Reconcile Phi's completed corpus and closed grammar boundary

project/deferred_questions.md

@@ -1,6 +1,6 @@
 # Deferred questions and parked work
 
-This ledger contains the questions and work that survived maintainer review on 2026-07-14. Each entry records where the matter stands and what would bring it back into active work. It is a memory aid, not a source of language law. [`canon.md`](../canon.md) and the [status roadmap](roadmap.md) stay authoritative until a new decision changes them.
+This ledger contains only questions and work that remain deliberately open or parked after maintainer review. Each entry records where the matter stands and what would bring it back into active work. Completed work belongs in the [status roadmap](roadmap.md) or [development log](development_log.md), not here. This file is a memory aid, not a source of language law; [`canon.md`](../canon.md) remains authoritative until a new decision changes it.
 
 ## Design and language questions
 
@@ -22,43 +22,6 @@ The effort will inventory existing body, birth, consent, and Medical vocabulary,
 
 ## Parked corpus and tooling work
 
-### Shelf visual identities
-
-Status: Complete. The primer, text, manual, and pamphlet shelves each have a renderer-led identity.
-
-The website should give each shelf a presentation suited to its work instead of extending the book's magazine treatment everywhere. A common frame keeps the shelves recognisably Phi: the site's colour, type, navigation, and accessibility behaviour do not change with the genre. Presentation belongs in the renderer and maintained site assets; the Markdown remains clean unless a necessary distinction cannot be inferred safely.
-
-Current Phi words and passages set in backticks inside English paragraphs keep the inline treatment established by the book: their type follows the surrounding prose at a slightly smaller proportion, and a quiet raised background makes the change of language visible without interrupting the line. The renderer applies that treatment only when every token is current canonical Phi. Paths, labels, retired forms, and experimental forms remain ordinary code.
-
-Work proceeds in this order:
-
-1. **Primer:** Complete. The full shelf is a quiet illustrated reader shaped around the household story, with a restrained motif for each of its four parts. Part and chapter progress remain easy to see without becoming a scoreboard. Phi passages hold the centre of each scene, compact word ledgers and speaker labels support them, and the manual pointer closes as a note. The pronunciation prelude, contents ladder, and capstone keep distinct layouts within the same family.
-2. **Texts:** Complete. The translation pages use one anthology design with individual masthead motifs, Tengwar titles, exact interlinear source rows, continuous Phi readings, and translation limits. D079 places the six formerly paired works in this form and removes their comparison ledgers; D080 gives the three Little Prince excerpts the same source-led structure; D081 brings *The Velveteen Rabbit* into it while retaining the story's eighteen-scene map. *The Prophet* collection gives each translated teaching its own page under a shared masthead and selection map. The Ring refusal examines Tolkien's source couplet before Phi answers it. Original Phi work keeps Phi in the leading line and follows it with proposition or vocabulary records. The contents page presents translations, the refusal, and original works while keeping *News from Nowhere* apart as one book. Its own contents page, chapter maps, navigation, and strict source-shape record remain separate from the shorter shelf.
-3. **Manual:** Complete. The website opens the manual as a catalogue rather than a loose table of contents. A jump strip moves among Parts I through VII and the closing matter; 23 chapter groups carry 139 numbered readings in the same order as the footer navigation. Each part keeps its established mark. The final cross-read gave every teaching page its proper part and chapter name, replaced the stale repository README and over-detailed outline, and made the build compare that outline with the renderer metadata. Browser checks cover the catalogue and every reading at desktop and phone widths, with dark and print views.
-4. **Pamphlets:** Fourteen workbooks complete. *The Three Slots* established the practice-book design from its opening through the appendix. Adverbial clauses, complementizers, coordination, and relative clauses form the clause family, with a bracket-and-join mark suited to clauses that open, close, modify, and meet. Tense and aspect and evidentiality keep that workbook frame without borrowing the same badge merely because both live in Slot 1. Time receives a returning clock beside a focus mark; evidentiality receives the route a claim took beside the eye that met it. Pronouns and topic-drop and nouns and classifiers form the reference-and-noun-phrase family. People and a recoverable absence mark the first; kinds of thing and an ordered branch mark the second. Naming and source material make the boundary pair without becoming the same lesson. A tag meets a person in the Naming mark; a document meets two separate panels in Source Material. Naming opens through its cast and household scenes, then closes with a six-part answer key. Source Material stays compact: five short lessons need no invented subsections, exact artifacts remain ordinary code, answers receive a reasoning form, and the quick reference begins with its table rather than a false introduction. Spoken Punctuation and Tengwar Mode meet at the period, then turn in different directions. Their marks make that fork visible: an audible line sits beside the dot that stays on the page, while a pen faces a reading frame. Spoken Punctuation keeps its named contrasts, dictation scene, six-part exercise, and two-table appendix. Tengwar's seven compact lessons remain unsectioned where the source is unsectioned. Every fenced example puts Tengwar above romanized Phi and its gloss; a translation stays at the foot when the source supplies one. The exercise page recovers its six divisions, and the final page is a numbered copybook of twenty-two lines. Ternary Numerals closes the workbook sequence with an abacus beside an ascending scale. The child's first eight numbers become a cover-and-check grid, while the higher forms become a numbered recital. Worked decompositions sit apart from the running lesson. The market exchange keeps its spoken dash within an interlinear figure; falling back to raw code is now a build failure. Ten answer groups make the long key navigable, and the five-table appendix still works as a quick reference on a 390-pixel phone. The older sources keep their own manners. The complementizer files retain plain page titles and deeper headings, while the evidentiality contents page keeps its colons beside page headings that use parentheses. The renderer accepts that precise punctuation difference and no broader drift. Interlinear glosses receive their own line, wide structural diagrams scroll inside the example that owns them, and consecutive answers that Markdown swallowed into one item regain their individual markers. The strict manifest pins all 154 pamphlet pages and their continued exercise numbering. Every page was loaded at desktop and phone widths, then once without the stylesheet: 462 renderings. The day and night palettes both held, as did print. Direct inspection covered the final opening and both numeral drills. Arithmetic and Market Day received their own checks. So did the error repairs, answer key, and five-table appendix. The shelf catalogue completes the family. It lists fourteen workbooks in the maintained order and places each reading total beside its mark. The layout is a two-column practice desk on wide screens and one column at phone width; without CSS, it remains a plain ordered list. No Markdown source changed anywhere in the visual sequence.
-
-Each pass begins with one representative page. Browser checks cover desktop, mobile, light, dark, and print behaviour before the treatment reaches the whole shelf. Contents and catalogue pages may need their own layout within the same visual family.
-
-### Entrance and index pages
-
-Status: Complete. Kia, the Short Road, and the Phi Book contents page each have their own renderer-led entrance treatment.
-
-With the shelves styled, attention has moved to the pages that lead into them. Kia stays light and introductory instead of compressing the manual. Its greeting sits inside an open doorway, and the Tengwar hand appears at arrival and farewell. Six brief encounters lead to seven destinations; each door is one large link rather than a small word to aim at. Inline Phi keeps the book's quiet treatment. The renderer checks Kia's expected headings and destinations so a source change cannot silently leave the design behind, while `kia.md` stays untouched.
-
-The Short Road now keeps walking instead of becoming another catalogue. Its map shows all twelve stops at a glance, then a single waypoint line carries the reader down the page. Thirteen examples have room to read as interlinear figures, inline Phi keeps the established quiet background, and the last invitation appears in Tengwar as well as Romanized Phi. The renderer checks the source structure before it builds. Responsive, print, and stylesheet-free layouts keep the route legible, and `short_road.md` stays untouched.
-
-The Phi Book contents page completes the sequence as a map of an argument rather than a spare list. Its cover opens beside the Thames, four movements gather chapters 1 through 14, and a boatman at one end answers a door at the other, which is perhaps as much furniture as a contents page ought to need. The bibliography follows as the book's gathered notes. The renderer compares all seventeen links and labels in `book/README.md` with the published files, so a stale contents list now stops the build. The page holds in both site themes, at phone width, in print, and without a stylesheet. Its source Markdown remains untouched.
-
-### Legacy vocabulary prose audit
-
-Status: Complete.
-
-The workflow gives every new or revised entry a separate Humanizer pass and checks it against the [Phi voice guide](../documents/reference/voice_for_models.md). All 1,297 canonical entries now follow the target contract: definitions live in `description`, physical pronunciation in `articulatory_notes`, and usage in structured `examples`. Content entries also carry accurate semantic-domain rationales. Search terms, usage notes, sound symbolism, and direct pillar connections remain optional.
-
-Articulatory notes follow the word through the mouth and describe breath, contact, release, stress, and hiatus accurately. Sound symbolism has a narrower optional role: it may preserve a genuine Phi-specific association, but it cannot turn a syllable into a hidden morpheme or treat a phoneme's meaning as universal.
-
-The completed lexicon audit worked in `vocabulary/`, the canonical source, and regenerated every dependent reference. [`documents/validation/vocabulary_prose_coverage.json`](../documents/validation/vocabulary_prose_coverage.json) records 1,297 target entries with no legacy, partial, or dual remainder. The contextual retrofit checked active Phi against that finished vocabulary and is also complete: the literary shelf, the book, all seven manual parts, the primer, all nine pamphlets, and the current documents are current. Later source-fidelity reviews may still reopen a passage for a real semantic, factual, voice, or corpus problem, as the On Giving review did for `parelu` and `thaweno`.
-
 ### Lexical relations between content words
 
 Status: Parked until curated navigation would justify maintaining the field.
@@ -67,11 +30,11 @@ Revised definitions already cite the nearest Phi neighbour when the distinction
 
 A future schema revision may add an optional `lexical_relations` array. Each item would name a canonical word and its controlled relationship, with a brief note where the boundary needs explanation. A link might record a contrast or place one word beneath a broader term. Validation can confirm that the target exists and apply reciprocal rules only where the relationship requires them.
 
-This would be a curated lexical graph, not a substitute for vector search. Embeddings could later infer looser similarities from definitions and examples; curated links would preserve distinctions that the lexicon states outright. Revisit the field when the lexicon explorer needs related-word navigation or the project chooses to maintain an explicit lexical graph. The active prose migration does not depend on that decision.
+This would be a curated lexical graph, not a substitute for vector search. Embeddings could later infer looser similarities from definitions and examples; curated links would preserve distinctions that the lexicon states outright. Revisit the field when the lexicon explorer needs related-word navigation or the project chooses to maintain an explicit lexical graph. The current schema and completed prose migration do not depend on that decision.
 
 ### Tengwar renderer verification
 
-Status: Parked until Tengwar work resumes or the mode approaches publication-ready status.
+Status: Parked until the maintainer reopens Tengwar glyph tuning.
 
 The renderer covers the full Phi inventory and builds every example in the Tengwar pamphlet. The gap is visual reference evidence for tehta placement. `h` has manual offsets but lacks independent confirmation; `p`, `k`, `w`, and internal `r` still rely on default bounding-box placement without preserved comparisons.
 

project/development_log.md

@@ -133,6 +133,7 @@ Implementation status, dependencies, optional evaluation work, and the next deve
 | D115 | Accepted | Close the completed lexicon prose migration at the schema boundary. | Every lexicon entry must carry `description`, `articulatory_notes`, and structured `examples`. The schema rejects the retired `concept` and `grammatical_notes` properties instead of accepting them as fallbacks. Regression tests prove both requirements and both rejections, while the completed migration ledger remains available as evidence. No vocabulary entry or generated lexicon reference changes. |
 | D116 | Accepted | Verify ordered source reconstruction against an independently stored witness. | A manifest selects each of the six current *News from Nowhere* chapters from the complete Morris source, names the `morris` citation stream, and applies a stated Project Gutenberg line-wrap normalization. The checker proves that 1,054 ordered citations reconstruct 75,243 normalized characters exactly and reports missing, duplicated, reordered, or altered spans directly. A required chapter pattern stops later chapters from bypassing the check, and synthetic regression tests exercise every failure class in CI. No translation layer changes. |
 | D117 | Accepted | Preserve *News from Nowhere* source-to-Phi continuity in one maintained record. | The record fixes seven onym forms for six named people, the source-bound treatment of recurring places, durable social and material renderings, narrator and evidential practice, one-frame continuing speech, and the chapters 1-6 checkpoint. It records rejected renderings and leaves later names, place onyms, and demonstrated lexical pressure open without reserving answers. The file is available during source-to-Phi work and cross-chapter review but is forbidden to fresh Phi-to-English contexts. No translation layer, vocabulary entry, registered compound, or grammar changes. |
+| D118 | Accepted | Treat Phi's grammar and particles as closed except through an explicit defect-led amendment. | Ordinary corpus, translation, and vocabulary work must use the existing grammar and function-word inventory. A proposed amendment requires a demonstrated repeatable language-level defect, failed resolution within current Phi, and an explicit maintainer decision to reopen the relevant canon before implementation. The same state correction marks the three-essay original-corpus milestone complete, keeps spontaneous conversation as separate evidence work, removes incidental live lexicon totals from the roadmap, and leaves only genuinely open or parked work in the deferred ledger. No grammar, particle, vocabulary entry, or corpus text changes. |
 
 ## Completed discretionary lexical migrations