cronli5 0.3.1 → 0.7.2

package/src/lang/pt/notes.md

@@ -0,0 +1,199 @@
+# Português (pt, target pt-BR) — Language Notes
+
+**donor: es.** Derived by sibling-derivation (tooling/docs/language-pipeline.md):
+the Spanish module supplies the structure, plan override, OR-frame, predicates,
+re-strategies, and dialect mechanism; this doc records only where **pt-BR
+diverges** from that donor. The shipped table today is **pt-BR**; pt-PT is a
+future dialect axis (below). The corpus translation (Stage 2) and renderer port
+(Stage 4) both follow this contract. The donor's own contract is
+[`../es/notes.md`](../es/notes.md).
+
+## Anchors
+
+Brazilian norm (VOLP / Academia Brasileira de Letras, plus cronstrue `pt_BR`):
+lowercase month and weekday names, **24-hour zero-padded clock by default**
+("às 09:30", "às 17:00"; `{ampm: true}` opts into the 12-hour clock), day
+periods on the 12-hour clock (madrugada 1–5, manhã 6–11, tarde 12–18, noite
+19–24 — see the note below; this is the one boundary that may differ from es),
+"meio-dia" / "meia-noite" for exact 12:00 / 0:00 (12-hour clock), colon time
+separator.
+
+**Clock — decided: "às 09:00" (24h zero-padded), not the colloquial "às 9h".**
+Rationale: parallels the es donor and cronstrue `pt_BR`'s reference rendering,
+and keeps the corpus comparable field-by-field with es; the "9h"/"9h30" form is
+genuinely common in pt-BR casual writing but is a *separate register*, deferred
+to a future custom/dialect style exactly as es kept `hSuffix` opt-in.
+
+**Article agreement (the a+a contraction):** the preposition *a* + the feminine
+article contracts — *a + a hora 1* → **à 1h / à 01:00** (grave accent, singular),
+*a + as horas* → **às** otherwise. This mirrors es's singular-article hold for
+one o'clock ("a la 1") but in pt the contraction is *written with the accent*:
+"à 01:00" at hour 1, "às 09:00" at every other hour, on **both** clocks. Hours
+zero-pad to two digits on the 24-hour clock ("às 09:00"); the 12-hour clock
+leaves the hour unpadded ("às 9 da manhã").
+
+## Day periods (12-hour)
+
+"da madrugada / da manhã / da tarde / da noite" (contraction *de + a* = *da*).
+**Decided (panel-confirmed): madrugada 1–5, manhã 6–11, tarde 12–18, noite
+19–24.** es uses tarde 12–19 / noche 20–24, but pt-BR puts *noite* earlier;
+the blind pt-BR panel unanimously affirmed the 19h boundary (broadcast/weather
+register and the "jornal da noite" cultural anchor place noite firmly at 19h —
+tighter than the loose popular sense that some extend to 18h, and the better
+choice for an unambiguous description). 18h reads *da tarde*, 19h+ *da noite*.
+"meio-dia" / "meia-noite" for exact 12:00 / 0:00.
+
+## Weekday recurrence
+
+es uses the plural article "los lunes" = every Monday. **Decided (panel-
+confirmed): keep *-feira* throughout** — the full forms are the standard pt-BR
+written/spoken register and dropping *-feira* is too informal for an unambiguous
+description. The *-feira* element attaches to weekdays Mon–Fri (segunda-feira …
+sexta-feira); sábado and domingo have none. The resolved sub-rules:
+
+- **Single weekday recurrence + a clock time → "toda segunda-feira às 9 da
+  manhã".** The plural-article recurrence "às segundas-feiras" before an "às …"
+  time clashed aurally (the double-"às"); the panel's fix is the singular
+  "toda X" head, which reads naturally and keeps the meaning. This applies
+  wherever a single weekday leads a clause that a clock time follows (incl.
+  "toda segunda-feira de junho às 9 da manhã").
+- **Standalone single weekday recurrence (no following time) keeps the plural
+  article** "às segundas-feiras" (e.g. "a cada 15 minutos às segundas-feiras",
+  trailing-qualifier "… às segundas-feiras").
+- **Lists carry the *-feira* suffix on the last *-feira* day only**, the
+  idiomatic pt-BR suffix-ellipsis: "às segundas, quartas e sextas-feiras";
+  "às terças, quintas-feiras, sábados e domingos" (terça bare, quinta is the
+  last *-feira* day so it carries the suffix, sábado/domingo never do). All
+  panels affirmed this is correct and unambiguous, not an inconsistency.
+- **Ranges carry *-feira* on the last term only**: "de segunda a sexta-feira"
+  (the asymmetric form is the idiomatic pt-BR range shorthand — not
+  "de segunda-feira a sexta-feira").
+- **Single weekday in an OR-union arm reads the Brazilian recurrence**
+  "às [weekday]s-feiras" / "aos domingos" (NOT "em qualquer [weekday]", which
+  reads slightly Iberian); a **range** arm keeps the nominal head
+  "em qualquer dia de segunda a sexta-feira" (a range needs the head "dia").
+- **Quartz nth-weekday ordinal collision:** when the ordinal word would collide
+  with the weekday name ("segunda segunda-feira" for `1#2`), use the ordinal
+  digit "na 2ª segunda-feira do mês". Non-colliding ordinals keep the word form
+  ("na última sexta-feira", "a primeira segunda-feira").
+
+## Ordinals / dates
+
+es: "el 1 de junio" / "el día N" / "el N de cada mes". **Decided for pt-BR:
+"(no) dia 1 de junho"** — pt-BR routinely uses the cardinal with the noun *dia*
+("dia 1", "dia 13"), so the donor's "el día N" maps cleanly to "dia N". Ranges:
+"do dia 1 ao dia 15 do mês" (contractions *de+o=do*, *a+o=ao*). The es bare
+"el 1 de junio" → "**dia 1 de junho**".
+
+- **Decided (panel-confirmed): the 1st of the month is the ordinal "dia 1º"**;
+  every other day stays cardinal. The ordinal first is a deep pt-BR norm
+  (calendars, official/legal texts, speech); cardinal "dia 1" reads as a typo or
+  informal shorthand. The "1º" carries into the date-range and OR-union arms
+  too: "do dia 1º ao dia 15", "seja no dia 1º …". (Ranges carry the ordinal on
+  the first term and cardinal on the rest, the normal pt-BR pattern.) The
+  W-operator proximity preposition is the dative "próximo **ao** dia 15" (not
+  "próximo do dia 15") — proximity-to-a-target takes *a+o=ao*.
+- Quartz nth-weekday ordinals: primeiro/primeira, segundo/segunda, terceiro,
+  quarto, quinto — **gendered** (see below).
+
+## Contractions (the big es→pt divergence — renderer logic, not string swaps)
+
+Portuguese fuses prepositions with the following article; the renderer must
+form these wherever es emitted a bare preposition + article:
+
+- *de* + o/a/os/as → **do / da / dos / das** ("do mês", "da manhã", "das 9").
+- *em* + o/a/os/as → **no / na / nos / nas** ("no dia 1", "no minuto 30",
+  "na hora").
+- *a* + a/as → **à / às** (clock and weekday recurrence; grave accent).
+  *a* + o/os → **ao / aos** (date ranges "ao dia 15").
+- *por* generally stays separate in these phrasings (not needed as a fused
+  form for the cron domain; noted for completeness).
+
+This contraction layer is the principal structural divergence from es and is
+where most RED in the TDD port is expected — it is **gender/number-driven
+formation**, not a lexical substitution.
+
+## Connectives
+
+- and → **e** (RAE-style coma ante "y" has **no pt-BR analog** — pt does *not*
+  put a comma before *e* in a simple series; **FLAGGED**: the donor's "coma
+  ante 'y'" re-strategy in the day-period join must be *dropped*, not ported.
+  This is a real renderer divergence, not a string swap.)
+- or → **ou** (the OR-union connector; see re-strategies).
+- range / until → **a** ("de … a …") and **até** where a terminal "until"
+  reads better; default to **a** to mirror es "de … a …".
+
+## Names, gender, agreement
+
+- Lowercase months and weekdays (confirmed pt-BR norm, VOLP).
+- **Gender/agreement the renderer must handle (es→pt divergence):**
+  - Weekdays are **feminine** in pt (a segunda-feira) — the recurrence article
+    is *as* → *às*; es's masculine "los lunes" does not carry over. This drives
+    "às segundas-feiras", "qualquer segunda-feira".
+  - Quartz nth ordinals agree with the (feminine) weekday: "a primeira
+    segunda-feira", "o último domingo" (domingo masculine), "a última
+    sexta-feira". The renderer must select ordinal gender by weekday gender —
+    es used invariant "primer/último". **FLAGGED** as needing real agreement
+    logic.
+  - "todo(s)" / "cada" agreement: "todos os dias" (m.pl.), "cada mês" (m.),
+    "cada hora" (f.) — gendered determiners where es had "todos los días" /
+    "cada".
+
+## Ported re-strategies (language-neutral; pt forms)
+
+- **Per-hour windows for wildcard minutes over hour lists** (es §"wildcard
+  minutes over hour lists render as per-hour windows"): keep the strategy; pt
+  form "das 9 às 9:59 da manhã" (note *das* = de+as, *às* = a+as).
+- **OR-union unified frame:** es "ya sea X o Y" → **"seja X ou Y"
+  (panel-confirmed).** All three personas read it as an unambiguous inclusive
+  OR; the "seja" frame is cleaner than a bare "X ou Y" and there is no
+  intersection misreading. "ou seja" is avoided — it means "that is/i.e." The
+  shared month is fronted once and the arms are month-less, exactly as in es.
+  The weekday arm wording is resolved under *Weekday recurrence* above (single
+  weekday → "às [weekday]s-feiras"; range → "em qualquer dia de segunda a
+  sexta-feira").
+- **No-fold month range:** a month range never folds into another phrase
+  ("dia 1 de junho a setembro" parses as "(dia 1 de junho) a setembro"); dates
+  scope it instead ("dia 1 de cada mês, de junho a setembro"); mixed lists
+  repeat the preposition per piece ("em janeiro e de março a junho"). Same rule
+  as es and English.
+- **Step-flattening:** step segments inside lists always flatten into their
+  fires — months, weekdays, dates, minutes, seconds — no raw step token reaches
+  the output. Identical to es.
+- **Anchored minutes/seconds** read as "no minuto 30 de cada hora" (em+o=no),
+  the donor's "en el minuto 30 de cada hora" — not a calque of "past the hour".
+
+## Dialect axis (future)
+
+pt-PT is a **future dialect** (clock/lexical divergences from pt-BR, e.g. some
+date/register differences), mirroring es's es-ES / es-419 split. **One `pt`
+table today = pt-BR.** A future `pt-PT` (and any regional pt-BR style such as a
+"9h" colloquial-clock custom field) would clear its own native panel before
+shipping, per the dialect rules in the pipeline.
+
+## Residuals inherited from es (NOT fixed here — es+pt follow-up)
+
+The blind pt-BR panel's technical reviewer flagged two issues that are **shared
+artifacts of the es donor corpus**, not pt regressions, so they were left in the
+pt corpus to keep it field-comparable with es and are tracked as a joint es+pt
+follow-up (docs/backlog.md, per-language follow-ups):
+
+- **Hour-window overlap in `* 2/4,18-20 * * *`.** Hour 18 is named twice — once
+  as the 2/4 step arm's per-hour window ("das 6 às 6:59 da tarde") and again as
+  the left endpoint of the 18-20 range window ("das 6 da tarde às 8:59 da
+  noite"). The fire set is correct (no value dropped or understated); the
+  overlap is a rendering-clarity artifact present identically in es.
+- **OR DOW-arm "e" bracketing** in `… ou de segunda a sexta-feira e aos
+  domingos` (`0 0 1 * 0,1-5`, `0 0 1 6-9 0,1-5`). The internal "e" joining
+  Mon–Fri + Sun inside the second OR arm could be misparsed as a top-level
+  conjunction. The meaning is correct and the construction is the same one the
+  es donor uses ("o de lunes a viernes y los domingos"); fixing it is an es+pt
+  bracketing change, not a pt-only one.
+
+## Known trade-offs
+
+- `short` only switches spelled numbers to digits; pt name abbreviations
+  (seg., qua.) are not yet implemented (same residue as es).
+- The grave-accent contraction (à/às) is correctness-critical for the 1-o'clock
+  and weekday-recurrence forms; the renderer forms it programmatically rather
+  than hard-coding strings.

package/src/lang/pt/status.json

@@ -0,0 +1,8 @@
+{
+  "name": "Portuguese",
+  "status": "beta",
+  "humanReview": null,
+  "modelReview": "sibling-derived from es (the validated Romance sibling): the es renderer's structure was ported (plan override, OR-union frame, parity predicates, re-strategies, dialect scaffold) and its lexicon translated to pt-BR, then TDD'd to green against the reviewed pt-BR corpus (277 entries). The corpus was a candidate translated from the reviewed es corpus and finalized by a blind 3-persona pt-BR Sonnet panel (everyday / copy-editor / technical) before the port (corpus -> review -> port; tooling/docs/language-pipeline.md). The es->pt divergences the panel and TDD surfaced and the renderer now handles: preposition+article contraction (do/da/no/na/à/às/ao/aos, formed gender/number-driven, not as strings), gender agreement (feminine -feira weekdays vs masculine domingo/sábado, gendered Quartz ordinals and the 'duas horas' feminine cardinal), the 'toda X' single-feminine-weekday recurrence head (kills the double-'às'), the suffix-ellipsis weekday list ('às segundas, quartas e sextas-feiras'), 'na 2ª segunda-feira' (ordinal digit on the weekday-name collision), the dropped comma before 'e', and the ordinal 'dia 1º' for the 1st of the month. Objective gates (round-trip, fuzz dropped-value detector, both-side OR-scope, cRonstrue pt_BR reference) plus the panel gate beta.",
+  "note": "BETA — model-validated: sibling-derived from es (a proven structure/style anchor), TDD-green over a blind-panel-reviewed pt-BR corpus, and clean on the corpus-independent mechanical gates (fuzz, round-trip, OR-scope, cRonstrue pt_BR). Two residuals are inherited unchanged from the es donor and are NOT pt regressions (tracked as a joint es+pt follow-up; notes.md §Residuals): (1) the hour-window overlap in '* 2/4,18-20 * * *' (hour 18 named twice — the rendering-clarity artifact, no value dropped); (2) the OR DOW-arm 'e' bracketing in '… ou de segunda a sexta-feira e aos domingos'. The 'short' style switches spelled numbers to digits but pt name abbreviations are not yet implemented (same residue as es). Graduates to stable only on fluent-pt human review. pt-PT is a future dialect axis (notes.md §Dialect axis); no regional dialect ships yet.",
+  "dialects": {}
+}

package/src/lang/zh/index.ts

@@ -25,6 +25,40 @@ type StepSegment = Extract<Segment, {kind: 'step'}>;
 const UNITS = {hour: '小时', minute: '分钟', second: '秒'};
 const WEEKDAYS = ['周日', '周一', '周二', '周三', '周四', '周五', '周六'];
 
+// Simplified → Traditional (zh-Hant) Han glyph map. Schedule prose differs
+// between the two scripts only by character form — within this domain every
+// Simplified glyph that has a Traditional form maps 1:1 with no context
+// sensitivity — so the Traditional variant is the reviewed Simplified output
+// with this map applied at the render boundary, NOT a second word table that
+// would duplicate the renderer's logic. The Taiwan-standard form is chosen for
+// each glyph (週 for week, 點/時/鐘/個/數/單/雙/後/間/從/內); 啟 (not the 啓
+// variant) for 啟動. Two whole-word choices are kept faithful to the 1:1 map
+// and flagged for native review in notes.md: 運行時間 (a Taiwan-native may say
+// 執行時間) and 表達式 (Taiwan tech register may prefer 運算式 / 表示式).
+const HANT: {[glyph: string]: string} = {
+  个: '個', 从: '從', 内: '內', 别: '別', 动: '動', 单: '單', 双: '雙',
+  后: '後', 启: '啟', 周: '週', 数: '數', 无: '無', 时: '時', 点: '點',
+  统: '統', 识: '識', 达: '達', 运: '運', 钟: '鐘', 间: '間'
+};
+
+// Apply the Traditional glyph map to a finished Simplified string. The default
+// Simplified (zh / zh-Hans) variant returns the input untouched, so its output
+// is byte-identical to before this variant existed.
+function toVariant(text: string, variant: ChineseStyle['variant']): string {
+  if (variant !== 'Hant') {
+    return text;
+  }
+
+  return Array.from(text, (glyph) => HANT[glyph] ?? glyph).join('');
+}
+
+// The variant the most recent `options()` call resolved. `cronli5()` always
+// calls `options()` before reading `reboot`/`fallback` or invoking `sentence`,
+// none of which receive `opts`; this lets those contract-fixed members honor
+// the dialect without changing the shared Language contract. The library is
+// synchronous and single render per call, so a module-private latch is safe.
+let activeVariant: ChineseStyle['variant'] = 'Hans';
+
 // "A、B和C" — enumerate with 、 and join the final item with 和.
 function joinAnd(items: string[]): string {
   if (items.length < 2) {
@@ -67,7 +101,7 @@ function renderStride(stride: Stride): string {
   const {interval, start, last, cycle, unit, mark, anchor} = stride;
   const lead = anchor + '从' + start + mark + '起' + cadence(interval, unit);
 
-  return chooseStride({start, interval, cycle}, {
+  return chooseStride({start, interval, last, cycle}, {
     bare: () => cadence(interval, unit),
     offset: () => lead,
     bounded: () => lead + '，至' + last + mark
@@ -1041,8 +1075,26 @@ function render(schedule: Schedule, plan: PlanNode, opts: Opts): string {
 
 // --- Day-level qualifier (date / month / weekday / year). ---
 
-// The month phrase: "" (wildcard), "每个奇数月"/"每个偶数月" (step ×2),
+// Whether the month is a BOUNDED parity step ("2-10/2") — an interval-2 step
+// that does NOT span the open parity set. It enumerates as a list of singles
+// ("2、4、6、8、10月"), so it takes the multi-month comma like an explicit list,
+// unlike a single month or a non-parity bounded step ("3-11/3", glued).
+function boundedParityMonth(schedule: Schedule): boolean {
+  if (schedule.shapes.month !== 'step' ||
+      isOpenStep(schedule.pattern.month)) {
+    return false;
+  }
+
+  const segs = segmentsOf(schedule, 'month');
+
+  return segs.length === 1 && segs[0].kind === 'step' && segs[0].interval === 2;
+}
+
+// The month phrase: "" (wildcard), "每个奇数月"/"每个偶数月" (OPEN step ×2),
 // "1月至3月" (range), else the enumerated numbers sharing one 月 ("1、4、7、10月").
+// A BOUNDED parity step ("2-10/2" = months 2,4,6,8,10) fires a finite set, so
+// it enumerates through the number path below rather than the open parity class
+// — the "每个偶数月" wording asserts December too, which the bound excludes.
 function monthPhrase(schedule: Schedule): string {
   if (schedule.pattern.month === '*') {
     return '';
@@ -1051,7 +1103,8 @@ function monthPhrase(schedule: Schedule): string {
   const segs = segmentsOf(schedule, 'month');
   const first = segs[0];
 
-  if (segs.length === 1 && first.kind === 'step' && first.interval === 2) {
+  if (segs.length === 1 && first.kind === 'step' && first.interval === 2 &&
+      isOpenStep(schedule.pattern.month)) {
     return '每个' + (first.fires[0] % 2 ? '奇' : '偶') + '数月';
   }
 
@@ -1074,13 +1127,14 @@ function monthPhrase(schedule: Schedule): string {
   return nums.join('、') + '月';
 }
 
-// The day-of-month list. A pure list of singles shares one trailing 日
-// ("1、3、8日"); any range gives each segment its own 日 ("1至5日、10日").
+// The day-of-month list. A list of singles — or a bounded step enumerated to
+// its fires (9-17/2 = 9,11,13,15,17) — shares one trailing 日 ("1、3、8日",
+// "9、11、13、15、17日"); any range gives each segment its own 日 ("1至5日、10日").
 function dayList(schedule: Schedule): string {
   const segs = segmentsOf(schedule, 'date');
 
-  if (segs.every((seg) => seg.kind === 'single')) {
-    return segs.map((seg) => (seg as {value: string}).value).join('、') + '日';
+  if (segs.every((seg) => seg.kind === 'single' || seg.kind === 'step')) {
+    return fireValues(segs).join('、') + '日';
   }
 
   return segs.map(function day(seg) {
@@ -1088,7 +1142,9 @@ function dayList(schedule: Schedule): string {
       return seg.bounds[0] + '日至' + seg.bounds[1] + '日';
     }
 
-    return (seg as {value: string}).value + '日';
+    return seg.kind === 'step' ?
+      seg.fires.join('、') + '日' :
+      (seg as {value: string}).value + '日';
   }).join('、');
 }
 
@@ -1152,7 +1208,11 @@ function datePhrase(schedule: Schedule): string {
     return quartzDate(date, month || '本月');
   }
 
-  if (schedule.shapes.date === 'step') {
+  // An OPEN day step ("*/N") is a frequency — the bare "每N天" cadence. A
+  // BOUNDED step ("a-b/N") fires a finite set of days, so it enumerates them
+  // through the day-list path below, never the cadence (which would drop the
+  // bounds).
+  if (schedule.shapes.date === 'step' && isOpenStep(date)) {
     return month + cadence(stepSegment(schedule, 'date').interval, '天');
   }
 
@@ -1160,12 +1220,13 @@ function datePhrase(schedule: Schedule): string {
     return '每月' + dayList(schedule);
   }
 
-  // A multi-month scope (range/list) ends in 月 and would run straight into the
+  // A multi-month scope (range/list, or a bounded parity step that enumerates
+  // like a list — "2、4、6、8、10月") ends in 月 and would run straight into the
   // day — "6月至8月1日" reads "8月1日" as August 1st. The comma keeps the month
   // scope distinct from the day ("6月至8月，1日"). A single month stays glued
   // ("6月1日"), which is unambiguous.
   const monthMulti = schedule.shapes.month === 'range' ||
-    schedule.shapes.month === 'list';
+    schedule.shapes.month === 'list' || boundedParityMonth(schedule);
 
   return month + (monthMulti ? '，' : '') + dayList(schedule);
 }
@@ -1178,7 +1239,9 @@ function dateCore(schedule: Schedule, quartzPrefix: string): string {
     return quartzDate(schedule.pattern.date, quartzPrefix);
   }
 
-  if (schedule.shapes.date === 'step') {
+  // An open day step is the bare "每N天" cadence; a bounded step enumerates its
+  // days through dayList (see datePhrase), so the bounds are not dropped.
+  if (schedule.shapes.date === 'step' && isOpenStep(schedule.pattern.date)) {
     return cadence(stepSegment(schedule, 'date').interval, '天');
   }
 
@@ -1318,7 +1381,12 @@ function composePoint(schedule: Schedule, core: string): string {
 
   const dateSet = isSet(schedule.pattern.date);
   const weekdaySet = isSet(schedule.pattern.weekday);
-  const comma = dateSet && weekdaySet || schedule.shapes.date === 'step';
+  // The comma separates an OR union or the open "每N天" cadence from the core. A
+  // bounded date step renders as a glued day list ("每月9、11…日"), not a
+  // cadence, so it takes no comma — only an open step does.
+  const dateCadence = schedule.shapes.date === 'step' &&
+    isOpenStep(schedule.pattern.date);
+  const comma = dateSet && weekdaySet || dateCadence;
 
   return qual + (comma ? '，' : '') + core;
 }
@@ -1371,6 +1439,13 @@ function hourCadenceApplies(schedule: Schedule): boolean {
 }
 
 function describe(schedule: Schedule, opts: Opts): string {
+  return toVariant(describeHans(schedule, opts), opts.style.variant);
+}
+
+// The Simplified rendering of a schedule; `describe` maps it to the active
+// variant. The body owns every Simplified glyph; the variant is applied once,
+// at the boundary, so no emit point has to know which script it is writing.
+function describeHans(schedule: Schedule, opts: Opts): string {
   const {kind} = schedule.plan;
   const core = render(schedule, schedule.plan, opts);
   let composed = core;
@@ -1419,22 +1494,36 @@ function describe(schedule: Schedule, opts: Opts): string {
 function normalizeOptions(options?: Cronli5Options): Opts {
   options = options || {};
 
+  const style = resolveDialect(options.dialect);
+
+  // `cronli5()` reads `reboot`/`fallback` and calls `sentence` without `opts`;
+  // latch the variant here (always called first) so they can honor the dialect.
+  activeVariant = style.variant;
+
   return {
     ampm: typeof options.ampm === 'boolean' ? options.ampm : false,
     lenient: !!options.lenient,
     seconds: !!options.seconds,
     short: !!options.short,
-    style: resolveDialect(options.dialect),
+    style,
     years: !!options.years
   };
 }
 
 const zh: Language<ChineseStyle> = {
   describe,
-  fallback: '无法识别的 cron 表达式',
+  // `reboot`/`fallback` are contract-fixed strings the core reads without
+  // `opts`; getters honor the variant `options()` latched, keeping the shared
+  // Language contract unchanged while the Traditional dialect still applies.
+  get fallback(): string {
+    return toVariant('无法识别的 cron 表达式', activeVariant);
+  },
   options: normalizeOptions,
-  reboot: '系统启动时',
-  sentence: (description) => '运行时间：' + description + '。'
+  get reboot(): string {
+    return toVariant('系统启动时', activeVariant);
+  },
+  sentence: (description) =>
+    toVariant('运行时间：', activeVariant) + description + '。'
 };
 
 export default zh;

package/src/lang/zh/notes.md

@@ -22,9 +22,21 @@ conventions (this doc) → coverage-spec pattern set → panel-validated
   for duration vs 分 for clock position (never swapped); **每天** not 每日;
   suppress the numeral 1 (每分钟, not 每1分钟); no 第 before day/month numbers;
   no redundant 每.
-- **Simplified (zh-Hans) is the default; Traditional (zh-Hant) is a separate
-  top-level locale**, not a sub-dialect — a distinct vocab/surface table
-  (週/號 swaps), per the copy-editor.
+- **Simplified (zh-Hans) is the default; Traditional (zh-Hant) is a within-zh
+  variant** selected by the `dialect` option, NOT a separate top-level locale.
+  A separate `zh-Hant` module would have to duplicate or import zh's assembly
+  logic, both forbidden (a language never imports another language — see
+  docs/i18n-design.md; only the core is shared). Within this domain the two
+  scripts differ only by character form (no register-level grammar split), so
+  the variant is the reviewed Simplified output with a 1:1 Han glyph map applied
+  at the render boundary (`toVariant` in `index.ts`:
+  時/鐘/點/週/個/數/單/雙/後/間/從/內/無/識/別/啟/統/達/運). It ships
+  **experimental** — a model-drafted glyph/register mapping, not yet validated
+  by a Traditional-native or blind Hant panel (the same gate that graduates zh).
+  Two whole-word choices are flagged for native review: `運行時間` (a
+  Taiwan-native may say `執行時間`) and `表達式` (Taiwan tech register may prefer
+  `運算式` / `表示式`); both are widely-accepted, and the faithful 1:1 map is
+  kept so the variant stays a pure transliteration of the reviewed Hans oracle.
 - **Confinement uses a frame, never juxtaposed cadences:** 在9点至17点之间，
   每15分钟 — the 在…之间 frame binds the cadence to the window (the same
   confinement-vs-juxtaposition rule as the other languages).
@@ -95,7 +107,7 @@ every field value preserved — `npm run fuzz zh` is clean):
 |---|---|---|---|
 | `numerals` | `'arabic'` / `'chinese'` | arabic | 9点 vs 九点 |
 | `clock` | `'24h'` / `'12h'` | 24h | 14点 vs 下午2点 (maps to today's `ampm`) |
-| `locale` | `'zh-Hans'` / `'zh-Hant'` | zh-Hans | Simplified vs Traditional vocab table |
+| `dialect` | `'zh-Hans'` / `'zh-Hant'` | zh-Hans | Simplified vs Traditional Han glyph form (within-zh variant; experimental) |
 | `quarterHour` | bool | false | enable 半 / 一刻 / 三刻 (default is explicit 分) |
 | `useHao` | bool | false | 号 vs 日 |
 

package/src/lang/zh/status.json

@@ -3,5 +3,14 @@
   "status": "beta",
   "humanReview": null,
   "modelReview": "blind 3-persona Sonnet style panel + author/audit corpus workflow (2026-06-20); npm run fuzz zh clean (0 throws / degenerate / missing-value)",
-  "note": "BETA — model-validated by a blind 3-persona Sonnet panel (src/lang/zh/notes.md), corpus authored via an author/audit/fix workflow and converged to the renderer's canonical forms. Graduates to stable only on fluent-human review."
+  "note": "BETA — model-validated by a blind 3-persona Sonnet panel (src/lang/zh/notes.md), corpus authored via an author/audit/fix workflow and converged to the renderer's canonical forms. Graduates to stable only on fluent-human review.",
+  "variants": {
+    "zh-Hant": {
+      "name": "Chinese (Mandarin, Traditional)",
+      "status": "experimental",
+      "humanReview": null,
+      "modelReview": null,
+      "note": "EXPERIMENTAL — a within-zh variant (the reviewed Simplified output with a 1:1 Han glyph map applied at the render boundary, selected by {dialect: 'zh-Hant'}), NOT a separate locale. Model-drafted glyph/register mapping, not yet validated by a Traditional-native or blind Hant panel. Graduates to beta on a Traditional-native / blind Hant review (the same gate model as Simplified zh). Whole-word choices flagged for native review: 運行時間 (vs 執行時間), 表達式 (vs 運算式 / 表示式)."
+    }
+  }
 }

package/types/core/cadence.d.ts

@@ -22,6 +22,7 @@ interface StrideParts {
 declare function renderStride(spec: {
     start: number;
     interval: number;
+    last: number;
     cycle: number;
 }, parts: StrideParts): string;
 declare function hourListStride(values: number[]): {

package/types/lang/fr/dialects.d.ts

@@ -0,0 +1,11 @@
+import type { Cronli5Options } from '../../types.js';
+/**
+ * French's own resolved style shape: a separator and the spacing of the `h`
+ * clock mark. fr is 24-hour only, so there is no `ampm`/`meridiem` axis.
+ */
+export interface FrenchStyle {
+    sep: string;
+    unspaced: boolean;
+}
+declare function resolveDialect(dialect: Cronli5Options['dialect']): FrenchStyle;
+export { resolveDialect };

package/types/lang/fr/index.d.ts

@@ -0,0 +1,4 @@
+import type { Language } from '../../core/schedule.js';
+import { type FrenchStyle } from './dialects.js';
+declare const fr: Language<FrenchStyle>;
+export default fr;

package/types/lang/pt/dialects.d.ts

@@ -0,0 +1,13 @@
+import type { Cronli5Options } from '../../types.js';
+/**
+ * Portuguese's own resolved style shape has a separator,
+ * clock default, meridiem form, and `h` suffix.
+ */
+export interface PortugueseStyle {
+    ampm: boolean;
+    hSuffix: boolean;
+    meridiem: 'descriptors' | 'english';
+    sep: string;
+}
+declare function resolveDialect(dialect: Cronli5Options['dialect']): PortugueseStyle;
+export { resolveDialect };

package/types/lang/pt/index.d.ts

@@ -0,0 +1,4 @@
+import type { Language } from '../../core/schedule.js';
+import { type PortugueseStyle } from './dialects.js';
+declare const pt: Language<PortugueseStyle>;
+export default pt;