source `shared/policy_data/nature_policy.md` drives rendering; step 1b's venue database (v2) does not need to contain every Nature Portfolio journal**. Conflicting pair → reject with explicit error.
+- `--venue=<v>` only → step 1b (venue path).
+- `--policy-anchor=<a>` only → step 1c (anchor path).
+- Neither supplied → prompt the user to specify selector.
+
+**Step 1b — venue lookup (v3.2 + #596 venue path):**
+- If venue is in the database (v2) → load policy from `venue_disclosure_policies.md`.
+- If venue is unknown → return `UNKNOWN/HALTED/UNCURATED_POLICY` and print: "I do not have a curated executable policy for {venue}. You may paste the venue's current official AI policy for manual review, but I cannot render a venue-aligned bundle until its selectors, predicates, required facts, and placement channels are curated together." Do NOT fabricate or infer an executable policy.
+- A pasted policy remains session evidence only for human review or a future curated database update; it does **not** enter Phase 2a/2b, produce a checklist, or authorize disclosure rendering. Do NOT auto-persist it — policies drift, and the database needs curation.
+
+**Step 1c — anchor lookup (#108 path):**
+- Validate `--policy-anchor=<a>` ∈ `{prisma-trAIce, icmje, nature, ieee}`. Other values → reject with the closed-enum error.
+- For `--policy-anchor=prisma-trAIce`: confirm `slr_lineage=true` (pipeline signal) or `mode=systematic-review` (cold-start input) per the G2 invariant track gate. Otherwise refuse with G2 invariant citation.
+- Reuse only Phase 2's generic AI-usage category shape, then delegate the anchor-specific decision, input expansion, rendering, and placement contract to `policy_anchor_disclosure_protocol.md`. Do not run venue Phase 2a/2b or venue placement rules on the anchor track.
+
+### Phase 2: Categorize AI usage
+
+Produce a categorized list of how AI was used in the manuscript:
+
+| Category | Examples |
+|---|---|
+| Research assistance | Literature search, annotated bibliography, claim verification |
+| Citation checking | Reference/source verification, citation accuracy/completeness checking, claim-to-source checking (distinct from citation-format conversion) |
+| Drafting assistance | Section drafting, paraphrasing, outline generation |
+| Revision assistance | Reviewer response drafting, tracked changes, consistency checking |
+| Editing assistance | Grammar, style, formatting, citation format conversion |
+| Analysis assistance | Not applicable to pure writing flows; flag if the paper reports any analysis the AI did |
+| Visual / artwork / media assistance | Explanatory figures, data visualizations, primary-research images, research-method images, clinical illustrations, graphical abstracts, cover art, or other submitted media |
+| Peer review simulation | `academic-paper-reviewer` was used on the draft pre-submission |
+| Other / unclassified AI use (logged or external) | Preserve the source and the author's verbatim description of any use that does not fit the rows above; do not force-map it from prose |
+
+For each category, mark: USED / NOT USED / UNCERTAIN. UNCERTAIN items require user confirmation before the disclosure text is finalized.
+
+On either track, a `USED` `OTHER / UNCLASSIFIED` category preserves the
+author's verbatim description and halts before venue Phase 2a or anchor
+evaluation until the author maps it to an executable category. It is never
+dropped or relabelled without author confirmation. The venue path returns
+`UNKNOWN/HALTED/UNRESOLVED_INPUT`; the anchor path returns the intake-pending
+annotation defined by its own protocol.
+
+For the #596 venue track, create one author-confirmed use record per distinct
+`tool × task/run/artifact` for every `USED` category. Each record retains the
+tool identity, purpose, category, affected section or artifact, whether the use
+was manuscript preparation or part of the research, `operations[]` from the
+closed set `GENERATED`, `SUBSTANTIVELY_DRAFTED`, `EDITED`, `ANALYSED`,
+`SEARCHED`, `CHECKED_CITATIONS`, `FORMATTED_CITATIONS`, `OTHER_CONFIRMED`, and
+`affected_targets[]` from the closed set `WHOLE_PAPER`, `TITLE`, `ABSTRACT`,
+`INTRODUCTION_OR_BACKGROUND`, `METHODS`, `RESULTS`, `DISCUSSION`, `CONCLUSION`,
+`RESULT_INTERPRETATION`, `CORE_ARGUMENT`, `INNOVATION_CLAIM`,
+`RESEARCH_FIGURE_OR_MEDIA`, `ORIGINAL_RESEARCH_DATA`, `RESEARCH_PROCESS`,
+`SUPPORTING_DATA_FILE`, `REFERENCE_OR_CITATION`, `CODE`, `PEER_REVIEW_MATERIAL`,
+`OTHER_SUBMISSION_TEXT`, `OTHER_CONFIRMED`.
+Classify operations by what the tool actually did, not by the high-level
+category label: a rewrite that creates, replaces, or restructures substantive
+manuscript content is `SUBSTANTIVELY_DRAFTED` even when the category is
+Revision assistance; `EDITED` is limited to non-substantive surface correction
+that does not supply or replace scientific claims, reasoning, interpretation,
+or conclusions. If that boundary is uncertain, keep the operation unresolved
+and halt rather than choosing the less restrictive label.
+Every `OTHER_CONFIRMED` operation or target carries the author's verbatim
+description and stays unresolved for any venue predicate that depends on it
+until the author maps it or supplies the predicate-specific confirmation and
+basis. Do not collapse multiple tools, prompt runs, figures, or research tasks
+into a single generic use. These records feed both the prohibited-use
+discriminators and the venue-required fact ledger. A mapped record must also
+resolve its operation and target before Phase 2a. The anchor track keeps its
+separate per-anchor input contract after the shared unclassified-use gate
+above.
+
+### Phase 2a: Decide whether the venue requires a disclosure (#596 venue path)
+
+Populate the venue-path result envelope defined above. Base applicability on
+confirmed uses, not merely on the selected venue.
+
+Resolve Phase 2a.1 for every triggered use record before finalizing this
+decision. A visual-only, data-only, or research-method use participates in
+applicability even when all writing categories are `NOT USED`; it must not be
+classified `NOT_REQUIRED` before its venue-specific predicates are resolved.
+
+- `REQUIRED`: at least one confirmed use falls within the target's disclosure
+  rule.
+- `ACTION_ONLY`: no confirmed use requires manuscript disclosure text, but a
+  recorded policy condition requires an author action or permission check. Emit
+  the audit ledger and a labelled action checklist with zero disclosure blocks.
+  In this database, the defined case is Elsevier AI-assisted cover art after its
+  permissions/rights gate. If another use in the same invocation requires
+  disclosure, the aggregate result is `REQUIRED` and the cover-art checklist is
+  included alongside the relevant blocks.
+- `NOT_REQUIRED`: every category is confirmed and either no AI was used or every
+  use falls entirely within an explicit policy exemption. Emit the decision and
+  policy basis; do not generate a disclosure paragraph.
+- `UNKNOWN`: a category or an exemption predicate is unresolved. Halt and ask
+  only for the fact needed to decide applicability.
+
+Explicit exemption predicates relevant to this database include:
+
+| Target | `NOT_REQUIRED` exemption (only when this is the complete AI use) |
+|---|---|
+| ACL / EMNLP | Language-only paraphrasing/polishing, predictive-keyboard input, or literature-search assistance covered by ACL's no-special-disclosure cases |
+| JAMA | Basic grammar or spelling checking only |
+| Nature | AI-assisted copyediting only, as defined by the canonical Nature policy: readability/style/error correction of human-generated text, excluding generative editorial work and autonomous content creation |
+| The Lancet / Elsevier | Basic grammar, spelling, or punctuation checking only; or specialist assistive technology used solely for accessibility |
+
+For every other target, `NOT_REQUIRED` is available when all categories are
+confirmed `NOT USED`; never invent an exemption absent from the recorded policy.
+If a confirmed use is prohibited by the target, halt with the prohibited-use
+finding instead of rendering language that could make the use acceptable.
+
+#### Phase 2a.1: Resolve explicit prohibited-use discriminators
+
+As part of Phase 2a, resolve every decision fact below when its trigger is
+true. These facts decide whether rendering is permitted; they are audit-ledger
+inputs, not prose to infer from a tool brand, manuscript text, or a broad use
+category. Give each predicate `KNOWN` or `UNKNOWN` state and retain its value
+and author-confirmed evidence basis. `UNKNOWN` halts before Phase 2b; a known
+prohibited combination halts with the policy finding.
+
+Before the target-specific checks below, record the manuscript-level
+`ai_listed_or_proposed_as_author` fact as `KNOWN(true)`, `KNOWN(false)`, or
+`UNKNOWN` for every target whose cached `Authorship rule` explicitly says that
+AI does not qualify or must/should not be listed (all current targets except
+PLOS). `UNKNOWN` halts as `UNRESOLVED_INPUT`. `KNOWN(true)` retains the
+applicability result established from the use records but halts as
+`INCOMPATIBLE_FACT`, refuses to produce author-list text, and asks the author to
+correct the list; it is not silently converted into a disclosure sentence.
+This manuscript-author fact is distinct from the per-citation
+`ai_cited_as_author` predicates below. The current PLOS evidence row does not
+state an AI-authorship rule, so this protocol does not invent one for PLOS.
+
+For Frontiers and the two Chinese-language policy targets, first record the
+author-confirmed `policy_tool_scope` of every use record as `GENAI_OR_AIGC`,
+`OTHER_AI`, or `UNKNOWN`; never infer it from a brand name. For Frontiers,
+`GENAI_OR_AIGC` means the generative-AI scope stated by that policy, including
+LLMs and text-to-image generators. `UNKNOWN` halts. `OTHER_AI` does not trigger
+the GenAI/AIGC-specific children below. If any record is `OTHER_AI`, preserve a
+separate ledger for the records already classified, return
+`UNKNOWN/HALTED/POLICY_SCOPE_GAP`, and direct the author to the target's current
+non-generative-AI instructions. Do not render a partially evaluated bundle for
+a mixed `GENAI_OR_AIGC` + `OTHER_AI` inventory. Only an all-`GENAI_OR_AIGC`
+inventory proceeds through the target-specific children below.
+
+| Target | Trigger | Decision facts and outcome |
+|---|---|---|
+| Chinese Nursing Journals Publishing House (中华护理杂志社) | A `GENAI_OR_AIGC` use record is `USED` | For each record, confirm `ai_performed_scientific_or_intellectual_contribution`, `generated_research_figure_or_media`, `altered_original_research_data_process_or_results`, and `used_unverified_genai_reference`. The first predicate is forced true when `GENERATED` or `SUBSTANTIVELY_DRAFTED` is paired with `WHOLE_PAPER`, `TITLE`, `ABSTRACT`, `INTRODUCTION_OR_BACKGROUND`, `METHODS`, `RESULTS`, `DISCUSSION`, `CONCLUSION`, `RESULT_INTERPRETATION`, `CORE_ARGUMENT`, `INNOVATION_CLAIM`, `RESEARCH_PROCESS`, or another target the author confirms is a scientific contribution or intellectual-labour product. For every other operation/target combination — including `ANALYSED` contributions and all uses affecting `OTHER_SUBMISSION_TEXT`, `CODE`, `REFERENCE_OR_CITATION`, `ORIGINAL_RESEARCH_DATA`, or `OTHER_CONFIRMED` — that predicate stays `UNKNOWN` until the author confirms, with a basis, whether the AI performed scientific contribution or intellectual labour; it cannot pass merely because an operation or target is outside the illustrative list. The second predicate is forced true when `GENERATED` is paired with `RESEARCH_FIGURE_OR_MEDIA`; the third is forced true when the record says original research data, process, or results were altered; and the fourth is forced true when a GenAI reference was used without verification. A contradictory mapping is `INCOMPATIBLE`; any true predicate is a prohibited-use halt. |
+| Frontiers | A `GENAI_OR_AIGC` use record is `USED` | Preserve whether the content was created or edited and whether it is written or visual so Phase 5 can select the applicable pre-submission actions. For a figure, also preserve the author-confirmed `figure_represents_data` routing fact. These are action-routing facts, not prohibited-use predicates or Phase-2 render prerequisites: an unresolved action or routing state remains on the Phase-5 checklist and does not halt disclosure rendering. `OTHER_AI` records never enter this GenAI-specific action path. |
+| ICMJE | A use record affects `REFERENCE_OR_CITATION`, or the author reports citing AI-generated material or an AI tool | Separately confirm `ai_generated_material_used_as_primary_source` and `ai_cited_as_author`; `KNOWN(true)` for either is a prohibited-use halt. `UNKNOWN` or an `OTHER_CONFIRMED` reference target without both applicable statuses resolved halts. |
+| International Eye Science (国际眼科杂志) | A `GENAI_OR_AIGC` use record is `USED` | For each use record, classify `use_scope` as `LANGUAGE_POLISHING`, `LITERATURE_RETRIEVAL`, `DATA_ORGANIZATION`, `CHART_ANNOTATION`, `OTHER_CONFIRMED_NON_CORE_RESEARCH_STEP`, or `CORE_RESEARCH_STEP`. Record the author-confirmed basis for an `OTHER_CONFIRMED` classification; `CORE_RESEARCH_STEP` is prohibited and an unknown scope halts. For each tool, also record `tool_is_overseas`. If false, the qualification child is `NOT_APPLICABLE`. If true, record `lawful_compliance_qualification` and its verification basis; `KNOWN(false)` is a prohibited-use halt. Never infer origin, qualification, or research-step scope from the product name, provider, or manuscript prose. For every record confirm `generated_core_main_text_conclusion_analysis_viewpoint_or_innovation_claim`, `fabricated_experimental_plan_technical_route_or_citation`, `replaced_author_in_experimental_design_or_data_validation`, `fabricated_data_invented_results_or_tampered_conclusions`, `rewrote_plagiarized_work_to_evade_detection`, `generated_peer_review_response_grant_contribution_or_integrity_statement`, and `uploaded_secret_research_data_or_unpublished_results_to_public_ai_platform`. Also record `use_involves_data`; `DATA_ORGANIZATION` and `CHART_ANNOTATION` force it true, and a contradictory false value is `INCOMPATIBLE`. For every other scope, an explicit false value makes the data-detail children `NOT_APPLICABLE`, while a true value requires data types, verification status, and `data_involves_clinical_or_case_data` in Phase 2b. A true clinical/case predicate additionally requires de-identification measures. Confirm `aigc_generated_or_tampered_data`, `aigc_replaced_core_analysis`, `uploaded_undeidentified_data_to_aigc`, `uploaded_data_lacking_required_ethics_review_to_aigc`, and `fabricated_data_or_ethics_proof`; `KNOWN(true)` for any prohibited predicate is a prohibited-use halt. |
+| JAMA | A use record has `GENERATED` or `SUBSTANTIVELY_DRAFTED` and `affected_targets[]` contains one of `WHOLE_PAPER`, `TITLE`, `ABSTRACT`, `INTRODUCTION_OR_BACKGROUND`, `METHODS`, `RESULTS`, `DISCUSSION`, `CONCLUSION`, `RESULT_INTERPRETATION`, `CORE_ARGUMENT`, `INNOVATION_CLAIM`, or `OTHER_SUBMISSION_TEXT`, regardless of whether its category is Drafting or Revision; or a record has any `OTHER_CONFIRMED` operation/target | For an `OTHER_CONFIRMED` trigger, evaluate `jama_generated_or_substantively_drafted_submission_text` with the author's basis whether its value is initially known or unknown; `UNKNOWN` halts and false makes the manuscript-class child facts `NOT_APPLICABLE`. The ordinary generated/substantively-drafted manuscript-target combination forces that predicate true. When true, retain the author's exact target-level `submission_type` and separately record `jama_submission_type_is_prohibited` as `KNOWN(true)`, `KNOWN(false)`, or `UNKNOWN`, linking both facts to every triggering use record. The predicate is forced true for `OPINION_MANUSCRIPT`, `LETTER_TO_THE_EDITOR`, `ONLINE_COMMENT`, `A_PIECE_OF_MY_MIND`, or `POETRY`; `KNOWN(true)` is a prohibited-use halt, `KNOWN(false)` continues with the exact non-prohibited type and author-confirmed basis, and `UNKNOWN` halts. Do not classify the manuscript type or predicate from its prose. |
+| JAMA | A use record affects `RESEARCH_FIGURE_OR_MEDIA` or any submitted clinical image/illustration, regardless of its high-level category | For each triggering record, confirm `clinical_image_or_illustration_created_or_manipulated`. If false, its formal-research child is `NOT_APPLICABLE`. If true, record `part_of_formal_research_design_or_methods`; `KNOWN(false)` is a prohibited-use halt and `KNOWN(true)` continues to the method and rights facts in Phase 2b. |
+| Nature | A venue-only use record is `EDITED` or `FORMATTED_CITATIONS` and affects human-generated manuscript text or reference formatting | Confirm `nature_ai_assisted_copyediting_only` against the canonical definition in `shared/policy_data/nature_policy.md`. If true and this is the complete use, return `NOT_REQUIRED`; if false, continue as disclosable use; `UNKNOWN` halts. Generated editorial work or autonomous content creation cannot satisfy the predicate. |
+| Nature | A venue-only use record affects `RESEARCH_FIGURE_OR_MEDIA` or any submitted image/media | Return `UNKNOWN/HALTED/CONTRACT_GAP` before Phase 2b with diagnostic code `NATURE_VENUE_IMAGE_CONTAINMENT`. The stale venue summary is not executable for images; direct the author to the canonical Nature policy/anchor path (`--policy-anchor=nature`) rather than permitting or disclosing the image from the venue row. This containment does not rewrite the maintainer-owned Nature evidence row. |
+| NEJM | A use record affects `REFERENCE_OR_CITATION`, or the author reports citing AI-generated material | Confirm `ai_generated_material_used_as_primary_source`; `KNOWN(true)` is a prohibited-use halt. `UNKNOWN` or an unresolved `OTHER_CONFIRMED` reference target halts. |
+| PLOS | A use record affects primary research data, results, or supporting data files | Confirm `ai_fabricated_or_misrepresented_primary_research_data`; `KNOWN(true)` is a prohibited-use halt and `UNKNOWN` halts. |
+| The Lancet / Elsevier | Any `USED` use record | Confirm `ai_replaced_authors_intellectual_contribution` with the author's basis for every operation/target combination, including `SUBSTANTIVELY_DRAFTED` use affecting `CORE_ARGUMENT`, `RESULT_INTERPRETATION`, `INNOVATION_CLAIM`, or `CONCLUSION`; operation and target alone do not prove replacement when genuine author contribution was retained. `KNOWN(true)` is a prohibited-use halt; `UNKNOWN` halts. If a use affects `REFERENCE_OR_CITATION` or the author reports citing an AI tool, separately confirm `ai_cited_as_author`; `KNOWN(true)` is prohibited and `UNKNOWN` halts. |
+| The Lancet / Elsevier | A use record affects research data, results, references, or figures | Confirm the applicable predicates `ai_fabricated_or_altered_results`, `ai_invented_or_altered_underlying_data`, `ai_fabricated_or_altered_references`, and `figure_not_faithfully_derived_from_underlying_data_and_methods`; mark children `NOT_APPLICABLE` only from explicit false target predicates. Any applicable `KNOWN(true)` is a prohibited-use halt; `UNKNOWN` halts. |
+| The Lancet / Elsevier | A use record affects `RESEARCH_FIGURE_OR_MEDIA` or any submitted visual/media artifact, regardless of its high-level category | First record the positive predicate `generated_media_duplicates_or_refers_to_protected_subject`, where a protected subject is an existing copyrighted image, a real person, another party's identifiable product/brand, or an individual's voice likeness; `KNOWN(true)` is prohibited. Next classify non-primary artifacts as `EXPLANATORY_IMAGE`, `DATA_VISUALIZATION`, `GRAPHICAL_ABSTRACT`, or `COVER_ART`. For a record involving primary observed/experimental data, do not preselect a class: collect `ai_is_formal_research_design_or_method`, `image_output_directly_obtained_in_research_through_that_method`, and reproducible-method details, then resolve it to exactly one of `PRIMARY_RESEARCH_IMAGE` or `RESEARCH_METHOD_IMAGE`. If both predicates are confirmed true, use `RESEARCH_METHOD_IMAGE` even when the output concerns primary data. If AI created or altered an image presented as primary evidence but the output was not directly obtained in the research through a confirmed formal method, use `PRIMARY_RESEARCH_IMAGE` and halt as prohibited. Unknown provenance/method facts halt; do not use the method label without the two affirmative facts and reproducible detail. An unmodelled nonvisual-media use halts as a contract gap after the protected-content check. For every otherwise permitted submitted visual, confirm accuracy and originality; record whether it is based on existing artwork/graphics and, when true, the rights-holder permission and attribution. `KNOWN(false)` for an applicable accuracy/originality/rights fact is incompatible. A `DATA_VISUALIZATION` must be confirmed as directly derived from underlying data through a reproducible method; `KNOWN(false)` is prohibited/incompatible. A `GRAPHICAL_ABSTRACT` record must confirm `graphical_abstract_used_ai_or_ai_assisted_illustration=true`; false is `INCOMPATIBLE` with its AI-use artifact classification and requires correction/re-evaluation, while `UNKNOWN` halts. Then record `graphical_abstract_tool_class` as `GENERAL_PURPOSE_GENERATIVE_AI_IMAGE_TOOL` or `DEDICATED_SCIENTIFIC_OR_PROFESSIONAL_ILLUSTRATION_TOOL`; the first is prohibited and the second continues to Phase 2b. For `COVER_ART`, additionally collect prior permission from both the journal editor and publisher, `cover_art_contains_third_party_material`, and appropriate content-attribution details. A false third-party predicate makes only its permission child `NOT_APPLICABLE`; a true predicate requires the corresponding permission evidence. Content attribution remains an affirmative action fact; if no attribution is applicable, record `KNOWN(none_applicable)` with the author-confirmed basis rather than `NOT_APPLICABLE`. An unknown/false required permission or unknown attribution fact halts, and a cover-art-only invocation resolves to `ACTION_ONLY`. |
+
+The `KNOWN` / `NOT_APPLICABLE` / `UNKNOWN` and incompatibility rules in
+Phase 2b apply to these render and prohibition discriminators as well, except
+for Frontiers facts explicitly designated only as non-blocking Phase-5 action
+routing. A halt here reports the facts already established but does not draft
+language that could imply a prohibited use becomes acceptable through
+disclosure.
+
+JAMA's statement that AI, LLMs, and chatbots **should not** generate or format
+references is policy guidance, not the same modal strength as its explicit
+"not permitted" uses. Surface a non-blocking advisory warning when that use is
+reported, but do not turn the sentence into either a prohibited-use halt or a
+disclosure-rendering prerequisite.
+
+Chinese Nursing Journals Publishing House separately says that another author's
+content already labelled as AI-generated generally should not be cited as an
+original source and, when genuinely necessary, should be explained. When such
+a citation is reported, record its necessity and explanation when available and
+surface a non-blocking advisory/action note. Do not merge this qualified rule
+into the hard `used_unverified_genai_reference` predicate, and do not make an
+unknown explanation block the AI-use disclosure renderer.
+
+### Phase 2b: Build the venue-required fact ledger (#596 venue path)
+
+Run this phase only when Phase 2a returns `REQUIRED` with
+`execution_status=READY`. A terminal halt never continues into fact expansion.
+Select the target row below and build a field-level ledger. "Required" here
+means required before ARS may render a venue-aligned AI-use disclosure bundle;
+it does not certify the rest of the submission or convert policy advice into a
+prohibition. A conditional field becomes required only when its predicate is
+confirmed `true`.
+
+| Venue | Required facts before render | Conditional facts |
+|---|---|---|
+| ACL | tool name; specific AI-produced content or task; affected section/content | For low-novelty generated text: author confirmation that output accuracy and source/idea citations were checked |
+| BMJ | AI technology; why it was used; how it was used | Research-related use: method-level description |
+| Chinese Nursing Journals Publishing House (中华护理杂志社) | tool/service name; purpose; affected text/figure/code; human review/editing performed; author acceptance of full responsibility | Use affecting research methods: method-level description |
+| EMNLP | Same fact contract as ACL | Same conditional facts as ACL |
+| Frontiers | For each `GENAI_OR_AIGC` record: tool name; version; model; source/provider; whether content was AI-produced or AI-edited; affected written/visual content | Use that forms part of the research method: method-level description. The factual-accuracy, plagiarism-free, and figure accuracy-to-data checks are Phase-5 pre-submission actions, not fields required before render. Phase 2b receives only an all-`GENAI_OR_AIGC` inventory; any `OTHER_AI` record has already halted the run in Phase 2a.1. |
+| ICLR | tool name; specific assisted tasks; author acceptance of full responsibility | None |
+| ICMJE | description of the AI-assisted technology; how it was used | AI-generated quoted material: attribution and full-citation details. Exact tool name is ARS-recommended metadata, not an ICMJE-only render blocker. Human review/editing remains policy advice and responsibility practice, not a separate render-blocking condition. |
+| International Eye Science (国际眼科杂志) | tool name; version; purpose; scope; proportion of generated content | When `use_involves_data` is true: data types; verification status; `data_involves_clinical_or_case_data`. When the clinical/case predicate is true: de-identification measures. This covers data organization, chart annotation, and every other data-involving use. The applicable data-upload and fabricated-proof prohibitions must already be false in Phase 2a.1. |
+| JAMA | affected use class; author review/accuracy confirmation; author acceptance of responsibility for content integrity | AI-assisted content creation, revision, or formatting beyond basic grammar/spelling: platform/program/tool name; model/tool version; per-tool/use `extension_numbers_applicable`; extension number(s) only when that predicate is true; manufacturer; date(s) of use; what/how/affected portions. `ai_used_in_scientific_study` is the same author-confirmed per-use research-use field carried by the Phase-2 record: a research-use record forces it true, a manuscript-preparation-only record forces it false, and any contradiction is `INCOMPATIBLE`. When true: specific research use plus the three predicates `study_uses_llm`, `copyright_protected_content_entered`, and `ai_generated_content_included_in_submission`. Only when `study_uses_llm` is true: platform/program/tool name and version; manufacturer; date(s); prompt(s), prompt sequence, and prompt revisions. If `copyright_protected_content_entered` is true: (a) a copy/evidence pointer for the copyright-holder permission or license that must accompany the submission and (b) a separate description of that permission/license for Methods. If `ai_generated_content_included_in_submission` is true: affected item/content type plus the publication-rights or permission basis **as determined by the AI service or owner**, for Methods or the relevant legend. AI-created/manipulated clinical images within a confirmed formal research design: method and image-rights details. JAMA's broader study-design/reporting requirements remain outside this AI-disclosure renderer and must be surfaced as a separate scope note. |
+| Nature | tool name; how it was used; affected content; author acceptance of accountability | Venue-only image/media records are halted by `NATURE_VENUE_IMAGE_CONTAINMENT` and never reach Phase 2b. Non-image research elements: method details supported by the current executable row; any evidence/contract mismatch halts under ledger rule 9. |
+| NEJM | AI-assisted technology description; what the technology produced; human review/editing performed; author assertion that AI-produced text/images contain no plagiarism | AI-generated quoted material: attribution and full-citation details |
+| NeurIPS | tool name; version when known; specific tasks; human review of AI-generated content | None |
+| PLOS | tool name; how it was used; how outputs were validated; affected parts of the work; `article_has_methods_section` | If `article_has_methods_section` is true, placement is `METHODS`; if false, placement is `ACKNOWLEDGEMENTS_NO_METHODS`. The placement enum is closed and the false branch is not an unknown location. |
+| Science | tool name; affected manuscript parts; author verification of AI-generated content | None |
+| The Lancet | affected use class; actual tool identity for every USED task; author acceptance of full responsibility | All permitted submitted visuals first satisfy the global Phase-2a.1 accuracy, originality, protected-subject, rights, attribution, and data/result/reference integrity gates. Substantive manuscript-preparation use: tool/service name, purpose/reason, human oversight, and review/editing performed. AI used as part of research: reproducible method details. Explanatory image use: per-image tool, version, and how used. AI-generated data visualization: model/tool name, version, developer/manufacturer, and reproducible Methods details. Research-method image use: name, version, per-record `developer_or_manufacturer_applicable`, developer/manufacturer only when that predicate is true, and reproducible Methods details. If `graphical_abstract_used_ai_or_ai_assisted_illustration` is true and Phase 2a.1 confirms a dedicated scientific or professional illustration tool: tool name for the caption and publication-rights/license basis. Cover-art action facts are collected in Phase 2a.1 and preserved in the audit/action checklist, not rendered as manuscript text. Primary-research-image alteration or general-purpose GenAI graphical-abstract use: prohibited-use halt, not a disclosure field. |
+
+**Ledger rules (fail closed):**
+
+1. Record the Phase-2a applicability result. Phase 2b cannot begin while it is `UNKNOWN`; neither `NOT_REQUIRED` nor `ACTION_ONLY` may render a disclosure paragraph. `ACTION_ONLY` returns the already-confirmed action facts and zero disclosure blocks unless another use makes the aggregate result `REQUIRED`.
+2. Give every unconditional field, every conditional predicate, and every field expanded by a `true` predicate one state: `KNOWN`, `NOT_APPLICABLE`, or `UNKNOWN`.
+3. `NOT_APPLICABLE` is valid only for a conditional child field whose predicate is explicitly recorded as false. An unconditional field cannot be `NOT_APPLICABLE`; an unknown predicate does not make its children not applicable. In particular, record `extension_numbers_applicable` before JAMA extension numbers, `developer_or_manufacturer_applicable` before the conditional Lancet research-method-image identity field, and `article_has_methods_section` before selecting the closed PLOS placement enum.
+4. A session or pipeline value may count as `KNOWN` only when its exact value is available and presented to the author for confirmation. Generic labels such as "an AI tool" do not satisfy tool/model/source fields.
+5. `KNOWN` means epistemically known, not policy-compatible. When the selected Phase-2 row explicitly designates an affirmative fact as required before render — such as human review/editing, author responsibility, figure/data verification, or no-plagiarism confirmation — record both the confirmed value and the required value. `KNOWN(false)` is `INCOMPATIBLE`, not a pass; halt without writing the affirmative sentence. This rule does not promote the Frontiers Phase-5-only action checklist into the render ledger.
+6. If any required field or triggering predicate is `UNKNOWN`, or any required value is `INCOMPATIBLE`, **halt before Phase 3**. Output the incomplete/incompatible ledger and ask only for remediable missing facts. Do not ask the author to falsely change a historical fact, generate a draft with placeholders, or silently drop the field.
+7. Continue only after every required field is `KNOWN` with a compatible value or validly `NOT_APPLICABLE`. Preserve the applicability decision and ledger with the rendered disclosure so an author can audit where every statement came from.
+8. Key tool- and use-specific facts by their `tool × task/run/artifact` record. Do not satisfy one tool's version, prompt sequence, rights basis, review confirmation, generated proportion, per-image disclosure, or other child fact with a value belonging to another record. Where the policy asks for an aggregate (for example, total affected portions), retain links to every contributing record.
+9. After expanding all conditional facts, rerun the Phase-2a.1 compatibility sweep against each use record. A record with an unresolved render/prohibition discriminator, an unmodelled use/artifact class, an `OTHER_CONFIRMED` operation/target whose predicate-specific meaning is unresolved, or a fact combination that conflicts with the selected row halts before Phase 3. Explicitly non-blocking Frontiers Phase-5 action-routing facts are excluded and remain outstanding actions instead. Do not infer a new hard condition from policy prose; report a contract gap when the executable table and evidence row cannot be reconciled.
+10. A `USED` `OTHER / UNCLASSIFIED` category never reaches rendering. Preserve its author-reported description in the ledger and halt until it is mapped to the closed category/operation/target contract. Re-labelling it without author confirmation is prohibited.
+
+The ledger applies to the venue path only. It does not weaken or replace the anchor path's three-state and per-anchor input rules.
+
+### Phase 3: Match categories to the venue's required phrasing
+
+Run Phase 3 and Phase 4 only for a `REQUIRED` result with
+`execution_status=READY`. `ACTION_ONLY` skips prose generation and proceeds
+directly to Phase 5 with its audit-backed action checklist. `NOT_REQUIRED`
+emits the applicability decision and policy basis with zero venue disclosure
+blocks. When `NOT_REQUIRED` rests on a confirmed venue exemption rather than an
+all-`NOT USED` inventory, continue to Phase 5 only for applicable advisories;
+an all-`NOT USED` result terminates after the decision and basis.
+
+The Phase-2a predicates and Phase-2b table are the venue track's executable category-to-field mapping. `venue_disclosure_policies.md` is the evidence layer: it supplies the policy summary, wording elements, prohibited uses, and placement channels. Do not claim that the seven-field policy row itself contains a machine-readable category mapping, and do not infer extra mandatory fields from free prose during rendering. If the evidence row appears to require a field absent from Phase 2b, halt and report a contract gap so the two coordinated surfaces can be updated together. If a category remains UNCERTAIN, a required fact is UNKNOWN, or a required value is INCOMPATIBLE, Phase 3 does not run.
+
+### Phase 4: Generate the disclosure text
+
+Generate a disclosure bundle containing one tailored block per required placement channel, using:
+- The venue's preferred voice (first person vs passive, past tense vs present)
+- The venue's required phrasing elements (many venues require the phrase "The authors take full responsibility for the content" or equivalent)
+- The confirmed tool/service, provider/developer, model, and version values from the ledger, including one identity per tool when more than one tool was used
+- The specific categories marked USED
+
+Never hard-code Claude, ChatGPT, or any other product as the current tool. Never
+emit square-bracket placeholders. When a version/provider field is policy-required,
+an absent value remains `UNKNOWN` and the render halts; when it is genuinely
+optional, omit it rather than inventing it. Nature's venue row does not make a
+model version mandatory, so an unknown version alone does not halt that venue
+render.
+
+For JAMA scientific-study use, label the generated Methods text as the **AI-use
+disclosure portion only**. When the corresponding predicates are true, include
+the confirmed description of the copyright-holder permission/license for
+protected model input and the confirmed publication-rights/permission facts for
+included AI-generated content. The permission/license copy itself is a separate
+submission-package action: report its confirmed evidence pointer in the audit
+bundle, but do not paste the license document into Methods. Put item-specific
+image rights in the relevant legend when that is the confirmed channel, and
+generate a separately labelled block rather than hiding it in the general
+Methods paragraph. Append a scope note directing the author to JAMA's remaining
+current study-design and reporting instructions; do not call the output a
+complete or submission-ready Methods section.
+
+### Phase 5: Placement instructions
+
+Output includes explicit placement instructions matching the venue's policy for every block:
+
+```
+Placement: Methods section (Nature policy, accessed YYYY-MM-DD from
+https://www.nature.com/.../policy-url). Include as the final
+subsection of Methods, before Data Availability.
+```
+
+If the venue requires placement in multiple locations (e.g., Methods + cover letter + Acknowledgements), the mode generates separately labelled, purpose-tailored text for each location rather than one paragraph copied into every channel. A policy phrase such as "at submission" is a timing instruction, not a manuscript location; when the source does not name a section, say that the section is unspecified and direct the author to the submission system/current journal instructions instead of inventing one.
+
+For the ICMJE-member targets in this database (`BMJ`, `JAMA`, `NEJM`, and `The
+Lancet`), append a separate **ICMJE-alongside advisory** whenever at least one AI
+use is confirmed, including a `NOT_REQUIRED/READY` result based on the selected
+venue's exemption. Put it after the selected venue's placement instructions
+when those exist, or after the exemption decision and basis when the venue emits
+zero blocks. The advisory states that ICMJE calls for AI use to be described in
+both the cover letter and the submitted work, with writing assistance in
+Acknowledgments and study/data-collection/analysis/figure use in Methods as
+applicable. This is an additional author-facing reminder, not a merge,
+precedence, or deduplication engine: do not overwrite the selected venue's
+channels, do not silently collapse two requirements, and surface any unresolved
+conflict for author/editor confirmation. Do not emit this advisory for an
+all-`NOT USED` inventory.
+
+For Frontiers, append a clearly labelled **pre-submission action checklist**
+after the disclosure placement instructions. This checklist is a separate
+action carrier, not a disclosure fact ledger or a new schema. Select actions per
+use record from this matrix:
+
+<!-- 619-frontiers-action-matrix:BEGIN -->
+
+| Use record | Factual-accuracy check | Plagiarism-free check | Accuracy-to-data check |
+|---|---|---|---|
+| Created written content | REQUIRED | REQUIRED | — |
+| Edited written content | — | REQUIRED | — |
+| Produced non-data visual content | REQUIRED | REQUIRED | — |
+| Edited non-data visual content | — | REQUIRED | — |
+| Produced figure representing manuscript data | REQUIRED | REQUIRED | REQUIRED |
+| Edited figure representing manuscript data | — | REQUIRED | REQUIRED |
+
+<!-- 619-frontiers-action-matrix:END -->
+
+Here `REQUIRED` means display that action for the author to complete before
+submission: check factual accuracy for applicable GenAI-created content; check
+that GenAI-produced or GenAI-edited written or visual content is plagiarism-free;
+and check accuracy to the underlying manuscript data only for a figure that
+represents those data. A confirmed true action may be labelled `CONFIRMED`.
+An UNKNOWN or false action state is OUTSTANDING and must remain visibly
+outstanding; it never becomes a disclosure `UNKNOWN` halt, an incompatible fact,
+or a falsely affirmative sentence. If whether a figure represents manuscript
+data is unresolved, keep that classification as an outstanding routing action;
+do not select or confirm the accuracy-to-data action until the author resolves
+the routing fact as true, and do not halt or withhold the otherwise complete
+disclosure. AI authorship and editor/reviewer
+upload of manuscript content to external GenAI tools remain separate hard
+prohibitions and are never represented as checklist actions.
+
+For The Lancet / Elsevier, when Phase 2a.1 confirms the permitted AI-assisted
+graphical-abstract path, generate a distinct caption disclosure naming the
+illustration tool and place it in the graphical-abstract image caption. Preserve
+the confirmed publication-rights/license basis and accuracy/originality checks
+in the audit ledger; add attribution to the caption when applicable. Never reuse
+the general manuscript-preparation declaration as a substitute for this caption.
+
+For an Elsevier `ACTION_ONLY` cover-art result, emit zero manuscript disclosure
+blocks and a labelled pre-submission checklist recording the confirmed editor
+permission, publisher permission, whether third-party material is present and
+its permissions when applicable, appropriate content-attribution details (or a
+confirmed `none_applicable` basis), and the global image responsibility facts. Do not
+invent a manuscript section for those actions. When cover art accompanies
+another disclosable use, append the same checklist to the normal placement
+instructions.
+
+For Chinese Nursing Journals Publishing House, also emit a non-blocking
+submission reminder that authors should cooperate with the editorial office in
+submitting and archiving AI-assisted text, figures, or code as supplementary
+material. Do not invent a first-submission attachment slot or make that
+operational cooperation clause a disclosure-rendering field.
+
+---
+
+## Failure cases this mode does NOT cover
+
+- **Venues outside the database**: the mode halts and may accept pasted official policy text only as manual-review/curation evidence. It does not guess or execute uncurated prose.
+- **Policies that have changed since the database snapshot**: the mode records the access date in the placement instructions. Users should verify against the current venue page before submission.
+- **Analysis assistance**: if the AI actually ran computations or generated analysis results (not just writing), the renderer evaluates the target's research-use and data-use conditions, then emits any additional location-tailored block the policy requires. It does not assume a Code Availability or Analysis location when the source names a different section or no section.
+- **Co-authored AI**: for every cached row with an explicit AI-authorship rule, the manuscript-level gate refuses to produce author-list text and halts until a proposed AI author is removed. The current PLOS evidence row does not state such a rule, so the mode reports that policy-scope limit instead of inventing one from the other venues.
+
+---
+
+## Integration with existing journal_submission_guide.md
+
+`journal_submission_guide.md` retains two generic examples (Minimal / Detailed)
+for human adaptation outside this mode. They are **not** an unknown-venue
+fallback: standalone disclosure mode remains halted when lookup fails, even if
+the user pastes policy prose for manual review. For a known venue, the
+protocol-driven disclosure bundle supersedes those examples.
+
+---
+
+## References
+
+- `venue_disclosure_policies.md` — policy database (v2: the 6 ML/NLP venues plus 9 medical-publishing policy targets incl. the ICMJE umbrella and the database's first Chinese-language entries; see its Scope line)
+- `policy_anchor_table.md` — #108 4-anchor × 16-field matrix (PRISMA-trAIce, ICMJE, Nature, IEEE) for the policy-anchor track
+- `policy_anchor_disclosure_protocol.md` — #108 policy-anchor track render protocol (per-anchor flows, G10 7-row precedence table, auto-promotion forbiddance, §4.4 11 concerns resolved paths)
+- `journal_submission_guide.md` — existing generic examples for manual adaptation outside disclosure mode (not a runtime fallback)
+- `credit_authorship_guide.md` — existing CRediT authorship best practices
+- Lu et al. (2026). Towards end-to-end automation of AI research. *Nature* 651, 914-919 — the ethics statement for Lu 2026 was drafted in compliance with Nature's policy; their methodology is a worked example of what this mode should produce.
+- `docs/design/2026-05-14-ai-disclosure-schema-decision.md` — #108 Decision Doc (G1-G10 + §4.3 invariants + §4.4 11 open concerns)
+- `docs/design/2026-05-14-ai-disclosure-impl-spec.md` — #108 implementation spec (resolved-paths table)
+- ROADMAP_v3.2.md item 6 — design decisions (v1 venue set, unknown-venue halt, education/QA venues deferred to v2)

package/vendor/ars/academic-paper/references/domain_evidence_profiles.md

@@ -0,0 +1,38 @@
+# Domain Evidence Profiles
+
+> **Advisory only.** A domain evidence profile changes which evidence types the literature screening *admits*. It never changes the A-F Overall Grade and never blocks manuscript ship. It never changes which databases are queried (that stays `Discipline`-driven in `literature_strategist_agent` Step 2). Consumed only by `academic-paper`'s `literature_strategist_agent`; produced by `intake_agent` Step 12 as the PCR `Domain Evidence Profile` row.
+
+## Domain Evidence Profiles
+
+The ship-ready enum is exactly four values. `unknown_user_defined` is the default and the neutral fallback.
+
+| Profile | Standard evidence types | Common provenance requirements | Critical gaps to surface | Reserved-note |
+|---|---|---|---|---|
+| `general_social_science` | Peer-reviewed empirical studies, mixed-methods, expert-panel / context-dependent policy analyses | Journal or proceedings; expert-panel reports acceptable where context-dependent | Single-context generalization; weak external validity | — (ship-ready) |
+| `cs_ml` | Peer-reviewed papers AND archival preprints (e.g. arXiv), conference proceedings; industry technical reports | Preprint server or proceedings acceptable; peer-review lags the field | Non-reproducible results; benchmark cherry-picking | — (ship-ready) |
+| `humanities_interpretive` | Primary sources, archival material, canonical/older texts, monographs | Primary-source provenance; recency is not a quality signal | Interpretive over-reach; missing primary-source grounding | — (ship-ready) |
+| `unknown_user_defined` | Neutral single-pyramid (pre-#259 behavior) — peer-reviewed default | Standard peer-review expectation | (no profile-specific loosening) | Default / neutral fallback |
+
+**Reserved profiles (documented, NOT in the enum).** Selecting one records effective `unknown_user_defined` and surfaces a reserved-fallback advisory — its checklist does not exist yet, so it falls back to neutral to prevent false rigor:
+
+`clinical` · `wet_lab` · `materials_physics` · `legal_case_based` · `education` — *not in enum yet; selecting this falls back to `unknown_user_defined`.*
+
+## Field-guidance carry-forward (seeded from the deep-research evidence hierarchy)
+
+The substance of the field-centric `## Field-Specific Adjustments` table in `deep-research/references/source_quality_hierarchy.md` is carried forward here so no per-field guidance is silently dropped. The deep-research file is **read, not edited** — this is a one-time authoring copy, not a runtime dual-read.
+
+Normative (folded into a ship-ready profile row above):
+- **Social Science** → folded into `general_social_science` (Level III-V; mixed methods common).
+- **Technology** → folded into `cs_ml` (Level III + industry reports; peer review lags reality).
+- **Humanities** → folded into `humanities_interpretive` (Level VI primary sources; different epistemology — "evidence" means different things).
+- **Policy** → folded into `general_social_science` (Level IV-V + VII expert panels; context-dependent, expert opinion valued). There is no dedicated Policy profile.
+
+**Historical reference — non-normative; current behavior for these domains is neutral `unknown_user_defined` until the `clinical` / `education` profile ships:**
+- **Medicine/Health** — Level I-II (RCTs, meta-analyses); Level I-III common; evidence-based-medicine tradition. *(maps to reserved `clinical`)*
+- **Education** — Level III-IV (quasi-experimental); Level IV-VI common; randomization often impractical. *(maps to reserved `education`)*
+
+These two rows are preserved verbatim for the eventual `clinical` / `education` profiles. They do NOT change runtime behavior today: until those reserved profiles ship, Medicine/Health and Education runs use the neutral `unknown_user_defined` pyramid, exactly as every other unmapped selection does.
+
+## #246 forward reference
+
+Discipline-relative *grade aggregation* (how these evidence expectations roll up into an Overall Grade) is tracked separately in #246 and is **not yet implemented**. Until #246 ships, the A-F Overall Grade lookup in `deep-research/references/source_quality_hierarchy.md` applies unchanged. #259 ships no aggregation logic and no placeholder aggregation code.