docs(specs,plans): amend #230 — drop we/th aliases, collision risk

#230

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: pajoma

docs/plans/2026-06-01-fix-230-smart-input-week-and-weekday-aliases.md

@@ -15,7 +15,7 @@ Both bugs are addressed in the #177 tokenizer branch. Bug 1 is already resolved
 
 1. **Verify Bug 1 is resolved** — add regression test: `parseInput("task week Finalize Shoppinglist")` → `flags="task"`, `text="Finalize Shoppinglist"`, `hasWeek()=true`. No code change needed.
 
-2. **Add 2-letter English aliases to `weekdayVocab()`** — append `'mo', 'tu', 'we', 'th', 'fr'` after the 3-letter English block (`mon/tue/wed/thu/fri/sat/sun`). Longest-first ordering preserved (3-letter tried before 2-letter). `fr` already present in German section; adding it to English section makes intent explicit and matches first (same result from `getDayOfWeekForString`).
+2. **Add 2-letter English aliases to `weekdayVocab()`** — append `'mo', 'tu', 'fr'` after the 3-letter English block (`mon/tue/wed/thu/fri/sat/sun`). `we` and `th` excluded: both are high-frequency English words that would destructively consume text. `fr` already present in German section; adding it to English section makes intent explicit (same result from `getDayOfWeekForString`). Longest-first ordering preserved.
 
 3. **Invalidate pattern cache** — `_weekdayPatterns` is built lazily from `weekdayVocab()`. Adding entries to the array changes the return value; since the cache is built per `MatchInput` instance, no reset needed.
 
@@ -33,9 +33,8 @@ Both bugs are addressed in the #177 tokenizer branch. Bug 1 is already resolved
 | Bug 1 regression — w15 still numeric | `"w15"` | week=15 |
 | 2-letter alias mo | `"mo"` | same offset as `"mon"` |
 | 2-letter alias tu | `"tu"` | same offset as `"tue"` |
-| 2-letter alias we | `"we"` | same offset as `"wed"` |
-| 2-letter alias th | `"th"` | same offset as `"thu"` |
 | 2-letter alias fr | `"fr"` | same offset as `"fri"` |
+| No alias we/th | `"we"`, `"th"` | text-only (not weekday aliases — collision risk) |
 | No regression — task do | `"task do the shopping"` | flag=task, text="the shopping" (do = German Thursday) |
 
 ## Dependencies
@@ -46,7 +45,7 @@ Both bugs are addressed in the #177 tokenizer branch. Bug 1 is already resolved
 
 | Risk | Mitigation |
 |------|-----------|
-| `we`/`th` false-positive on common words | Accepted per spec; tokenizer marks them `ambiguous`, yellow border visible to user |
+| `we`/`th` collision with English words | Excluded from aliases — not added |
 | `fr` duplication (English + German section) | `getDayOfWeekForString("fr", locale)` returns 5 (Friday) for all locales — result identical |
 | `_weekdayPatterns` cache stale | Cache is per-instance, built lazily on first `recognizeWeekday` call — no stale state possible |
 

docs/specs/2026-06-01-fix-230-smart-input-week-and-weekday-aliases.md

@@ -11,7 +11,7 @@ Both defects break user-visible output in the current 1.1.0 milestone.  The week
 ## In scope
 
 - Fix `weekPattern` in `MatchInput.getExpression()` so it does not consume any character beyond the week token itself.
-- Add `mo`, `tu`, `we`, `th`, `fr` to `getWeekdayPattern()` as recognized English aliases (longest-first ordering; after the 3-letter set).
+- Add `mo`, `tu`, `fr` to `getWeekdayPattern()` as recognized English aliases (longest-first ordering; after the 3-letter set). `we` and `th` are excluded — both are high-frequency English words that would destructively consume text (e.g. `"task we need to buy milk"` → `we` parsed as Wednesday).
 - Add regression tests in `src/test/suite/input.test.ts` and/or `week-input.test.ts` covering the exact inputs from the issue report.
 
 ## Out of scope
@@ -25,7 +25,7 @@ Both defects break user-visible output in the current 1.1.0 milestone.  The week
 1. `parseInput("task week Finalize Shoppinglist")` → `flags="task"`, `text="Finalize Shoppinglist"`, `week` = current week number.
 2. `parseInput("task do the shopping")` — behaviour unchanged (deferred; `do` still fires as German Thursday, text = "the shopping"). No regression.
 3. `parseInput("mo")` → `weekday` group matched, `offset` resolves to next/prev Monday relative to today (same as `parseInput("mon")`).
-4. Same for `tu`, `we`, `th`, `fr`.
+4. Same for `tu`, `fr`. `we` and `th` are NOT added — they collide with common English words.
 5. `parseInput("w15")` → `week=15` (no regression from week-pattern change).
 6. `parseInput("week")` alone (no following text) → `hasWeek()` true, no crash.
 7. Full test suite green.
@@ -50,7 +50,7 @@ const weekPattern = '(?<week>w(?:eek)?)(?=\\s(?!\\d)|$)';
 
 `getDayOfWeekForString` (dates.ts:86-90) already maps `mo → 1`, `tu → 2`, `we → 3`, `th → 4`, `fr → 5`.  `getWeekdayPattern()` only lists 3-letter English forms (`mon`, `tue`, `wed`, `thu`, `fri`), so the 2-letter variants never reach `resolveWeekday`.
 
-Fix: append `'mo', 'tu', 'we', 'th', 'fr'` after the 3-letter English group in the `alternatives` array (longest-first ordering preserved).
+Fix: append `'mo', 'tu', 'fr'` after the 3-letter English group in the `alternatives` array (longest-first ordering preserved). `we` and `th` are excluded — `we` is a common English pronoun and would destructively consume words like "we" in `"task we need to buy milk"`; `th` appears at the start of many English words.
 
 ## Entities / interfaces