De AI

Normative authority: docs/SCIPAPER_STANDARD.md. Where this file and the standard disagree, the standard wins. docs/architecture/DEAI_SUBSYSTEM.md explains the design; docs/architecture/EVALUATION.md records the empirical evidence. Style profiles and learned models supply evidence and positive anchors, never competing policy.

Provenance (MIT, attribution retained): this skill chains three de-AI sources into one pipeline — the in-house de-AI subsystem (the L0-L4 tools and the standard's feedback contract) and two vendored humanizers: AIScientists-Dev/academic-humanizer v0.3.3 (MIT; the Layer 1-5 catalog) and blader/humanizer (MIT; the structural families 2.12-2.16, the Pass-2 self-interrogation, and the false-positive guards). The upstream word lists are not normative here — corpus evidence is (§7 Corpus overrides).

de-ai — one skill, three passes

0. What this skill does

Three sequential passes over one target, all inside the standard's measure -> type -> rank -> edit -> re-measure -> disposition loop:

1. Pass 1 — measure (de-AI subsystem). The unified linter and the document-scale detectors produce a machine-readable sci-paper.feedback.v1 report: L0 lexical/punctuation, L1 distribution, L2 sentence + document structure, L3 learned similarity, L4 positive voice.
2. Pass 2 — humanizer audit (two vendored repos). The structural-tell catalog (Layers 1-5 plus families 2.12-2.16) runs in audit form; its findings merge into Pass 1's ranked list.
3. Pass 3 — claim-first rewrite. Each selected paragraph is rebuilt from its protected claim graph; only fidelity-eligible candidates within the §5.3 length budget survive; re-measure, then apply with author-visible before/after.

--audit-only stops after Pass 2 (findings only; this is the mode paper-review dimension D and final-review invoke). --no-apply runs Pass 3 to proposals only. For technical prose, neutral and precise IS the human voice — this skill strips machine tells; it never adds personality, humor, or first-person flavor.

Boundary with /sci-paper:condense (canonical home — this statement; mirrored verbatim in condense SKILL.md §0): de-ai removes the authorship fingerprint (L0-L4 signals, structural tells, voice; its Pass-3 length cap only guards its own rewrites against growth). condense removes redundancy and length (cross-document deduplication under one-canonical-home-per-fact). At the overlap (verbose AI-isms such as rule-of-three padding or connective stacking), de-ai DETECTS the tell; when the right fix is deletion rather than rewrite, the deletion EXECUTES under condense's ranked sweep.

0.1 Non-negotiable rules

1. Scientific fidelity is eligibility, not a score. A candidate that drops or alters a number, unit, citation, mathematical expression, acronym, comparison direction, negation, causal direction, named entity, scope, stance, or qualifier is ineligible regardless of its style score.
2. Never add specificity from imagination. Every number, entity, citation, and quantitative qualifier must already exist in a source read in the same turn. Specificity means retaining supported detail, not inventing it.
3. Rebuild from claims, not surface swaps. Synonym substitution, connector deletion, and punctuation cleanup are not a structural rewrite. Extract the claim graph and compose new sentences.
4. Imitate style, never content. Exemplars guide rhythm, register, information distribution, and transition practice; claims, citations, data, and distinctive wording are never copied from them.
5. Learned scores are advisory field-similarity signals. They do not establish authorship and are not paper gates. Missing calibration is degraded or unmeasured, never a clean result. No universal style verdict exists.
6. L0 is the only axis rewritten to zero. Tier A, em-dash, and Tier B above the section cap are l0_target. L1-L4 findings are ranked advisories with dispositions, never forced to zero at the cost of clarity.
7. Condense, do not accumulate (§5.3). The default direction of every rewrite is shorter; the explanatory patch — answering a finding by appending a clause instead of rewriting the flagged text — is the canonical violation. Forward narrative: never narrate the drafting history.
8. Minimum effective edit + explicit human control. Rewrite only the paragraphs whose findings or the author's instruction justify it; show before/after, fidelity evidence, and residual dispositions.

1. Field resolution and calibration

1. Resolve --field: one available field under style-profile/ selects automatically; multiple require --field; zero means corpus guidance is unavailable — model-free axes still run and the gap is reported.
2. Read docs/SCIPAPER_STANDARD.md and the /sci-paper:paper writing guidance (Anti-AI-isms tiers, forward narrative, formula and citation standards, canonical L0 examples).
3. Read style-profile/<field>/style_dossier.md in full when it exists, and check freshness: if the corpus under style-corpus/<field>/ is newer than the dossier, stop and regenerate with python tools/extract_style.py. A stale profile is unmeasured, never treated as conformity.
4. Retrieve section-typed positive anchors only when the bank exists: python tools/retrieve_exemplars.py --field <field> --section <type> --topic "<verified one-sentence topic>" --k 5. Read every returned exemplar; record provenance; use for rhythm and register only.

2. Pass 1 — measure (de-AI subsystem)

python tools/ai_ism_lint.py <file> --field <field> \
  --structure --distribution --document-structure --register --salience \
  --oracle --voice --summary --format json --output <scratch>/feedback-before.json

Read the output as sci-paper.feedback.v1; record every axis state (measured / degraded / unmeasured / not_applicable) — silence is never clean. Exit semantics: 0 = no L0 targets (advisories may remain), 1 = L0 targets present, 2 = input/config/execution failure. Never re-parse JSON from printed prose.

Axes and tools:

• L0 lexical + punctuation (ai_ism_lint Tier A/B + em-dash) — the only to-zero axis. The Tier A/B word lists are the TIER_A_PATTERN and TIER_B_PATTERN regexes in tools/ai_ism_lint.py, mirrored for human reading in skills/paper/SKILL.md; do not re-derive them by hand. style-profile/<field>/lexicon.json is separate calibration data: only its llm_words_absent_from_corpus key is read, and only for the advisory corpus-zero: rule.
• L0 register (deai_register) — terms the manuscript leans on that the field's own corpus does not carry: the tell of a paper written in a neighbouring discipline's vocabulary. Evidence is corpus document frequency, never a curated list of another field's words, because a list cannot separate AUC (1 astronomy passage in 15,599) from epoch (402) and accuracy (774). Advisories only; a corpus-rare term may be a borrowed method's accepted name or the concept the paper introduces. The same tool's zero-hit audit (register-zero:<term>) is exhaustive rather than thresholded: every body word the manuscript uses that no corpus passage carries is listed, strong unless the word is a mechanical formation of an attested stem (clamped from clamp). A strong hit is answered by one of the §4.2 dispositions, never by silence.
• L2 collocation (deai_collocation) — sentences that join common words the field never joins (physical cells, controlled grid): the fraction of a sentence's adjacent content-word pairs that no corpus passage attests, against a per-section leave-one-out reference. A pair the field writes by chance would have been seen; a pair it never wrote is a coinage or a figure of speech, and the action names the relation it compresses. Advisory; a term this paper defines keeps its pair.
• L2 salience hierarchy (deai_salience) — whether a passage ranks the quantities it reports or recites them. Not numeric density (a quantitative abstract is supposed to carry numbers) but how far the numerals run without an interpreting sentence between them, against a per-section human reference. This is the only axis that reads the numeral-preserving LaTeX projection; the others read latex_to_plain, which replaces math with [math] and therefore measures zero numerals on any .tex file.
• L1 distribution (deai_metrics, deai_oracle) — burstiness, connective openers, surprisal/UID. Degraded without a field operating point; rank-only.
• L2 sentence structure (deai_structure) — template families (announced enumeration, ordinal runs, setup-list-wrap, repeated modal/anaphoric frames, symmetric closers) plus the auxiliary families (antithesis clusters, short reversal beats, paper-as-agent subjects, wh-cleft openers, modifier stacks of three-plus tokens with two hyphenated compounds). Auxiliary families never enter the template score; they name the sentence.
• L4 residue (deai_residue) — the trace an edit leaves: self-referential drafting history (we initially, no longer), edit-meta text (TODO, see previous version), a sentence that defines the paper's own object by what it never does or has (residue-absence: never strong, carries no ordinary, exempt with a citation in the sentence; delete the clause where its neighbours already carry the positive statement, rewrite only what the page lacks, and record the dropped negation the §6 gate reports), and a heading or caption that promises what the body never delivers (negative-label). With --before or --git-ref the diff rule reports a label the edit added and the body does not earn, and exit 1 gates the pass; a residue finding is fixed by rewriting the sentence to the current state of the science, never by adding an explanation.
• L2 document structure (deai_docstructure) — the de-AI centre of gravity: per-stratum cross-paragraph dispersion manifold, role coupling, split-conformal operating points. Word-level rewriting cannot move this signal; the only sanctioned lever is deai_partition (§4).
• L2 claim anchoring (deai_anchoring) — a writing-quality band, not an AI-discrimination axis.
• L3 learned voice (deai_voice) — a decided-degraded offline audit instrument; per-paragraph findings are confidence-capped at 0.5. A low score never proves machine authorship.
• L4 cooperative repair (deai_partition, deai_provenance, deai_personal) — provenance/personal are honestly unmeasured until the author supplies their own draft history or prior papers.

3. Pass 2 — humanizer structural-tell audit (vendored)

Audit against the catalog below, then merge every hit into the Pass-1 ranked list (Layer 1-2 structural hits are style-class advisories, density excess upgrades to strong; Layer-4 claim-evidence failures are integrity-class). Skip LaTeX comments and scaffolding macros. This pass never edits.

Layer 1 — general AI tells

Inflated significance; superficial "-ing" tails that fake depth; promotional or figurative language; vague attribution without a citation; copula avoidance ("serves as" -> "is"); negative parallelism ("not just X, but Y"); rule-of-three padding; elegant variation (one referent keeps one name for the whole paper); filler ("it is worth noting that", "in order to"); overlong clause-stacked sentences (2.11); em-dashes (remove entirely; recast with commas, colons per the colon rule, parentheses, or separate sentences). Lexical tells are enforced mechanically by the linter — run it, do not re-derive the lists here.

Layer 2 — academic AI tells

• 2.1 Over-claiming verbs. prove / establish / confirm / guarantee / demonstrate, checked against the evidence (§7 corpus overrides).
• 2.2 Significance hype. paves the way for, a crucial/pivotal step toward, opens new avenues, sheds light on, bridges the gap -> the specific failure mode or result addressed.
• 2.3 Empty intensifiers. extensive/comprehensive/thorough experiments, a wide range of, numerous, various -> enumerate or quantify.
• 2.4 Novelty padding. "novel" more than once per section; "to the best of our knowledge"; "for the first time" -> state the specific gap.
• 2.5 Formulaic openers. "In recent years...", "With the rapid development of...", "Despite recent advances,..." -> open with the structural fact or the problem.
• 2.6 Connective overuse. No consecutive sentences opening with Moreover/Furthermore/Additionally/In particular; let logic carry.
• 2.7 Contribution-list cliches. Each contribution names a specific result with its number, not a restatement of the abstract.
• 2.8 Citation dumping. Evidence-conditional, never a length rule: flag a bracketed list only when entries do not support the sentence or duplicate a role. Never delete a relevant source merely to shorten a list.
• 2.9 Hedging-by-vagueness. somewhat, relatively, fairly, to some extent, quite -> quantify or cut (distinct from calibrated hedging — Layer 3).
• 2.10 Boilerplate emphasis. "It is worth noting that", sentence-initial "Notably,"/"Importantly," -> if it matters, the sentence shows it.
• 2.11 Overlong clause-stacked sentences. Past ~30 words, 3+ subordinate clauses chained by which/that/while/with, or double-nested parentheticals. Split: one idea per sentence. The highest-yield structural check — the mechanical linter does not measure it.

The next five (2.12-2.16, from blader/humanizer) are structural, not lexical — flag the construction, honour each corpus caveat, never the bare word.

• 2.12 False ranges. "from X to Y" with categorical endpoints ("from theory to observation"). Caveat: a genuine quantitative range (redshift 0.1 to 0.5) is correct and required; flag only the rhetorical range whose endpoints share no scale.
• 2.13 Aphorism formulas. "X is the Y of Z" / "X becomes a Y" epigrams for resonance -> state the literal fact. Caveat: a formal definition of the same shape ("the aperture mass is the convolution of the shear with the filter") is a definition; keep it.
• 2.14 Persuasive authority tropes. "at its core", "fundamentally", "the real question is", "it is important to realize" -> cut, or give the specific reason. Caveat: "fundamentally different" naming a real physical distinction (regime, symmetry, scaling) is substantive; keep it.
• 2.15 Manufactured staccato drama. A run of terse fragments strung for impact -> measured sentences; academic emphasis comes from content. A single emphatic short sentence is fine (Layer 3); only a run is a tell.
• 2.16 Hyphenated-pair predicate overuse. Compound modifiers piled in the predicate ("the result is model-dependent") -> prefer the plain relation ("the result depends on the model") where it reads naturally. Fine as an attributive modifier; flag density, not any single use.

Pass-2 self-interrogation (blader/humanizer two-pass audit)

After the first draft of any Pass-3 rewrite, ask of it: "what still reads as machine-written here?" Answer in 2-4 concrete bullets (a specific phrase, a rhythm, a residual tell), then apply one further targeted rewrite that clears them and re-checks the fidelity and length gates. Do not manufacture voice to pass this step.

Layer 3 — preserve these (do NOT over-correct)

Evidence-tied hedging ("suggests", "is consistent with", "we hypothesize", "may indicate", "appears to") is correct and required when the claim is genuinely uncertain; strengthening a calibrated verb is a claim-evidence defect. Passive voice is fine when the actor is irrelevant; first-person plural "we" is standard; semicolons and an occasional triple are fine; formal definitions, named methods, symbols, and equations stay verbatim. False-positive guards: do not flag formal vocabulary on its own; a single mixed register; straight-vs-curly quotes in isolation; a lone emphatic short sentence; text inside quotations, a caption's verbatim labels, or a worked example; or field-common knowledge. Specific hard-to-fabricate detail, a genuine mid-sentence self-correction, and calibrated hedging are human signals, not tells.

Layer 4 — claim-evidence discipline

For every empirical claim: (a) is it backed by a number, figure, table, or citation in the text, and (b) does the verb match that evidence's strength? Unbacked -> add the pointer or soften. Verb too strong -> downgrade to what the data show. Vague magnitude -> a number or an attributed range; lead comparisons with the strongest baseline. Layer-4 failures are integrity-class, not style-class.

Layer 5 — voice and venue

If the author supplies prior papers, sample them first and match sentence rhythm, connective habits, hedging placement, and section openers. Otherwise default to clean, precise, venue-appropriate prose; the corpus dossier (§1) is the measured voice baseline. Funding proposals route to /sci-paper:proposal-polish, never here.

4. Pass 3 — claim-first rewrite

Select work in ranked order: L0 targets in the requested span first, then strong L1/L2/L3 advisories with concrete actions, then document-shape findings. For each selected paragraph:

1. Protect. Write the protected claim record (<scratch>/claim.txt): every claim and its evidence relation, all numbers/units, citations, named entities, inline math and acronyms, comparison/negation/causal direction, scope, stance, qualifiers. Re-read every numeric and citation source in the same turn. Snapshot the original verbatim (<scratch>/original.txt) as the §5.3 length baseline. 1b. Bind. Fill the binding ledger (§4.3) for the paragraph. It is not optional and no detector emits it; every none row is a clause the rewrite must drop rather than carry forward.
2. Generate. Produce N candidates (default 5) from the claim record, not by editing sentence-by-sentence. Each preserves every protected item, states the claim at the same strength and scope, adds no new fact, avoids Tier A and em-dashes, keeps Tier B within cap, breaks the specific L2 pattern that motivated the rewrite, keeps technically necessary lists, is no longer than the original unless a protected invariant forces it, and carries no clause that the ledger marked none.
3. Gate + rank.

   python tools/rewrite_reward.py --field <field> \
     --reference <scratch>/claim.txt --original <scratch>/original.txt \
     --candidates <scratch>/cand_0.txt <scratch>/cand_1.txt ...
   

   eligible=False (fidelity failure, or longer than the original → -inf) cannot be selected; the tool exits 1 when NO candidate is eligible (a measured outcome, not a failure — regenerate tighter) and 2 only on invalid input or a missing profile; --allow-growth "<reason>" only with an author-approved justification, printed into the run record. Among eligible candidates rank by L0 reduction, semantic fidelity, voice score, and condensation. If none is eligible, preserve the original and regenerate tighter — never pick the least-bad ineligible candidate. Then verify entities, scope, stance, and logical dependencies against the source by hand: the deterministic checker is necessary, not sufficient.
4. Re-measure. Re-run the Pass-1 linter on the candidate in enough section context for section caps. A good rewrite introduces no new integrity blocker or L0 target, stays eligible, acts on the selected advisory, worsens no higher-priority finding, does not grow the passage without a stated reason (report the length delta), and reports residual advisories.
5. Apply. Show before/after, protected invariants, eligibility, affected findings, and dispositions; apply with a minimal Edit and re-read the changed region in context. Run the Pass-2 self-interrogation on the applied text.

4.1 Ranking a recital passage (salience-recital)

A recital finding is not an instruction to delete numbers. It says the passage has not told the reader which of its quantities matter, and the repair is hierarchy, not subtraction. Work in this order, inside the same claim-record protocol as any other rewrite:

1. Name the passage's one claim. Write it as a sentence with no number in it. If the passage supports two claims it is two passages, and the partition suggestions of §4 apply instead.
2. Ask each quantity what it establishes. A number that is the evidence for the claim stays, and the sentence carrying it says what it establishes. A number that is merely something the work produced belongs to the section that argues from it. Moving a value is a fidelity-neutral operation only if it lands somewhere; dropping it is not, and a number that is the sole support of a claim never leaves.
3. Break the run. The measured defect is consecutive numeral-bearing sentences. One interpreting sentence between two reported quantities removes the finding and is the sentence a reader wanted; adding a filler clause to the same sentence is the §5.3 explanatory patch and is not the fix.
4. For an abstract, check three sentences exist. What was not known, what this work establishes, and what follows for someone else's analysis. A quantity list can satisfy none of these while every number in it is correct. State the contrast the result creates against what was assumed before, when the manuscript already supports that contrast; never manufacture novelty language to supply it.

An accepted disposition is the right outcome wherever the density is what the genre requires: a methods paragraph specifying a parameter grid is an enumeration by nature, and the finding only reports that it is denser than most human method passages.

4.2 Register dispositions (register-foreign, register-zero)

Three dispositions, in preference order, each recorded. A strong register-zero hit must take one of them explicitly: it is a word no passage of the field has ever written, and the only reasons to keep it are that this paper defines it (disposition 3, with the definition at first use) or that it names a method the field has not used before (disposition 2, with the citation that introduces it).

1. Name the field's own quantity. Prefer the term the audience reads in, when one exists and means the same thing. A weak-lensing reader asks about completeness and purity at a given false-alarm rate; a discrimination summary borrowed from another field may be exactly the same number under a name they do not use. Only make this swap when the replacement is the same quantity — renaming a statistic to something it is not is an integrity defect, not a register fix.
2. Keep the term and define it. Where the borrowed name is the accepted one, define it at first use and say what it measures in field terms.
3. Confirm the definition, for a term this paper introduces. A corpus frequency of zero is equally consistent with a foreign term and a new one; the detector cannot tell them apart and does not try.

Never swap a term whose replacement would change the claim, and never introduce a field term the manuscript has not earned.

Document-shape findings (dispersion manifold, role coupling) cannot be moved by word-level rewriting. Run python tools/deai_partition.py <file> --field <field> and surface its fidelity-free merge/split suggestions to the author verbatim; they change zero tokens and are applied by hand, never automatically.

4.3 The binding ledger (SCIPAPER_STANDARD §5.4)

No detector produces this finding. §4.1 and §4.2 act on measured advisories; this one is a mandatory unmeasured pass, run on every paragraph selected for rewrite. The reason it has no tool is recorded: three surface statistics were built for it and all three were refuted (EVALUATION §9.6, §14.5, §15.1). The defect is not a property of a clause but of the relation between a clause and the propositions around it, so the ledger is filled by judgement and made checkable by its output shape, not by a threshold.

Fill it before generating candidates, on the original paragraph:

• fact — introduces something checkable that was not in play: a number, a measurement, a named object, a concrete procedure.
• link — binds two propositions already in play. Name both.
• none — no antecedent can be named. This is a claim about the text, so write it only after trying to find one.

A causal, purposive, or evaluative frame with a none antecedent is the padding case: the sentence has the shape of an argument and none of the substance. Delete it in the rewrite; do not repair it by inventing the antecedent, which manufactures a claim the sources do not support.

Two traps make this pass fail in opposite directions:

• A connective is not an inference. Machine prose supplies "therefore", "thereby", "through", and "which in turn" and leaves the antecedent unbound; one AI bank carries inference markers two to three times more often than human abstracts (EVALUATION §15.1). Judge the antecedent, never the marker.
• A hedge is not padding. A clause that narrows a claim — a condition, a range, an uncertainty, a scope limit, a negation, a conceded limitation — is load-bearing by definition and is fact, not none, however evaluative it reads. §6 eligibility already protects the qualifier class and outranks this pass entirely.

Where a whole paragraph resolves to none rows, the repair is deletion and belongs to /sci-paper:condense, not to a rewrite; hand it over rather than rewriting inert text into better inert text. Where the paragraph's fact rows are unranked instead of unbound, that is the recital case — use §4.1.

Optional self-distillation (--distill)

Only paragraphs explicitly accepted by the author may enter style-profile/<field>/exemplar_paragraphs.jsonl as positive voice anchors, with provenance and section type recorded. Never feed automatically generated or merely detector-favored text back as human positive data. Retraining on the grown bank requires held-out and confound-aware evaluation; a multi-minute heavy run executes on the authorized compute environment, not locally.

5. Stopping rule and report

Stop only when: all integrity blockers are resolved or verified false positives; applicable L0 targets are zero; every strong advisory is acted, accepted, rejected_as_false_positive, or pending with a stated reason; ordinary advisories and unavailable axes are reported; and every changed paragraph passes scientific-fidelity verification. This is a disposition-complete feedback state, not a universal prose verdict. If the iteration budget is exhausted, leave the original text for unresolved cases and return the pending findings.

# de-ai — Feedback Report
Target: <file> | Field: <field> | Candidates/paragraph: N
Measurement: <axis=status list, all four states shown>
Passes run: measure / humanizer-audit / rewrite (or audit-only)
Paragraphs considered: K | rewritten: R | unchanged: U
L0 targets: before B -> after A   (Tier A, em-dash, Tier B excess)
Strong advisories: acted X | accepted Y | false-positive Z | pending P
Length delta (rendered prose): <per-paragraph words, §5.3>

## Paragraph changes
- <line/section>: <finding ids and rules>
  - eligibility, missing invariants, before/after, disposition

## Residual feedback
<ranked ordinary advisories, document-shape suggestions, unavailable axes>
No number, equation, or citation changed.

6. Anti-patterns

• Treating synonym swaps or connector deletion as a structural rewrite.
• Quoting a manuscript number or citation from the dossier or memory.
• Forcing every advisory to zero, or every sentence-length outlier to the mean.
• Manufacturing raggedness, voice, or personality to lower a detector score.
• Calling a compatibility threshold calibrated, or a stale profile fresh.
• Declaring a paper human- or machine-authored from a learned score.
• Answering a finding by appending an explanatory clause (the §5.3 patch).
• Deleting a relevant citation merely to shorten a list.
• Copying or lightly paraphrasing an exemplar.
• Doing condense's job: bulk redundancy deletion belongs to /sci-paper:condense; de-ai only rewrites what a tell finding selects.

7. Corpus overrides and interfaces

Corpus overrides (measured evidence beats the upstream word lists): landscape is a legitimate astrophysics term — never flag it lexically; demonstrate* and significantly are normal astro usage — flag only on a Layer-4 failure, never as bare words. The enforced Tier A/B lists live in tools/ai_ism_lint.py; this skill defers to them. The corpus-derived style-profile/<field>/lexicon.json feeds the advisory corpus-zero: rule only. When the corpus changes, re-run tools/extract_style.py and tools/build_profile.py.

Interfaces:

• docs/SCIPAPER_STANDARD.md — consequence classes, ranking, disposition, stopping, §5.2 de-AI-ization order, §5.3 length budget, §6 rewrite eligibility.
• /sci-paper:paper — writing guidance and canonical L0 prose examples.
• /sci-paper:condense — the redundancy/length action surface (boundary statement in §0).
• /sci-paper:paper-review — invokes this skill --audit-only as its dimension-D structural-tell step.
• /sci-paper:final-review — runs this skill --audit-only as an isolated reviewer.
• Tools: ai_ism_lint (L0 hub + aggregation), deai_metrics, deai_oracle, deai_structure, deai_salience, deai_register, deai_collocation, deai_residue, deai_docstructure, deai_anchoring, deai_voice, deai_partition, deai_provenance, deai_personal, rewrite_reward (fidelity + length gate), retrieve_exemplars, extract_style, build_profile, deai_feedback (schema + ranking).