Weksa/docs/targets/interlingua-lowering-map.md

Map: interlingua lowering (campaign weksa-lowering)

Prose map for the campaign whose target is interlingua-lowering.md. Rulings, questions, follow-ups and cut specs live in the Eureka mind under weksa-lowering:; this page is the model page and the cut sequence, and cites them by label. Written by Imagination for the Eureka-area Self, 2026-10-29.

The design in one paragraph

The interlingua spec has two layers. The schema is Rust types: packet families, node ids, role slots, value kinds. It changes rarely and by commit. The inventory is data in CultCache: the concepts, frames, roles, relation types, speech acts and nuance features (hedge, certainty, irony, stance affects and restraint, and whatever the loop finds missing) that the schema's slots name by id. Every open slot in a packet is an inventory id; a referent names an entity by id from the entity store, and a packet's only free text is the trace. No door judges text for leakage: what each model call may see, and what a score may count, own that (weksa-lowering:ruling:leak-owned-by-flow-and-scorer-3). The golf loop improves the spec mostly by growing and refining the inventory, which is fast; schema changes are the slow path. Concepts are open-class and grow forever; the other families are closed-class, and convergence is measured on them (weksa-lowering:ruling:convergence-test). This split was spec-is-schema-plus-inventory, now carried by leak-owned-by-flow-and-scorer-3.

Model page

Every persistent kind, what names it, its life, and who decides. "Derived" means no store holds it and nothing writes it.

Kind	What names it	Lifecycle	Who decides
Packet	Key packet:<16 hex>, an opaque digest of item, inventory revision and encoder (the item key lives in provenance.corpus_item, which the lowerer never sees); CultCache type weksa.interlingua_packet, carrying schema_version and inventory_rev. Every id inside is an opaque form type (n<k> nodes, encoder:<16 hex>, canonical <family>:<n>). From cut speaker-state (schema v4) it also carries the utterance stance: the speaker's felt affect and restraint, as inventory features of slot Stance (see Speaker state below)	Minted by an encoder for one corpus item under one spec version; immutable; a new spec version means a new packet, never an edit. Admitted only through PacketStore::admit, which validates against the inventory store at the packet's revision	The encoder writes; the form types refuse malformed ids at construction and load, the validator refuses broken references and over-capacity packets: a feature at most once per slot, at most 64 nodes (typed-packet); nothing else writes packets
Schema version	weksa.interlingua_packet.v<N> constant in crates/weksa; published JSON Schema at schemas/cultnet/	Bumped by a commit that changes the Rust types; the published schema is regenerated in the same commit and a test fails on drift	A cut (Hands, under a spec); never the loop
Inventory entry	Key <family>:<n>, n canonical decimal >= 1 (families: concept, frame, role, relation, speech_act, feature, referent_kind, extension); type weksa.inventory_entry, with a bounded gloss, value kind (option labels bounded), the slots a feature may sit in (Discourse, Referent, Predication, Relation; Stance from cut speaker-state, Speaker and Scene from cut speaker-scene), introduced_rev and optional retired_rev. Speaker axes, scene relations and affects are features of those slots, so the loop grows them like any closed-class family and convergence covers them	Added or retired by an inventory revision; never edited in place (a changed meaning is a retire plus an add); a retired id still resolves for packets of earlier revisions	The loop's accepted rounds, whose minting call the context builder assembles from Train items only; weksa inventory admit for one model-drafted EditDraft outside a run (cap 256, reviewed whole in its cut; same numbering and admission check as the loop); Self on merge to main. data/original revs 5 (233 concepts and frames) and 6 (five concepts closing named gaps) were drafted by claude-sonnet-5-5 (Weksa Hands, model-drafted original-bed lexicon from Train and Validation passages); Validation-informed: numbers on this bed measure lowering, never generalization or convergence. Residue after rev 6, not amended (each fresh coverage reader names new gaps; the loop's growth path owns them): simultaneity (ながら), distal deixis (あの), 'first' (まず), counterfactual mood, 'at least', emphatic question; heaven versus sky for the visible 天 stays weak. Admission checks form, not text
Inventory revision	Key rev:<n>, type weksa.inventory_revision, with parent, change list and frozen	Appended per accepted round; exactly one revision is ever marked frozen, by the convergence test	The loop's accept rule appends; the convergence test freezes; Self records the freeze as a ruling
Target profile	Locale or language-project id (en-US, ja-JP, a conlang's slug). Until entries grow in rounds there is no profile revision document: a run names the profile it lowered with by the ontology store's path and digest in its RunRecord (cut language-ontology); profile:<language>:<n> revision documents arrive with follow-up language-ontology-growth	The ontology store is revised by commit; a run pins it and refuses drift; it likewise pins the speaker-state store (RunRecord keys 17 and 18, Drift(SpeakerState), cut language-ontology-scene)	Profile authors and accepted rounds; never the lowerer's defaults (persona-drives-register)
Language ontology entry (cut language-ontology)	Key onto:<language>:<n>, n from 1 per language; type weksa.ontology_entry in data/ontology/ontology.cc: the language, the revision of the inventory it was admitted through (provenance: beds number revisions independently and a run reads each condition's feature at its own head), a layer (Lexicon, Grammar, Discourse, Phonology, Orthography, as the door spells them; the mint prompt ends with an example line serialized from the draft type itself), at most four conditions (a feature id of slot Stance, Discourse or Scene with AtLeast/AtMost/AtMostOrUnset a grade or Is an option or flag; a Scene condition reads the relation of the item's speaker to its addressee that with_scene selected; slot Speaker and community ids wait on follow-up language-ontology-communities), a choice of 1 to 200 bytes, at most three examples of 1 to 120 bytes, and an Annotator. Whether an entry's text keeps politeness on the scene is a model reader's judgment, owed per change (a mint owes it under language-ontology-growth), not a test	Admitted in all-or-nothing batches through OntologyStore::admit, immutable, never edited; retire arrives with language-ontology-growth. The run pins the store like the entity store (path and digest, Drift(Ontology)) and never writes it; slice 1 of growth is outside any run: weksa loop mint-ontology drafts entries from the run's Train items of one language into <root>/ontology-mint/<digest>.answer, weksa ontology admit --ontology <copy> admits them into a file copy, and a new run pins the copy (a run-owned copy waits on follow-up language-ontology-growth). Cut language-ontology-seed converts by drafting, not by parse: a fresh model drafts each language's seed from that language's YAML and the inventory's feature lines with no corpus text, then deletes data/target-language-ontology/ and .weksa/cultural-ontology/ in the same cut	Seed: a labelled model (Annotator::Model), no operator gate, never Test evidence (ja-ontology-owner-answer); attestation (operator, native reader) appends later. Accepted rounds later, through a mint call from Train items only. Selection is ontology::select(ontology, language, view), called only inside Prompt::lower, so no caller picks entries; the prompt prints each selected entry's id, layer, choice and examples. Its text is minted text for the leak diagnostic, which today has no runtime caller (follow-up leak-diagnostic-in-round). The store was rebuilt without ja-JP :14 and :15 (above-status Scene entries; a paired measure found them net-negative on register, follow-up ja-scene-entries-lower-register-agreement), so ja-JP ends at :13: an above-status out-group addressee, and an above-status in-group addressee below intimacy 700, select no ja-JP Scene entry (:12 still selects in-group at intimacy 700 or more regardless of status) until a native reader attests a replacement (ja-native-reader-scenes). ja-JP :5, :7 and :8 were revised by certainty and restraint AtMostOrUnset conditions (holds when the feature is unrated or at most the grade; AtMost still fails on an unrated feature) in a store rebuild (the store has no retire, so a conditions change is a rebuild), which makes ja-JP's hedging, certainty, affection, restraint and sadness entries disjoint on :5/:6, :4/:7, :6/:7, :6/:8, and ja-JP :3 (anger) carries AtMostOrUnset hedge and affection bounds, as en-US :3, which makes :3/:5 and :3/:7 disjoint; en-US :3 carries AtMostOrUnset hedge and affection bounds, :5 an AtMostOrUnset certainty bound (the clause it held in prose) and :8 one too (as ja-JP :7), which makes :3/:5, :3/:8, :5/:6 and :6/:8 disjoint, while :6/:10 stay co-selectable and are kept apart by scoping in :6's text; ja-JP :9 was reworded in a rebuild to state its question ending as available, so :6/:9, :5/:9 and :3/:9 stay co-selectable and are reconciled by wording, not by condition
Community (cut language-ontology, follow-up language-ontology-communities)	Key community:<slug>; type weksa.community, built fresh when the cut is mapped: data/cultural-ontology/ (draft YAML profiles, read by no code) is deleted from main (cut 5e delete-cultural-ontology) and stays in git history before that cut's commit	Revised by commit; a speaker names the communities it belongs to by id	The cultural ontology's author (owner-attested evidence only; Discord-only content stays deleted, discord-only-content-deleted)
Speaker (cut speaker-scene)	Key speaker:<slug>@<n>; type weksa.speaker: stable axes as feature values of slot Speaker (origin, socioeconomic standing, age cohort, education and whatever the loop adds), at an inventory revision, and an Annotator	A new revision appends, never edits; a corpus item and every prior turn name a speaker slug, and a round records the speaker revisions it read	The annotator it names: PersonaSnapshot (a projection of a gamecult.persona_state.v0 document, naming its source document and updatedAt), Operator, WorkMetadata (a public-domain work's own cast list and narration), or Model { family, generator } (the 2b ModelFamily)
Scene (cut speaker-scene)	Key scene:<group>@<n>, the item's split group; type weksa.scene: participants (speaker slugs) and, per ordered pair, relation features of slot Scene (relative status, intimacy, in-group, audience and whatever the loop adds), at an inventory revision (provenance; a run reads the features at its own head), and an Annotator	As Speaker. Keyed by the group, so a scene never straddles the split. data/original/speaker-state.cc holds Model-annotated keeper scenes, one per Train or Validation group of both languages (`weksa scene put	dump); a Test group has none and reads NoScene`
Corpus item	Key item:<source-slug>:<n>, n canonical decimal >= 1 (no sign, no leading zero, from cut 2b); type weksa.corpus_item with text, language, speaker, split group, prior turns, source and who placed its name spans (names_by, an Annotator, absent when unrecorded; cut pd-name-spans). Source ModelWritten carries a ModelFamily enum, the generator string as the fixture wrote it, and a prompt digest (cut 2b). The item carries no speaker state of its own: its speaker slug and group name the Speaker and Scene documents	Admitted once; the whole item is immutable (re-adding an identical item is a no-op, any difference is refused; a correction is a new item that supersedes the old); text already held under another key is refused unless the new item supersedes that key (cut 2b); withdrawn, never deleted, so old scores stay explainable	Importers and weksa corpus add, under corpus-cast-plus-public-domain; the source enum has no variant for Discord or any training-forbidden source (corpus-clean)
Split assignment	Derived: split(scheme, item.group)	Fixed for a scheme; a new scheme is a new constant, never a reassignment. A model-written item whose group derives test is in no split; a superseding item keeps its predecessor's group	The hash function. Nobody assigns by hand, so there is no writer to forbid
Loop round	Key run:<label>:round:<n>; type weksa.loop_round with spec version, profile revisions, model ids and prompt digests, sampled train items, Validation means and refusal counts (from cut ja-panel-loop: one set per language of the run's panel, no pooled mean), per-item Validation scores in corpus order (key 22, numbers only, no item key or text, private to run.rs for the paired floor of the cycle (rounds 1 to 4) and of relower; a run from before has none and opens), the digest of the inventory file it was scored under and, for a cycle's candidate round, its cycle k	Scored and recorded whole; immutable. A candidate round is put in one batch with its Decision. Lives in the run store on Yggdrasil	run::round and run::cycle (both through score_round)
Loop decision	Key run:<label>:cycle:<k>; type weksa.loop_decision with the base round and revision, the edit prompt's digest and Train keys, the editor's answer digest, the candidate revision, inventory digest and round, the base, candidate and floor readings (chrF, meaning, composite; the floor is the paired floor of round 1 against rounds 2 to 4), the threshold and the verdict (Accepted, or Rejected for edit-parse, edit-admit, no-gain, regressed chrF, meaning or composite); digests and numbers, no text	Immutable. The latest Accepted one names the run's head inventory (<root>/cycle-<k>/inventory.cc, its candidate digest, its candidate round); with none the head is the copy create made at <root>/inventory.cc. Lives in the run store on Yggdrasil	run::cycle's accept rule, run::decide: the validation composite gains more than max(noise floor, LADDER_STEP), no metric falls past max(its floor, one step), the composite does not fall past max(its floor, one step), and a non-finite reading rejects (the one regression test, run::regression); differences compared in billionths so a gain of exactly one step is a tie. From cut ja-panel-loop the readings, floor and threshold are kept per panel language and run::regression runs per language and is the one regression test (decide adds the gain test for a one-language panel): any language regressing rejects (verdicts regressed-chrf, regressed-meaning, regressed-composite), otherwise a one-language panel accepts as decide does and a panel of two or more accepts when the sum of composite deltas exceeds one LADDER_STEP (billionths). Given up: strict Pareto non-regression; a language may lose up to its own noise floor if the panel nets more than a step, because zero tolerance ratchets every base to a favourable noise draw and strangles the loop. Termination: each accept raises the composite sum by more than a step and composites lie in -0.5 to 1, so a run accepts at most 1.5 * languages / LADDER_STEP edits
Score	Key <lowering-key>:<metric>@<metric-version>; type weksa.score	Immutable; a new metric version scores again under a new key	The scorer, deterministic given the metric version and the embedder's model id. Cut scorer computes scores as values; cut golf-loop-en stores them, because the lowering keys they hang from arrive with the loop. Lands with follow-up golf-loop-score-docs, in the round's directory; nothing in the cycle reads them
Lowering trace	Key lowering:<packet-key>:<profile>@<rev>:<lowerer-digest>:<sample>; type weksa.lowering holding the surface text and its trace together	Immutable. A surface text cannot exist without its trace, because they are one record (lowering-traced)	The lowerer emits; the lowerer sees the packet without its trace (the input type has no such field), never the corpus item
Conlang lexicon entry	Key lex:<language>:<concept-id>; root form, morphology class	Root placed at its concept's embedding when the concept first needs a form in that language; stable once accepted; diachrony derives later forms, it does not overwrite	The language project's owner: the operator for her settings' languages, the loop proposing roots for her review
Conlang grammar and phonology data	<language>/grammar, <language>/phonology documents per language project	Revised by commit; every lowering names the revision	The language project's owner
Training pair (later)	Key pair:<frozen-rev>:<item-id>:<profile>	Created only by re-encoding the whole corpus under the frozen inventory revision; never before the freeze (staged-distill-every-language-equal); split inherited from the item	The re-encode job, which can only be constructed from a frozen revision

Authority map (the packet boundary)

• Owner: the validator in crates/weksa owns whether a packet exists.
• Inputs: the packet, the inventory at the packet's revision.
• Outputs: an admitted packet, or a typed refusal naming the rule.
• Derived state: split assignment (from the item's group); the lowering view of a packet (the packet minus its trace).
• Forbidden writers: the Node daemon's decomposeThoughtIntoInterlingua and normalizeInterlinguaPacket (parked, cut park-node-body); the Metame harness's packet-like YAML (parked); any loop step that writes a packet without the validator.
• Shared paths: the loop's encoder, the CLI, tests and the future live lowering all admit packets through the same validate call.
• Deletion line: the Node body goes before the crate lands, so no second packet shape survives on main.

Speaker state (persona-drives-ontology)

Operator, 2026-10-29: "The specific choice of words of a character can tell you exactly where they were from, their socioeconomic status, their relative status within the scene, and both how they feel and how restrained they are being with those feelings. The linguistic ontology holds this data and drives it from the persona." Speaker state is three layers, and each lives where its writer and lifetime put it:

Layer	Examples	Writer	Lifetime	Lives in
Stable axes	origin and community, socioeconomic standing, age cohort, education	the persona (live) or an annotator (corpus)	the speaker's	weksa.speaker
Scene relations	relative status, intimacy, in-group or out-group, who overhears	the scene (live) or an annotator (corpus)	the scene's	weksa.scene
Utterance stance	felt affect, restraint	whoever writes the utterance: the persona agent (live), the encoder (corpus)	the utterance's	the packet (stance)

Affect and restraint live in the packet, as stance, not in a speaker-state document. A per-utterance speaker-state record would have the packet's key, the packet's writer and the packet's lifecycle (minted once per utterance under one spec version, immutable), so it would be a second document under one authority; it collapses into the packet. It also has to be golfed: in the loop the encoder reads felt affect and restraint off the corpus line, as it reads the meaning, and the loop improves that reading round by round. A fixed annotation could not be golfed. Live, the agent writes its own packet (campaign-direction), so the persona drives stance through it. The stance is a separate section from the meaning (referents, predications, relations): a feeling the speaker holds back is not a claim, and packet-owns-meaning still forbids a lowering to add one. "I am angry" said outright is a predication; anger leaking through clipped syntax is stance.

The stance carries felt affect and restraint, never displayed affect. How much of a held-back feeling shows, and how, is a cultural display rule, and display rules differ by culture (Ekman and Friesen 1969; Matsumoto's cross-cultural display-rule studies; from recall, not a search this pass). If the packet carried displayed intensity, the English reading would be baked in and a ja-JP lowering would apply restraint twice. So each language's ontology maps felt affect, restraint, scene and speaker to its own surface, which is staged-distill-every-language-equal applied to affect. Whether restraint is one grade per utterance or one per felt affect is the speaker-state spec's call; the loop can split it later if one grade loses.

Demotions. formality, warmth and intensity (seed features 8, 9, 10, slots Discourse and Predication) are no longer packet owners. Formality is derived from scene relations and stable axes by the target language's ontology; displayed intensity is derived from felt affect and restraint; warmth toward the addressee is a felt affect in the stance. Cut speaker-state removes them from the seed (no committed packet store exists and nothing outside the seed and the standard names them; Body facts), so no packet can set formality beside the ontology. hedge, certainty and irony stay: they are the speaker's epistemic and pragmatic stance toward the content, language-neutral, not register.

Corpus items carry speaker state as typed ontology values, never free text. The item names its speaker slug and group; the Speaker and Scene documents hold feature values of the Speaker and Scene slots (grades, option labels, flags, and later community ids), validated against the inventory by the same feature check the packet uses. No field is free text, so no wording can reach the lowerer through annotation (no-surface-leak). Cast items: the speaker document is a PersonaSnapshot of the cast member's persona state where the standard has the value, and annotated where it does not (below). Annotations of stable axes read the whole work or persona, so a speaker's axes are not drawn from one scored line; a scene's relations read the scene, which includes its scored lines, and are bounded typed values, as a live scene would supply.

Persona state today has no field for origin, community or socioeconomic standing. gamecult.persona_state.v0 (Body facts) carries open trait maps (activationProfile.voiceStyle and five more), moodDimensions, socialBonds (trust, tension, intensity) and statusReads (authority, trust and others). So a PersonaSnapshot projects mood-like and relation-like values, and the stable axes the standard lacks are annotated. Whether the standard gains a typed speech-community reference or Weksa keeps the persona-to-community binding (the deleted data/cultural-ontology/persona-stacks/brokkr.yaml was its one draft) is question persona-speech-community-home (recommended: a Weksa-owned binding; follow-up persona-state-speech-community), decided when live-lowering is mapped; the corpus path needs neither. Putting Weksa feature ids into the open voiceStyle trait map is rejected: it would make an extension field portable authority, which the standard forbids.

Public-domain annotation and the annotator ceiling. A public-domain play has tens of speakers and scenes, so annotation is per speaker and per scene, not per line, and costs nothing per round. Stable axes come first from the work itself (WorkMetadata: the cast list, narration, stage directions). The ceiling: the loop can learn surface choices only as well as the speaker and scene values describe the line, so annotation errors become ontology errors (a servant labelled gentry teaches the ontology the wrong deference). And when a Claude annotator, a Claude encoder and a Claude lowerer share one family's idea of what low status sounds like, the loop agrees with itself and scores the agreement: the round-trip inflation the target already records. Every annotation names its annotator, so the scorer can report a test composite over operator-attested annotations alone. Who annotates is question pd-speaker-annotation.

Scoring (golf-metric-automatic, conlang-scoring, meaning-metric-embedding-cosine)

• English against its corpus item: chrF (character n-gram F-score, char order 6, beta 2, spaces removed; parity vectors against sacrebleu 2.4.3 in Body facts) and meaning cosine (sentence embeddings from qwen3-embedding:0.6b, served by Ollama on Yggdrasil; the adapter sends options.num_thread, Body facts). Probe on one reference: hedged paraphrase 0.948, flattened claim 0.643, register-shifted 0.720, negation 0.496. BERTScore proper needs per-token transformer embeddings and a torch stack; it is deferred, not adopted.
• Every language by the same meaning cosine: the embedder reads the output directly when it can (natural languages, since the embedder is multilingual); for a conlang it reads the deterministic gloss made from the language's own lexicon (Weksa owns the lexicon, so reading it back needs no model). Plus lexicon adherence for a conlang: the share of tokens that parse against its lexicon and morphology.
• A style axis follows as its own cut, behind the same scorer port.
• Names and leak (leak-owned-by-flow-and-scorer-3): no packet field carries free text, so the scorer masks nothing. A referent names an entity by id (EntityId, entity:<slug>); the entity store (weksa.entity) holds each entity's forms per language, citation form first, then inflected or particle-bound forms (data, as the inventory is). Entities are minted at intake: cast entities from the cast fixtures, public-domain entities by the importer, and in live use by the persona before it names one. A corpus item carries its name spans (byte ranges of its text, each naming an entity and on extended grapheme cluster boundaries of its NFC text and equal to one of that entity's NFC forms in the item's language; an entity holds at most 8 languages of BCP 47-form tags, 8 forms each, a form at most 64 bytes with no invisible format character), and its speaker and prior-turn speakers are entity ids. The packet store refuses a packet whose referent cites an entity the store lacks, and a packet with a corpus_item (required, typed CorpusItemId) that cites an entity neither annotated on that item nor one of its speakers: an id-set check, not a text rule. The scorer reads only the output and the reference and scores the whole of each. An honest name earns its credit ("Brokkr came back." against itself is 100), a wrong one loses it, filler costs what filler costs (Soul scorer s2: "Darling, I am truly not angry, darling." scores 46.84 whole; the output-only mask, fed the filler as a name, had raised it to 86.78), and no rule depends on script or word segmentation, so Thai, Korean particles and Turkish or Finnish suffixes are scored by the same character n-grams as English. Why this is not a leak: leakage is information the learner has that it would not have at prediction time (Kaufman et al., TKDD 2012). A live persona knows which entity it means; the annotation gives the corpus encoder exactly that, fixed at intake before any round and written by no loop writer, and the encoder chooses only which of the item's own ids to cite. The free text left is inventory labels and glosses (Train-only, the context builder's) and the trace (never lowered or scored). Diagnostics, reported and never gates: max pairwise chrF of minted text against held-out items; novel coverage pooled per order, for n = 1, 2 and 3 separately, the covered share of all novel n-grams (word n-grams of a held-out item that occur in no Train item) over a split's items, with the worst item at n >= 2. A word is a run of letters, digits, marks and joiners; any other character, and a lower-to-upper case transition, ends it, so kick:the:bucket and a snake_case OptionLabel read as words. Han, kana, Thai, Lao, Khmer and Burmese write no word boundaries and the scorer holds no dictionary, so a run of them is one word (per-character words would give every honest gloss shared bigrams); a leak there shows at n = 1 and in chrF. Honest glosses of single words saturate n = 1 (the 54 seed glosses read max 1.000, s2), so the leak signature is pooled n >= 2. Costs: a live persona mints before it names (one typed write, which also buys coreference and one rendering per language); public-domain intake annotates name spans (follow-up pd-name-spans, annotator per pd-speaker-annotation); the lowerer inflects a name from its stored forms or adds one, traced. Prior art (recall): AMR keeps :name strings inside the graph beside a :wiki link, and AMR-to-text systems copy those strings; Abstract Wikipedia renders constructors over Wikidata item ids, with labels per language and Lexeme forms for inflection; Grammatical Framework lexicons hold proper names (PN) with per-language inflection tables; referring expression generation (Reiter and Dale 2000; the GREC tasks, 2008-2010) chooses a name, pronoun or description from an entity id and the discourse. Rejected paths: text rules at the inventory and packet doors (Soul packet-store s2); any mask of names in the scored texts: masking reference and output (s1: "Kick Bucket" took chrF 53.73 to 100) and masking the output only (s2: named filler took chrF 46.84 to 86.78, cosine 0.728 to 0.886); equal-length placeholder masking (bounds chrF, not cosine, and still pays nothing for an honest name); character 6-gram coverage (one-word glosses read 0.000); max-over-items coverage (saturates on honest runs).

Body facts (probes of 2026-10-29, map pass for speaker state, 2b and scorer)

• Weksa main 869c82b (packet-store, 2b corpus labels, 2c entity store, scorer landed; packet schema v3). Annotators land with their consumers: Operator and Model { family, generator } with cut speaker-scene (the cast seeds are Model-annotated and an operator attests by appending an Operator revision), WorkMetadata with pd-importer, PersonaSnapshot with live-lowering.

• UAX #29 word probe (unicode-segmentation 1, scratch crate, 2026-10-29): unicode_words() counts (whitespace count in brackets) "Brokkr" 1 [1], "Mary Ann Evans" 3 [3], "de la Cruz y Garcia" 5 [5], Thai "สมชาย" 5 [1], "สมชาย ใจดี" 8 [2], "山田太郎" 4 [1], "李小龍" 3 [1], "ブロッカー" 1 [1], "김민준" 1 [1], "Jean-Luc" 2 [1], "O'Brien" 1, five letters glued by ZWSP 5 [1], glued by WJ U+2060 1 [1]. Without a dictionary UAX #29 splits Thai and Han per character, so a 4-word UAX #29 bound refuses ordinary Thai names (question form-word-bound, recommended bytes-only). NFKC, default case fold, NFKC, alphanumerics only: "fine" -> "fine", "Ｂｒｏｋｋｒ" -> "brokkr", ZWSP and WJ drop out.

• Weksa main 28284cc: cargo test -p weksa on Yggdrasil (ygg-verify, rust image) 73 passed. weksa corpus dump holds three items: item:brokkr:1 (group brokkr-edda-voice, Validation, ModelWritten { model_family: "local-source-aware-draft" }), item:nibu:1 (group nibu-proof-001, Test, CastAuthored, fixture examples/english-lowering/nibu-proof-001/lowered-output.yaml, which records no author) and item:nibu:2 (same group, excluded, ModelWritten { model_family: "unrecorded-agent-run" }; that string exists only in corpus.cc). The Brokkr fixture's generator was codex-local-source-aware-draft at 4197123 and reads local-source-aware-draft on main.

• CorpusStore::add compares only text (store.rs:166-171): re-adding a key with the same text and a different source returns Unchanged and writes nothing, so a relabel through add silently does nothing. key_slug parses n with u32::parse, which accepts 01 and +1.

• Seed inventory (2026-11-10, base 5e10d7c): features 6-8 are hedge, certainty and irony; 9-18 are the stance affects and restraint (slot Stance). The committed inventory holds revisions 1-2 only. Nothing calls with_scene except tests, and the proof packet's addressee carries no entity (its referent n3 does: entity:nibu).

• Yggdrasil's Ollama (127.0.0.1:11434) is the Docker container ollama-cpu, Huginn's embedder (runbook gamecult-ops/runbooks/huginn-yggdrasil.md, ruling eureka-body:ruling:huginn-embedder-yggdrasil-primary). /api/embed with an input array returns unit-norm 1024-dimension vectors. Under host load (load average 18-24 from ygg-verify rustc jobs) and no num_thread: a batch of 6 took 21.5 s and 29.8 s, a batch of 32 timed out at 120 s. With options.num_thread 8: a batch of 6 in 2.1 s, a batch of 32 in 5.8-6.4 s (0.18-0.20 s per item); num_thread 4: 8.8 s for 32. Cosine of "Take your hand off that control." against its lowercased, unpunctuated form: 0.943. One server serves Huginn and the loop, so a 32-item batch holds it about 6 s.

• chrF parity: sacrebleu 2.4.3 sentence_chrf(hyp, [ref]) and a Python reimplementation (orders 1-6, per-order precision and recall averaged over the orders where both sides have n-grams, F-beta with beta 2, spaces removed) agree to 4 decimals on six pairs (reference / hypothesis / score): "Take your hand off that control unless you want your nervous system to become my least useful circuit diagram." / "Take your hand off that control, unless you'd like your nervous system to be my least useful wiring diagram." / 74.4498; "The command has no receipt. It is not truth." / "The order carries no receipt, so it is not true." / 45.7164; "I am not angry." / the same / 100.0; "Forgive me, sir." / "Get out." / 3.9683; "その制御に触るな。" / "その操作盤に触らないで。" / 14.2931; "Well, aren't you a sight for sore eyes, love." / "" / 0.0.

• Scorer s1 follow-up probe (sacrebleu 2.4.3 in python:3.12-slim on Yggdrasil; qwen3-embedding:0.6b, one 8-text call). sacrebleu strips every whitespace char before gramming: hypothesis "a b c", "a b", "ab", "a…b", "a
b c", "line one line two", "その　制御" against the same text with that char removed (or "line one line two") all score 100.0; "a​b" against "ab" scores 45.4545 (zero-width space is not whitespace). Name scoring, output masked and reference whole (hypothesis / reference / chrF / cosine): "He is going to die tomorrow." / "He is going to kick the bucket tomorrow." / 53.7347 / 0.9081; "He is going to the tomorrow." (named or omitted) / same / 55.9583 / 0.8246; "Not ever again." / "Never again." / 85.1384 / 0.9169; "." (named) / same / 11.1111 / 0.6456; "came back." / "Brokkr came back." / 56.6137 / 0.7364; "Kestrel came back." / same / 51.3670; "not go." / "Will will not go." / 33.6248.

• gamecult.persona_state.v0 (F:\Projects\Epiphany\schemas\cultnet\gamecult.persona_state.v0.schema.json): required activationProfile (six open trait maps of mean, plasticity and current activation) and affect (needs, socialBonds, statusReads, moodDimensions, socialBiases, doctrineStances); presentation.voiceSummary is free text; no field names origin, community or socioeconomic standing; extensions are non-authoritative by the schema's own description.

• Probes of 2026-11-12 at d9ebbcf (map pass for cut 5): no code reads data/target-language-ontology/, data/cultural-ontology/ (since deleted) or .weksa/cultural-ontology/starter-targets.cc (a third, typed copy of en, ja-JP and pt-BR profiles); README.md, docs/multilingual-utterance-lowering.md, examples/cultcache-lowering/ and prompts/nibu-cultcache-target-*.md cite them. LoweringInput::with_scene has no caller in run.rs or main.rs, the run pins no SpeakerStateStore, and no Speaker or Scene document is committed (speaker-scene-seeds open), so only the packet's Stance and Discourse features reach a loop's lowering view. score::leak_diagnostic and TestReadings::at_checkpoint take minted text as &[&str] and have no caller outside tests: no round reads glosses or any minted text against Validation. data/daodejing/inventory.cc is a revision prefix of data/original/inventory.cc (revisions 1 to 4 equal, a test proves it; revs 5 and 6 add the bed's own lexicon); with data/interlingua/inventory.cc they share revision 1, where features 6 to 8 (hedge, certainty grades, irony flag) sit in Discourse and Predication and features 9 to 18 (anger to restraint, grades) in Stance. PROMPT_SOURCE_DIGEST covers context.rs, split.rs, lower.rs and packet.rs. 25 RunPaths { literals across run.rs, context.rs doctests, main.rs, tests/common/mod.rs, tests/golf_loop.rs and tests/original_corpus.rs; 10 Prompt::lower( calls. huginn known timed out (Unavailable, 15 s deadline) twice this pass.

Where things live and run (rust-single-crate, loop-runs-on-yggdrasil)

• Code: one Cargo workspace at the repo root, one package crates/weksa (a library and the weksa binary). CultCache through cultcache-rs pinned by git rev, as Huginn pins it. No daemon: the loop is a batch command. A daemon is earned later only by the live lowering service, if isolation protects a named invariant there.
• Stores: data/interlingua/inventory.cc and data/corpus/corpus.cc in the repo (small, reviewed with weksa dump); run stores (rounds, lowerings, scores) on Yggdrasil under the loop's state root, not in git.
• The loop is a batch command driven from a Claude Code session on Yggdrasil (golf-loop-in-claude), not a detached unit. Model calls go through a ModelPort whose one adapter, AnswerFiles, writes a .prompt file and reads a .answer file: Claude agents answer them, one fresh subagent per encode prompt (encoder-isolation-subagent-per-prompt) and a tool-less relay for lowering prompts (lowerer-call-reach-toolless-relay). Builds, stores, scoring and embeddings run on Yggdrasil; Starfire runs none of it, and Weksa needs no claude binary or token there. Never OpenAI (eureka-body:ruling:no-return-to-openai-supplier), never Discord data (gamecult-chat:ruling:discord-taints-training).

Convergence (convergence-test)

The spec has converged when, on a panel of en-US, ja-JP and the first conlang, for 20 consecutive rounds no accepted change to a closed-class inventory family or the schema raised the validation composite past the noise floor, and at that checkpoint the test-split composite is within the noise floor of validation. The test split is read only at checkpoints. Self then records the freeze, and the training set is the whole corpus re-encoded under the frozen revision.

Distil measure (distil-slice)

A lowerer is compared on round n's Validation packets with the prompts byte-identical to the round's, by weksa loop relower ROOT --round N --arm ARM --lowerer LABEL [--floor ARM2 --floor ARM3 --floor ARM4]. Every arm is answered through the one prompt door (RunStores::answer_files, the Test-text guard); a tuned model is a script that reads <digest>.prompt and writes <digest>.answer in <ROOT>/relower-<N>-<ARM>, not a new adapter. The recorded reading is the round's own lowering, read over the same items as the arm. An item the round refused before admission (no packet) or at the lowering prompt has no prompt to ask: it is skipped, so the readings exclude it, and it is counted as skipped= on the line; n is the items read, so recorded can differ from the round's own Validation reading, which scores those items 0. An arm answer that is not a usable lowering scores 0 and counts in unusable=; the same for the floor arm, counted (over all floor arms) after floor ... unusable=. The floor needs three or more floor arms, each a prompted repeat of the recorded lowering on the same prompts: it is FLOOR_Z (3) standard errors of the mean per-item difference between the recorded lowering and a repeat, per metric, with the variances pooled over the arms (paired_floor; lowering-only noise, no encoder noise). The verdict is decide_panel over (recorded, arm, floor): Accepted is a composite gain past the noise of the panel's sum (the root of the languages' composite floors squared, at least one step), Rejected(Regressed) a loss past a language's floor, Rejected(NoGain) neither. The cycle's own floor is the same paired_floor (floor_of): round 1 (the recorded lowering, the head's establishing round) against rounds 2 to FLOOR_ARMS_MIN + 1 (prompted repeats under the run's first inventory), per panel language, from the rounds' per-item Validation scores. It is measured once per run and is the same after an accept; a run needs four loop round calls before its first loop cycle. Calibration (runs/distil-base, 8 prompted draws by one model family on identical prompts, no draw sharing more than 7 of 28 answers with another; 14 en-US and 14 ja-JP Validation items; every base, candidate and 3-arm floor combination, 1120 trials from 28 unordered pairs; intervals bootstrap over draws, each degraded arm one fixed draw): a prompted repeat reads NoGain 92% (80 to 100), Regressed 8% (0 to 20), Accepted 0%, which with 28 pairs bounds false-accept only to about 10%; an all-"Yes." arm reads Regressed 100%; a mild degradation (one detail blurred per utterance) 100%; a subtler one 75% (50 to 100), whose mean composite drop (0.044) sits under the mean floor (0.060). FLOOR_Z was chosen on this data (2.5 gives NoGain 84% and 93% on the subtler arm), so the rates are in-sample. Without the noise term in the gain test a prompted repeat reads Accepted in 40% of the trials. Power (can the rule Accept a real gain?): the only better arm that could be built without Test or Validation answers is a lowerer shown the Train sentences of the corpus (no packets, no Validation or Test text) as a register guide. Four such draws beat the prompted draws' mean composite in both languages (en-US 0.742 to 0.759 to 0.795; ja-JP 0.559 to 0.578 to 0.620), so the premise that it is better is measured on the very Validation items the rule reads, not independent. Against the prompted bases with 3-arm floors the rule reads it Accepted in 29% of 1120 trials, NoGain 71%, Regressed 0%: a net gain of 0.079 against a mean noise term of 0.079. At 14 items per language the rule is therefore a low-power filter for a gain of this size; by 1/sqrt(n), about 35 or more items per language would bring the noise term near 0.05 (an extrapolation, not measured).

Validation growth (validation-growth, built and recalibrated)

Power needs about 35 or more Validation items per language. data/original now holds 35 per language (it held 14; Train 14 and Test 8 are unchanged): 42 model-written items, item:vgrow:1..21 (en-US) and item:vgrow-ja:1..21 (ja-JP), live Train+Validation 56 to 98. The slug is new on purpose: the item:original: prefix filter of DIGEST_OF_THE_28 still selects the first 28 and passes; DIGEST_OF_THE_42 freezes the new prose.

The split is derived from an item's group alone (SplitScheme::V1: FNV-1a of weksa-split-v1:<group> mod 100, under 40 Train, under 80 Validation, else Test). The names were chosen before any text existed: 90 candidate nouns per language, dropped by script when the group already existed or the noun occurred in a Test text, then the first 24 whose bucket fell in 40..=79 (the bucket came from SplitScheme::V1 itself, cross-checked on the 72 existing items). Each language was written by three blind writer agents (8 nouns each, a brief of language, register note, noun and speech act, no Test text and no existing item), 24 texts per language written, the first 21 in name order admitted through weksa corpus add; each group is its own group. The scenes are weksa scene put documents (keeper to unnamed, inventory rev 4), annotated by a separate agent; a second annotator measured agreement only and is not committed.

The writer is Claude (family Anthropic recorded): no cross-family generator was reachable from the body that ran this cut. The self-preference effect (a Claude lowerer favouring Claude prose, Panickssery et al. 2024) is therefore not measured. The writers' blindness is by brief content, not sandbox.

Expect encode refusals: the existing 56 were written onto the Dao r3 concepts and the inventory was drafted from them, while the new items were written free of the inventory (question validation-growth-vocabulary stays open). Some new packets will be refused at encode (inventory coverage) and score 0 as the means do; they stay in the corpus. That is a result about inventory generalization, not a defect.

Every calibration number in the Distil measure section above (92/0/8 prompted rates, the 29% power, the 0.079 noise term) is for the 14-item corpus and stays the stated baseline of that corpus.

Recalibration on the grown corpus (runs/growth-base, main 44b66f9, panel en-US and ja-JP, round 1 recorded by briefed subagents; 6 prompted draws including the recorded one, 2 mild degradations, an all-"Yes." control and 4 Train-guide draws by Claude sonnet-5-5; the same decision rule at FLOOR_Z 3 and 3-arm floors; 120 prompted trials from the 30 ordered pairs of 6 draws with the other 4 as floor candidates, so trials share draws and are not independent). Soul pass cut-validation-growth.s3 re-derived every number below; its findings shaped the wording.

• Live Validation items: 28 en-US and 32 ja-JP of 35 each; 10 are refused at parse (en-US original:18, 23, 25, 27 and vgrow:4, 5, 6; ja-JP original-ja:17, 22 and vgrow-ja:2); relower skips an item with no packet, so every reading here is over live items only. Old 14: 10 en-US and 12 ja-JP live; new 21: 18 and 20.
• Prompted repeat: NoGain 96.7%, Accepted 3.3%, Regressed 0% (14 items: 92 / 0 / 8). All four false accepts are one pair, base p6 and candidate the recorded lowering R: R comes from the round, the other five draws from relower, and in en-US R scores +0.024 per item above them [+0.002, +0.052], so R is not exchangeable with a repeat (a Plausible Soul finding, cause not isolated). No false-accept bound can be drawn from 6 draws.
• Degraded arms read Regressed: all-"Yes." 100% (as before); mild-1 31.7%; mild-2 0% (14 items: 100% and 75% on a subtler arm). The mild arms are weak and not independent: they were rewritten from R with one detail blurred, the original wording is not on disk, mild-2 equals R word for word in 5 of 60 items and mild-1 in 2, and mild-1 equals mild-2 in 27 of 60 although they were asked to blur different details. Their mean composite is 0.003 to 0.015 below the prompted mean (the old subtler arm: 0.037 and 0.052). The sensitivity to a small regression is not re-measured.
• Noise: the panel noise term of the sum rule is 0.045 on average over the base and floor choices (0.031 to 0.064), against 0.079 at 14 items. The 1/sqrt(n) forecast (about 0.05) assumed per-item variance stayed the same; it did not: the per-item SD of the difference between two prompted draws is about 0.085 on the old items and 0.034 on the new ones, so the pooled term landed near the forecast because less variable items were added. Not a confirmation of the forecast.
• Power: the Train-guide arm (14 Train sentences as a register guide) is Accepted in 21.2% of 240 trials, NoGain 78.8%, against 29% at 14 items, and its mean composite gain over the prompted draws is 0.0185, 0.41 of the mean noise term (0.37 on the first floor choice per base); with relower-only bases (R removed) it is Accepted in 35% (28 of 80). Split by item class (paired bootstrap over items, 95%): old items +0.029 [+0.005, +0.052] (en-US +0.049 [+0.023, +0.084] on 10 items, ja-JP +0.013 [-0.021, +0.037] on 12); new items -0.002 [-0.011, +0.007] pooled, but two opposite effects: en-US -0.017 [-0.027, -0.007] on 18 items and ja-JP +0.011 [+0.001, +0.022] on 20. So the old-versus-new contrast exists in en-US; in ja-JP old (+0.013) and new (+0.011) do not differ. Measured: the 14-item gain does not carry to the new en-US items. Inferred, not isolated: that it was in-sample register fit. The new items differ from the old in several ways at once (written free of the inventory, longer, a different scene convention, lower scores, a smaller vocabulary overlap with Train in en-US: 0.29 against 0.52) and were briefed to the same register as Train, by Claude as the old were. The 29% power figure stays the stated baseline of the 14-item corpus; no better arm is known on the new items, so the rule's power against a real gain is not re-measured.
• The two classes are different strata (Soul s2 register-stratum-shift): the new items score lower (composite 0.41 against 0.64 pooled) so a pooled 60-item mean mixes them; report old and new separately.
• The self-preference effect is not measured (ruling validation-growth-writer-claude-checked). The new scenes use a different in_group convention from the old ones (follow-up vgrow-scene-in-group-convention), so the new-item scores are for scenes that will be re-annotated; the speaker-state store changes then (a run pins its digest), the corpus does not.

For growth-base, weksa loop calibrate (docs/calibration.md) reproduces from run.rs's own scoring and accept rule the counts per class, the trial totals and the mean noise of this block: prompted 116 NoGain, 4 Accepted, 0 Regressed of 120; control-yes 60 of 60 Regressed, mild-1 19 of 60, mild-2 0 of 60; the Train-guide arm 51 Accepted, 189 NoGain, 0 Regressed of 240; noise 0.0451; items en-US n=28 skipped=7, ja-JP n=32 skipped=3. By class it adds: old items (item:original*) prompted 111/9/0, noise 0.1069, Train-guide 71 Accepted of 240; new items (item:vgrow*) prompted 116/4/0, noise 0.0334, Train-guide 0 Accepted and 74 Regressed of 240. The other figures of this block (35% with R removed, the noise range, the 0.0185 gain and 0.41 ratio, the bootstrap intervals, the per-item SDs, R's +0.024, the mild-arm duplicate counts and mean composite offsets, the Train-vocabulary overlap, the old/new class composites, the ten parse-refused item ids) were computed by scratch scripts outside git (Soul finding evidence-outside-git) and are not reproduced by the harness.

Recalibration on the relabelled scenes (runs/growth-scene, main d8afc4a; same panel, 35 live items per language, 0 refused, 5 prompted draws including the recorded one R, 2 mild, 4 Train-guide). It is not a replacement for the growth-base block above and no delta against it is drawn: the item set, the packets and the scenes changed, and encode was redrawn (the 98 encode prompts are byte-identical to growth-base, no encode answer is; the refusals fell from 10 to 0 on the redraw, not through the scenes, which reach only the lowering prompts). Soul cut-calibrate-harness.s3 re-derived all 33 counts and 3 noise terms below from the per-item answers without calling calibrate; they match.

Measured, and reproduced by weksa loop calibrate (trials: prompted 120, each mild 60, Train-guide 240; Accepted / NoGain / Regressed):

class	prompted	noise	mild-1 Regressed	mild-2 Regressed	Train-guide
all (35 per language)	11/100/9	0.0392	51	58	85/35/120
old (14)	0/105/15	0.0844	60	60	40/80/120
new (21)	10/97/13	0.0340	3	4	50/129/61

Measured by Soul scratch scripts outside git (not reproduced by the harness, except the R split in the first bullet, which calibrate prints as prompted-r-candidate, prompted-r-base and prompted-r-free since the calibrate-report-r-involvement cut; Soul re-ran growth-scene with it and the printed all lines read r-candidate 11/9/0, r-base 0/11/9, r-free 0/80/0, as below, in Accepted/NoGain/Regressed; likewise the per-arm Train-guide counts (candidate=h1..h4) and the mild arms' composite offsets from the prompted mean, R included (offset), which calibrate prints since calibrate-arm-counts-and-mild-offsets and which reproduce the figures below, Soul re-run on growth-scene):

• Every one of the 11 false accepts has R as the candidate (bases prompted-2, p4, p6); every one of the 9 false rejects has R as the base. The 80 trials without R read 0 Accepted and 0 Regressed in every class (95% upper bound 4.5%). R scores above the other prompted draws (+0.012 en-US and +0.023 ja-JP on all items), so R is not exchangeable with a repeat. Which of the two rates the live rule should be held to is undecided: the live loop meets R as the head's establishing lowering, so the R-candidate rate may be the real one for cycles, while the 80 R-free trials are the exchangeable repeat rate; neither number stands for the other.
• The old/new split of the false-accept lives only in the 20 trials where R is the candidate, from 5 bases: new 10/20 [0.27, 0.73], old 0/20 [0.00, 0.17] (Clopper-Pearson); by base 3 of 5 against 0 of 5 (Fisher p 0.17). The 10/120 against 0/120 comparison treats dependent trials as independent. The claim that the new items drive the prompted false-accept is not supported; the old items' larger noise term (2.4 times) raises their threshold, which alone can produce the gap.
• Mild arms: they overlap 8 of 35 answers (en-US) and 9 (ja-JP) over the limit of 7. They do not degrade the new items (composite -0.004 en-US [-0.021, +0.012], +0.013 ja-JP for mild-1, +0.005 for mild-2); the Regressed counts come from the old en-US items (-0.072). The rule's sensitivity on the new items is unmeasured.
• Train-guide pools unlike arms. h1 and h2 Regress in 120 of 120 trials; h3 and h4 gain (Accepted 45 and 40 of 60); on the new items h3 is Accepted 50 of 60 and h4 0 of 60. The 35% pooled figure is not a power.

Extrapolated only: the rule's power at 35 items against any real gain beyond h3 and h4, and any false-accept bound on repeat draws (5 draws, one of them R).

Smallest measurable distil slice (distil-incontext, not built)

Distil is a tuned lowerer measured against the prompted one; the tuned model needs a host (follow-up distil-host-question-owed). The smallest slice that measures whether Train pairs buy a gain with the rule that exists is training-free: an in-context arm, the prompted lowerer given Train (packet, text) pairs as few-shot examples, answered through the one prompt door and read by loop relower against three prompted repeats. Parts: 12b's Train-only export (follow-up distil-export-at-freeze, guarded against Test text) as the few-shot source; the arm's directory of answers; the existing floor and verdict. A weaker version ran on 14 items: a lowerer shown the 28 Train sentences (no packets) beat the prompted draws in both languages and the rule Accepted it in 29% of 1120 trials. The slice is decisive only with the Validation growth above: at 14 items it can only fail to reject. It decides the fine-tune question in one direction: if in-context Train pairs do not clear the floor at 35 items, a tuned model's gain over the prompted one is not measurable with this corpus either.

Training pairs are Train only, at a frozen revision; the export is 12b.

Lowerer first cut (golf-loop-lowerer-cut)

Authority map for the smallest lowerer, written before its code (Imagination pass by the Weksa Self run, 2026-11-01). The cut exists to give the golf loop a measurable baseline from a packet: lower gives text, score gives numbers.

Decision	Owner	Inputs	Outputs	Not the owner
Whether a packet exists	validate / PacketStore::admit (unchanged)	packet, inventory, corpus, entities	admitted packet or typed refusal	the lowerer
Wording of a scored item in English (golf loop, golf-loop-model-lowerer)	the ModelPort's answer to Prompt::lower (context.rs), read as LoweringDraft (encode.rs) and held in run::score_round's locals	the admitted view, the inventory at its revision, en-US entity forms	{"text"}, scored once, never stored	the corpus item, the scorer, the editor
Wording of a packet in English (weksa lower, proof fixtures)	crates/weksa/src/lower.rs: lower(&LoweringInput, &Inventory) -> Result<String, LowerError>	the validated lowering view and the inventory at its revision	text, or a typed refusal naming the missing template	the corpus item, the scorer, the golf loop
English template table (concept words, frame templates, speech-act openers)	the same module, as rows (inventory id, gloss it speaks for, surface)	none; it is data in code	rows the lowerer looks up	the inventory (it owns the ids and glosses, not the English)
The inventory's concepts and frames	an inventory revision; here inventory::proof_revision() (rev 2), hand-seeded as seed() is	none	the 6 concepts and 3 frames nibu-proof-001 and the consent-gated request packet name	the lowerer
What a score counts	Scorer::score (unchanged)	output text, reference text	ItemScore	weksa score, which only reads text and calls it
The proof packet	packet::proof_packet(), typed ids of rev 2, admitted by weksa packet proof <store> through PacketStore::admit. Not JSON: Packet's serde is CultCache's positional tuple, and JSON in is the xenos boundary the cut 7 encoder owns	none	admitted packet	the CLI decides nothing about validity

Decisions this map makes, and why:

• lower takes &LoweringInput, not &Packet. A Packet carries trace and provenance.corpus_item; the model page says the lowerer sees the packet without its trace and never the corpus item (lowering-traced, leak-owned-by-flow-and-scorer-3). validate already yields the view, so the CLI runs validate and hands it over. Cut 4 narrows the door to PacketStore::admit; this cut does not wait on it.
• The frame owns the verb phrase; a predication's concept must resolve but is not rendered. Referent concepts supply nouns. A frame template names its roles by inventory gloss ({agent}), resolved against the inventory at the packet's revision, so a retired or re-glossed role cannot be silently mis-keyed. A row whose gloss differs from the inventory's entry for its id is a LowerError, so a table row cannot speak for a different meaning than its id now carries.
• Rendering rules, all deterministic and all from packet fields: the discourse addressee renders "you" ("yourself" when it already filled a role in the same predication), other referents render "the ", a predication filling a role renders "that", the speech-act row supplies an optional opener, sentences are capitalised and end in a full stop. Entity-bearing referents render by concept word: the lowerer does not read the entity store in this cut (follow-up: entity forms in the lowerer).
• No style, no register, no features. Hedge, certainty, modality and polarity are not rendered, so the baseline scores what meaning alone buys. A lowerer that ignores polarity: negative would assert the opposite, so lower refuses any packet that carries a feature, relation or extension it does not render rather than render it wrong (packet-owns-meaning); features arrive as templates learn them.
• Deleted by later cuts, not this one: the table in lower.rs and proof_revision() are temporary owners. Cut 5 (language-ontology) moves the table to weksa.ontology_entry documents selected by the target profile; cut 7 replaces proof_revision() with the loop's minting round. Nothing is deleted today because no earlier lowerer exists; the YAML examples/english-lowering/ hand-lowerings are fixtures, not owners, and stay.
• weksa score is a thin reader of Scorer::score: weksa score <item>... [--output FILE] embeds the output text once against every named item. It stores nothing (Score documents arrive with cut 7) and gates nothing.

Cut sequence

One line per cut; specs exist for the cuts about to run (2d and 4, siblings on 869c82b; 2d merges first and 4 rebases onto it).

1. park-node-body: park the Node daemon, MiMo command, RUDP shims, launchers and the Discord-sourced Metame harness at tag parked/weksa-node-body; README front door tells the truth. Landed.
2. packet-store: Rust crate with packet, inventory and corpus types, the validator, the derived split, the stores and the weksa CLI. Landed at 28284cc. 2b. Landed. packet-store-corpus: corpus labels tell the truth (corpus-labels-tell-truth): ModelFamily and the generator string on ModelWritten; item:nibu:1 ModelWritten (Unrecorded); item:brokkr:1 and its fixture carry codex-local-source-aware-draft, family OpenAi; intake refuses aliased keys, a whole item changed under its key, and a held item's text under a new key; corpus.cc reseeded. Speaker state is not in 2b: it lives in Speaker and Scene documents and the packet's stance, not on the item. Waits on 2. (spec) 2c. Landed. entity-store: ProperName and Referent.name deleted; EntityId, Referent.entity, the weksa.entity document and its store; corpus name spans and entity-id speakers; the packet store's entity checks; seed entities and spans for the cast items; schema v2. Waits on 2b. (spec r1)
3. Landed at 869c82b. scorer: chrF, meaning cosine behind an Embedder port (Ollama adapter with num_thread), leak diagnostics, aggregation by derived split (Train and Validation only on the round path; Test only through TestReadings::at_checkpoint, which consumes a Checkpoint); library only, weksa score comes with the loop. h2 landed the r2 fix pass on its branch; r3 deletes the name mask (the scorer reads output and reference only) and pools novel coverage per order over word keys that split joined labels. Waits on 2c for merge order. (spec r3) 2d. entity-door-lows: invisible format characters refused in entity forms (ZWJ, ZWNJ, CGJ and variation selectors kept); the form length bound per question form-word-bound (recommended: bytes only); attested origins (Cast { persona, fixture }, Live { persona }) and CastAuthored.author; empty provenance strings refused at the store; S6 under NFKC plus case fold and no-alphanumeric texts refused; next_key refuses at u32::MAX; CLI flags that do not apply refuse and no error echoes a value. Owns follow-ups entity-door-lows and s6-unicode-normal-form. (spec r1) 3b. lowerer-first-cut (follow-up golf-loop-lowerer-cut): the section above. lower, weksa lower, weksa score, weksa packet proof, inventory rev 2 for nibu-proof-001.
4. speaker-state: packet stance (felt affect and restraint, slot Stance, packet schema v4); formality, warmth and intensity out of the seed and the standard; only PacketStore::admit yields a LoweringInput, and it refuses packets for withdrawn items (follow-up lowering-input-from-packet-store). (spec r1) 4a. speaker-relation-input: slot Scene and its first seed features (relative_status, intimacy, in_group; inventory revision 3, now speaker_scene_revision); LoweringInput::with_scene carries the speaker's relation to the addressee as typed Scene feature values. Since 4b it takes the SpeakerStateStore and the CorpusStore, reads the scene of the view's own item group for the item's speaker, and selects the relation by the addressee referent's entity (V16 when the scene does not list the item's speaker or the speaker of any live item of the group, the rule put_scene applies too; V17 refuses an addressee that names a predication; V13 is retired). The English lowerer does not render it. (PR for follow-up chat-s3352-3352) 4b. speaker-scene (spec r1; speaker-scene-seeds is its cut 4c: CLI, inventory revision 3, cast speakers and scenes; Party::Unnamed covers an addressee with no entity; Speaker and Scene share one SpeakerStateStore): slots Speaker and Scene with their seed features; weksa.speaker (speaker:<entity-slug>@<n>) and weksa.scene (scene:<group>@<n>, participants and ordered-pair relations) documents with Annotator Operator and Model { family, generator }, validated by the packet's shared feature check, append-only per key; their stores, weksa speaker and weksa scene commands, published schemas; Model-annotated cast speakers and scenes for the committed items. Waits on 2d and 4; pd-speaker-annotation blocks only the public-domain annotator. Part 1 landed (branch weksa-lowering/speaker-scene): speaker.rs, SpeakerStateStore, slots Speaker and Scene exclusive to their own feature lists, speaker_scene_revision() (Speaker features 22-24), with_scene on the store, schemas. The CLI, the committed inventory revision 3 and the seeds stay with 4c.
5. language-ontology, in sequential cuts on d9ebbcf. 5a language-ontology (spec r1): weksa.ontology_entry, OntologyStore, ontology::select over the view's Stance and Discourse features, Prompt::lower printing the selected entries of the item's language, RunRecord pinning the ontology store (keys 15 and 16), weksa ontology admit|dump; ontology.rs joins PROMPT_SOURCE_DIGEST. 5b language-ontology-seed (spec r1, after 5a): Model-annotated en-US and ja-JP seed entries drafted by an isolated model, data/target-language-ontology/, .weksa/cultural-ontology/ and schemas/target-language-cultural-ontology.md deleted. Lost until 5c': keigo, hierarchy and uchi-soto hang on the scene relation from 5c', so ja-JP politeness is still the lowerer's guess. 5c' language-ontology-scene (spec r1, after 5b): the run pins the SpeakerStateStore (keys 17 and 18, Drift(SpeakerState)), RunStores::assembled is the one door that reads the item's scene into a loop view, select reads slot Scene, Prompt::lower refuses a view whose scene was never read, weksa scene put|dump and weksa inventory speaker-scene (rev 4 in data/original and data/daodejing, rev 3 in data/interlingua, which holds no Dao revision), Model-annotated keeper scenes and Scene-conditioned entries for data/original. 5c language-ontology-view-assembly (follow-up, now only the Speaker read): select widens to slot Speaker; waits on speaker-scene-seeds. 5d language-ontology-communities (follow-up): weksa.community built fresh (data/cultural-ontology/ is deleted by 5e), community ids on weksa.speaker, community conditions; waits on 5c and question persona-speech-community-home. 5e. delete-cultural-ontology (steward ruling, 2026-11-13): data/cultural-ontology/ (five draft YAML profile files and a README; old-norse-edda-source-tradition.yaml did not parse) deleted. Nothing in crates/, tests, scripts or CI read it; examples/persona-lowering/brokkr-edda-voice/ cites its paths as text only (the one test that reads that example reads lowered-output.yaml). The typed weksa.community of follow-up language-ontology-communities is built fresh under cut 5; follow-up edda-referent-yaml-parse is moot.
6. pd-importer: public-domain play and fiction dialogue into the corpus with speaker, scene group, edition provenance, and Speaker and Scene annotations by the ruled annotator. Waits on 4, follow-up gutenberg-terms and question pd-speaker-annotation. 6a. pd-name-spans (spec r1): weksa corpus import <work> <names> mints a work's cast as PublicDomain entities with forms in the work's language and adds each line a name-span answer covers, speaker from the work file, spans from [[entity|form]] markup, labelled names_by (Model or Operator) per docs/name-span-annotation.md. Speaker and Scene documents, prior turns and fiction speaker attribution stay with 6. 7a. loop-context-builder: context.rs (TrainItems, EncoderOffer, Prompt: corpus text reaches a model call only through Prompt::mint over live Train items or Prompt::encode over one offered item; instructions are the consts MINT_INSTRUCTIONS and ENCODE_INSTRUCTIONS in context.rs; Checkpoint has no public constructor, offers each item once, and its offers borrow it (the gate is typed TestReadings and Test offers on the round path; operator tools such as weksa corpus dump and weksa score still read corpus text of any split); prior turns print the held Train text), SplitScheme::live, CorpusItem::cited_entities (one owner of the id set PacketStore::admit and the offer share), and checkpoint-only Test readings (Checkpoint, TestReadings; Aggregate and leak_diagnostic carry no Test). Library only; since 7b TrainItems::from_corpus and EncoderOffer::for_round are crate-private and RunStores is their one public producer. Merges follow-ups loop-context-builder-2 and scorer-test-readings-type. (spec r1) 7b. golf-loop-round (spec r1; branch weksa-lowering/golf-loop-round): the smallest loop that scores one round. run.rs: RunStores pins the corpus and entity stores of a run by canonical path and FNV-1a 64 digest at create, reopens only from <root>/run.cc and refuses drift (corpus, entities, PROMPT_SOURCE_DIGEST of context.rs, a second pin on lower.rs retired in 7d, model, embedder); round encodes every live Train and Validation item through a ModelPort (model.rs; the one adapter AnswerFiles writes <digest>.prompt and reads <digest>.answer, so the model is the Claude session), reads each answer into a PacketDraft (encode.rs, the one place a model's JSON is read), admits through PacketStore::admit into <root>/round-<n>/packets.cc, lowers, scores once, and keeps a RoundRecord (weksa.loop_round, no text, no Test; Train rows, Validation only as means and refusal counts). weksa loop open|round. Not here: the spec-edit step, accept rule and composite, noise floor, Score and lowering documents, a subprocess adapter, the checkpoint door, a model lowerer (follow-ups golf-loop-spec-edit, golf-loop-accept-rule, golf-loop-score-documents, golf-loop-subprocess-adapter, golf-loop-checkpoint-door-bound, golf-loop-model-lowerer). packet-store-inventory-binding is met for loop packets only (RunStores admits with the run's inventory); validate, packet proof|dao and lower still take caller stores. Waits on 7a, checkpoint-gate, 4b. 7c. golf-loop-edit-accept (spec r1; branch weksa-lowering/golf-loop-edit-accept): one edit-accept cycle. The run works on its own copy of the inventory (create copies the caller's file to <root>/inventory.cc, refusing a root that already holds a file of that name, and pins its digest as RunRecord key 11; RunRecord.inventory is provenance; a run.cc from before has no key 11 and does not open); the head is derived from the latest Accepted Decision. weksa loop cycle asks the editor for one inventory edit from Prompt::mint(train, inventory, rev, feedback) (Train blocks, the inventory listing and an EditFeedback line: whether the last edit was kept and the head's Validation composite rounded to LADDER_STEP, the Ladder release of Blum and Hardt 2015), reads it as an EditDraft (encode.rs; ids and revisions are the store's, into_revision), admits it through InventoryStore::admit_revision into <root>/cycle-<k>/inventory.cc, runs a candidate round under it, and decides on Validation against the head's establishing round past a floor (never under one LADDER_STEP; from cut cycle-floor-paired the paired floor over rounds 1 to FLOOR_ARMS_MIN + 1, so a cycle needs four recorded rounds, where 7c measured the spread between rounds 1 and 2). An edit that fails is a rejected Decision, not an error. RoundRecord gains keys 16 (inventory digest) and 17 (cycle); an encode answer that is not UTF-8 is that item's parse refusal. Not here: Score and lowering documents (golf-loop-score-docs), Train rows in the edit prompt (golf-loop-edit-train-feedback), a paired floor (golf-loop-paired-floor), the checkpoint door and merging to data/ (golf-loop-checkpoint-door-bound, golf-loop-freeze-merge), a model lowerer (landed as 7d: until then an entry an edit added had no lowering row). Waits on 7b. 7d. golf-loop-model-lowerer (spec r1; branch weksa-lowering/golf-loop-model-lowerer): the round lowers through the ModelPort. Prompt::lower (context.rs) takes a LoweringInput (no text, key, group or speaker getter), the item's speaker id, the inventory at the view's revision and the item language's entity forms, and prints LOWER_INSTRUCTIONS, the meaning of the packet, the entities line and the inventory listing; the item key goes only to item_keys, and the speaker id never prints: a referent citing the item's speaker holds the fixed token entity:speaker (SPEAKER_TOKEN, a constant the same for every speaker and item, never derived from an id) and the entities map holds the speaker's forms under that token. The promise guards the id that names an entity of this item; the speaker's forms and the role token do print, because forms are shared by every item of the language (follow-up speaker-alias-token-in-lowering-prompt, which restored the lowerer's ability to name its speaker after #84 removed it). A speaker the language has no form for, or the port lacks, does not refuse: its referents print no entity and the map has no entry. A referent citing the real id entity:speaker is refused (LowerError::SpeakerTokenTaken); the entity store does not reserve it. The answer is {"text"}, read as LoweringDraft in encode.rs, held in score_round's locals and scored once; it is never put in run.cc, a RoundRecord, a Decision or EditFeedback. A round has two answer phases (every encode prompt, then every lowering prompt), all or nothing: while any prompt is unanswered packets.cc is discarded and nothing reaches run.cc. lower::lower and the template table leave the loop (they remain weksa lower's; follow-up table-retirement), so an entry an edit adds can be said and a cycle can buy a gain. One source pin remains: PROMPT_SOURCE_DIGEST covers context.rs, split.rs and lower.rs (RunRecord key 7 and RoundRecord key 3 are retired). The run's record pins LowererReach (weksa loop open --lowerer-reach briefed-subagent|tool-less-relay, RunRecord key 12) and every printed round and cycle line says it; a briefed-subagent run says its numbers are smoke-round only (rulings lowerer-call-reach-toolless-relay, dao-round-seed-inventory). A lowering is refused at Lower when blank, over MAX_LOWERING_BYTES (4096) or holding a control character other than a line break or tab; PROMPT_SOURCE_DIGEST also covers packet.rs. Not here: per-item lowering text documents (golf-loop-score-docs), the subprocess adapter. Waits on 7c.
7. golf-loop-en (7b, 7c and 7d landed the round, ModelPort with AnswerFiles, encoder and lowering prompts, template digest, round records and the edit-accept cycle; the rest remains): the packet store bound to one inventory store everywhere (packet-store-inventory-binding), Score documents (golf-loop-score-docs, golf-loop-score-documents), the accept rule (golf-loop-accept-rule), the checkpoint door and freeze merge (golf-loop-checkpoint-door-bound, golf-loop-freeze-merge). No subprocess adapter and no detached unit: golf-loop-in-claude and encoder-isolation-subagent-per-prompt rule them out. Waits on 3, 5 and 6.
8. style-axis: a style-embedding score behind the scorer port. Waits on 3.
9. ja-panel: ja-JP into the loop's panel, in two cuts on e9cbd9b. 9a ja-panel-loop (spec r1): language is the item's and the panel the run's: Prompt::lower takes the item's language and prints it on a language line (no prompt names English), entity forms are read in that language, RunRecord pins the panel (weksa loop open --panel), offers, mint and recall read only panel items, round means, floors and readings are per panel language, the accept rule runs decide per language, the recall split works on unspaced scripts, and weksa corpus add has no default language. 9b ja-panel-corpus (spec r1, after 9a): the ja-JP slice of data/original, model-written Japanese Train and Validation passages and eight Aozora Bunko public-domain Test sentences with full provenance, and their entities. Cut 5 is not a prerequisite: neither language lowers with ontology entries until cut 5 gives both theirs.
10. conlang-lexicon: the first conlang's lexicon placed at concept embeddings, its morphology data, its ontology entries and the gloss reader. Waits on 3, 5 and question first-conlang.
11. convergence-freeze: the convergence test, Checkpoint's one door (built from the convergence window over the persisted round records), the freeze, the re-encode into training pairs. Waits on 7, 9 and 10.
12. distil, three cuts. 12a distil-slice (relower and the prompted baseline; spec r1). 12b distil-export: Train-only prompt/completion JSONL at a frozen revision, Test-text guarded (follow-up distil-export-at-freeze); waits on 11. 12c distil-train: waits on 12b and the host question (follow-up distil-host-question-owed).
13. live-lowering: agents emit packets with stance; Weksa reads speaker state from persona state (follow-up persona-state-speech-community) and lowers for the reader. Waits on 11.