polytypo 1.0.1 → 1.2.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: a60e1eac7173d0034f4846b2142a5b4168d0af023653b8b301e868d9dda7138f
-  data.tar.gz: f1911651b4d7f3014212b130e03e79bee1cb324aeee201e0e59d87bc5afedfc7
+  metadata.gz: 83d0c7e30bb5525df1fcde26cfc166534b11b92e83a68490b2fdb468af059a93
+  data.tar.gz: fba93f881b2fffc39b572cabf93dd5d0a0af8073ffe3179c9594b76c17e6d052
 SHA512:
-  metadata.gz: 3150585226505f0c29d8244f4c82c109a17d83e1a51e9cbe9f29368d2a9deddeda41603e2248508cd0fa372fb1d6361b36ed0e9175877e1d24ca2f23e338868f
-  data.tar.gz: 972340ac07969de576167bedf99918875727ae6877bbf55178b087b191c6bc1915b4d9378c01cd144e79d46b2e81d386a7fb71fa8c067f95cef91fde196a6e19
+  metadata.gz: 736115dda633084592307b6402ce41c40183c2b71d04b62ad95512f7ad8d62c5f43a29ea4380e718f8f1ee6d7b83da5af527f24a049e947b389b4b0e56b83980
+  data.tar.gz: dd20929d2101f1151bfd004bb9cdcfb89e3a7793442864fb35e897bf9295a21fbca7224f55efc3077ade7ed9d2079487493c8f26c051b70f6b0f39cfe701709c

data/lib/polytypo/data/VERSION

@@ -1 +1 @@
-1.0.0
+1.2.0

data/lib/polytypo/data/fixtures/de-CH.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "de-CH",
   "cases": [
     {
@@ -494,8 +494,16 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "spec 0.5.0: de-CH has no cited quotes.elisionIdioms entry for this idiom (quotes.md §7 item 8 — only en-US does, via Chicago Manual of Style Online and the American Heritage Dictionary). The general ambiguous-medial-span veto (quotes.md 3.2) preserves both marks as literal U+0027 rather than pairing them as an ordinary quotation, closing the false positive AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 3.1 named for \"equivalent cases in several locales.\""
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs."
+    },
+    {
+      "id": "de-ch-apostrophe-elision-before-guillemet",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "die Zeitung l'\"Express\"",
+      "out": "die Zeitung l’«Express»",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0): `quotes` gives de-CH's « », and the elision before « converts."
     }
   ]
 }

data/lib/polytypo/data/fixtures/de-DE.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "de-DE",
   "cases": [
     {
@@ -540,8 +540,40 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "spec 0.5.0: de-DE has no cited quotes.elisionIdioms entry for this idiom (quotes.md §7 item 8 — only en-US does, via Chicago Manual of Style Online and the American Heritage Dictionary). The general ambiguous-medial-span veto (quotes.md 3.2) preserves both marks as literal U+0027 rather than pairing them as an ordinary quotation, closing the false positive AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 3.1 named for \"equivalent cases in several locales.\""
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs."
+    },
+    {
+      "id": "de-de-spaces-dot-word-start-dotfile",
+      "rule": "spaces",
+      "mode": "text",
+      "in": "Die Datei .gitignore ist leer .",
+      "out": "Die Datei .gitignore ist leer.",
+      "note": "spaces.md §3.4 word-start clause (spec 1.2.0): the space before `.gitignore` survives because a letter follows the dot; the space before the final dot is deleted because nothing follows it."
+    },
+    {
+      "id": "de-de-apostrophe-possessive-before-closing-quote",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "Er sagte „Hans'“ und ging.",
+      "out": "Er sagte „Hans’“ und ging.",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0), §6 row 17: U+201C closes a de-DE quotation but is not in CLOSEISH, so case 3 missed the possessive; case 3a reaches the same U+2019 without reading locale data."
+    },
+    {
+      "id": "de-de-apostrophe-closing-single-before-closing-double-not-elision",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "\"Er sagte 'Hans'\" und ging.",
+      "out": "„Er sagte ‚Hans‘“ und ging.",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0), §7 item 7: in de-DE the closing double quote U+201C is in OPENQUOTE, so a closing single quote directly before it has exactly case 3a's shape. `quotes` (order 40) pairs both levels first, so no U+0027 reaches this rule and the nested quotation keeps its de-DE glyphs."
+    },
+    {
+      "id": "de-de-apostrophe-mixed-curly-and-straight-closing-single",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "„Er sagte ‚Hans'“ und ging.",
+      "out": "„Er sagte ‚Hans‘“ und ging.",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0): the same shape with the outer pair and the opening single quote already curly. `quotes` still pairs the straight closing single quote with ‚, so case 3a never sees it."
     }
   ]
 }

data/lib/polytypo/data/fixtures/el.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "el",
   "cases": [
     {
@@ -133,8 +133,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "\"Είπε 'όχι' σε μένα\", σημείωσε.",
-      "out": "«Είπε 'όχι' σε μένα», σημείωσε.",
-      "note": "quotes.md §6: one stack, inner pair at depth 2 takes the secondary pair U+201C/U+201D. Greek combines guillemets with curly double quotes, which no other v1 locale does. spec 0.5.0: the inner 'όχι' (3 Greek letters, space-flanked) is preserved by the general ambiguous-medial-span veto — el has no elisionIdioms entry for this shape."
+      "out": "«Είπε “όχι” σε μένα», σημείωσε.",
+      "note": "quotes.md §6: one stack, inner pair at depth 2 takes the secondary pair U+201C/U+201D. Greek combines guillemets with curly double quotes, which no other v1 locale does. spec 1.1.0 restores the nesting this case was written for: 0.5.0's general ambiguous-medial-span veto had preserved the inner 'όχι' as literal U+0027 because it was 3 letters, space-flanked; the veto that replaced it matches only a medial n (quotes.md §3.2), so an ordinary short quotation pairs again."
     },
     {
       "id": "el-quotes-already-guillemets",
@@ -205,8 +205,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "\"Είπε 'όχι' σε μένα\", σημείωσε.",
-      "out": "«Είπε 'όχι' σε μένα», σημείωσε.",
-      "note": "quotes.md §6 row 16: width drift with no gate intervention — el's secondary pair is WIDE (“ ”), so the inner NARROW pair drifts to WIDE on render and all four marks land on one stack in the gate's re-scan; the certification gate certifies without declining anything, evidence the gate's cost is near zero for the common case. spec 0.5.0: same as el-quotes-nested — 'όχι' is preserved by the general ambiguous-medial-span veto rather than pairing as an ordinary same-width quotation."
+      "out": "«Είπε “όχι” σε μένα», σημείωσε.",
+      "note": "quotes.md §6 row 16: width drift with no gate intervention — el's secondary pair is WIDE (“ ”), so the inner NARROW pair drifts to WIDE on render and all four marks land on one stack in the gate's re-scan; the certification gate certifies without declining anything, evidence the gate's cost is near zero for the common case. spec 1.1.0: same as el-quotes-nested — the width-drift path this case exists to exercise is reachable again now that a 3-letter medial span is no longer vetoed (quotes.md §3.2). Under 0.5.0 the inner marks never paired, so nothing drifted and the case proved nothing about the gate."
     },
     {
       "id": "el-quotes-030-adversarial-interleaved",
@@ -221,8 +221,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "spec 0.5.0: el has no cited quotes.elisionIdioms entry for this idiom (quotes.md §7 item 8 — only en-US does, via Chicago Manual of Style Online and the American Heritage Dictionary). The general ambiguous-medial-span veto (quotes.md 3.2) preserves both marks as literal U+0027 rather than pairing them as an ordinary quotation, closing the false positive AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 3.1 named for \"equivalent cases in several locales.\""
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs."
     },
     {
       "id": "el-ranges-none",
@@ -234,6 +234,14 @@
         "ranges": true
       },
       "note": "spec 0.5.0: el's dash.range is \"none\" (no verified convention), so `ranges` substitutes nothing even when explicitly enabled (ranges.md §2, §3.3) — the same invariant the pre-0.5.0 `dashes`-tagged no-op fixtures for this locale already proved for the combined rule."
+    },
+    {
+      "id": "el-apostrophe-elision-before-guillemet",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "η εφημερίδα l'\"Humanité\"",
+      "out": "η εφημερίδα l’«Humanité»",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0): `quotes` gives el's « », and the elision before « converts."
     }
   ]
 }

data/lib/polytypo/data/fixtures/en-GB.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "en-GB",
   "cases": [
     {
@@ -632,8 +632,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "\"He said 'no' to me,\" she noted.",
-      "out": "‘He said 'no' to me,’ she noted.",
-      "note": "quotes.md §6 case 2, inverted: depth 1 → single, depth 2 → double. Compare en-US-quotes-nested, same input. The inner 'no' (2 letters, space-flanked, no matching elisionIdioms entry — en-GB has none) is preserved by the general ambiguous-medial-span veto (quotes.md §3.2) rather than paired as a nested quotation: a documented false negative, not an unresolved gap (History: this was open per quotes.md §7 item 8 through spec 0.4.1, closed by the general veto in spec 0.5.0)."
+      "out": "‘He said “no” to me,’ she noted.",
+      "note": "quotes.md §6 case 2, inverted: depth 1 → single, depth 2 → double. Compare en-us-quotes-nested, same input. spec 1.1.0: the inner 'no' pairs as a nested quotation again — 0.5.0's general ambiguous-medial-span veto had preserved it as literal U+0027, and the universal medial-n veto that replaced it does not match a two-letter span (quotes.md §3.2)."
     },
     {
       "id": "en-gb-quotes-nested-inverted-input",
@@ -712,8 +712,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "quotes.md §7.1. The output is accidentally right here — not because a secondary level was selected, but because en-GB declares the single pair as PRIMARY and the pair is at depth 1. spec 0.5.0: en-GB has no cited elisionIdioms entry for this idiom (quotes.md §7 item 8 — elevated ambiguity risk, since en-GB's primary pair is itself the single quote). The general ambiguous-medial-span veto now preserves the marks as literal U+0027 instead of pairing them as an ordinary quotation, which is the pre-0.5.0 behaviour this fixture used to assert. AUDIT_REMEDIATION_AND_RELEASE_PLAN.md §3.1 named exactly this locale as an affected case."
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs. en-GB is the locale quotes.md §7 item 8 flags as carrying the highest genuine-dialogue collision risk, because its primary pair is itself the single quote — that risk is why no elisionIdioms entry was ever cited here, and 1.1.0's veto deliberately does not depend on one."
     },
     {
       "id": "en-gb-quotes-elision-inside-quotation",
@@ -1269,6 +1269,22 @@
       "in": "'90s were fun,' he said",
       "out": "‘90s were fun,’ he said",
       "note": "quotes.md §6 row E2s: the straight-input control, confirming the corruption in E2 predates mandate 1."
+    },
+    {
+      "id": "en-gb-spaces-dot-word-start-file-extensions",
+      "rule": "spaces",
+      "mode": "text",
+      "in": "CAD files (.DWG, .STEP) on request",
+      "out": "CAD files (.DWG, .STEP) on request",
+      "note": "spaces.md §3.4 word-start clause (spec 1.2.0), §6 row 10g: `.STEP` starts a token, so the space after the comma survives. Spec 1.1.0 produced `(.DWG,.STEP)`."
+    },
+    {
+      "id": "en-gb-apostrophe-elision-before-curly-quote",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "dell'“arte” moderna",
+      "out": "dell’‘arte’ moderna",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0), §6 row 16: `quotes` re-typesets the curly pair as en-GB's primary single quotes, and the elision before the opening glyph converts to U+2019."
     }
   ]
 }

data/lib/polytypo/data/fixtures/en-US.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "en-US",
   "cases": [
     {
@@ -755,8 +755,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "\"He said 'no' to me,\" she noted.",
-      "out": "“He said 'no' to me,” she noted.",
-      "note": "quotes.md §6 case 2: one stack, inner pair at depth 2 → secondary. spec 0.5.0: the inner 'no' is now an ambiguous medial span (2 letters, space-flanked, no matching en-US elisionIdioms entry — the idiom list holds only rock/n/roll) and is preserved as literal U+0027 rather than paired as a nested quotation (quotes.md 3.2, general ambiguous-medial-span veto). A documented false negative, not a regression: quotes.md 3.2 explains why a structural, portable predicate cannot distinguish this from rock 'n' roll's shape."
+      "out": "“He said ‘no’ to me,” she noted.",
+      "note": "quotes.md §6 case 2: one stack, inner pair at depth 2 → secondary. spec 1.1.0 restores the nesting this case was written for: 0.5.0's general ambiguous-medial-span veto had preserved the inner 'no' as literal U+0027 because it was 2 letters and space-flanked, and the veto that replaced it matches only a medial n (quotes.md §3.2), so an ordinary short quotation pairs again. Compare en-gb-quotes-nested, same input, inverted glyphs."
     },
     {
       "id": "en-us-quotes-nested-inverted-input",
@@ -836,7 +836,7 @@
       "mode": "text",
       "in": "rock 'n' roll",
       "out": "rock ’n’ roll",
-      "note": "quotes.md §3.2's listed elision veto (spec 0.4.0): quotes.elisionIdioms = [{ left: \"rock\", elided: \"n\", right: \"roll\" }] (en-US.json sources: Chicago Manual of Style Online on the mark's function, American Heritage Dictionary on the spaced variant) makes both marks decline pairing when the full left/elided/right context matches, so they survive to apostrophe (order 50), which independently renders each as U+2019 — leading elision (case 4) then trailing elision (case 3). apostrophe.md §6 row 11a. Contrast en-GB/fi/sv, which have no citable elisionIdioms entry: the general ambiguous-medial-span veto (quotes.md §3.2) preserves the same bytes unconverted there instead — see en-gb-quotes-rock-n-roll. Matching elided content alone (a bare word list) was tried first and withdrawn: it falsely elided ordinary quotations of the letter n — see en-us-quotes-letter-n-not-elided and en-us-quotes-nested-quote-n-not-elided."
+      "note": "quotes.md §3.2's listed elision veto (spec 0.4.0): quotes.elisionIdioms = [{ left: \"rock\", elided: \"n\", right: \"roll\" }] (en-US.json sources: Chicago Manual of Style Online on the mark's function, American Heritage Dictionary on the spaced variant) makes both marks decline pairing when the full left/elided/right context matches, so they survive to apostrophe (order 50), which independently renders each as U+2019 — leading elision (case 4) then trailing elision (case 3). apostrophe.md §6 row 11a. This output has been constant since 0.4.0 and is the model every other locale was brought into line with in spec 1.1.0, whose universal medial-n veto (quotes.md §3.2) now vetoes a superset of these positions — see ru-quotes-rock-n-roll for the same output reached without a citation. Matching elided content alone (a bare word list) was tried and withdrawn in 0.4.0; 1.1.0 readmits that cost knowingly and only for a single code point — see en-us-quotes-letter-n-elided."
     },
     {
       "id": "en-us-quotes-elision-inside-quotation",
@@ -844,7 +844,7 @@
       "mode": "text",
       "in": "She said \"it's the '90s\" and left.",
       "out": "She said “it’s the ’90s” and left.",
-      "note": "quotes.md §3.3 and §5.5: pass 2 keeps TWO STACKS, one per kind, and a candidate only ever touches the stack for its own kind — no reader opens with `\"` and closes with `'`, in any language. So the two DQ marks pair with each other, and the SQ elision marks find no SQ partner, survive unmatched, and `apostrophe` renders them U+2019. A PORT WITH ONE SHARED STACK PRODUCES `She said \"it’s the “90s” and left.` — the `'` of `'90s` is canClose, it pops the opening `\"`, and the real closing `\"` is stranded. That is damage rather than a miss (§5.5 calls it a release blocker). Contrast the `rock 'n' roll` case, which is a MEDIAL elision between two SQ marks, not this rule's leading/trailing single-mark elision shape — see en-us-quotes-rock-n-roll for how spec 0.5.0's general ambiguous-medial-span veto handles that different shape. History: before the per-kind stack split landed (pre-0.3.0), this exact case's canonical fixture asserted the single-shared-stack damage shown above (the stranded-quote output) as its expected `out` — a previous implementation defect, not a documented limitation; the two-stack design fixed it, and the `out` actually asserted above (the correct, two-stack form) has been correct ever since."
+      "note": "quotes.md §3.3 and §5.5: pass 2 keeps TWO STACKS, one per kind, and a candidate only ever touches the stack for its own kind — no reader opens with `\"` and closes with `'`, in any language. So the two DQ marks pair with each other, and the SQ elision marks find no SQ partner, survive unmatched, and `apostrophe` renders them U+2019. A PORT WITH ONE SHARED STACK PRODUCES `She said \"it’s the “90s” and left.` — the `'` of `'90s` is canClose, it pops the opening `\"`, and the real closing `\"` is stranded. That is damage rather than a miss (§5.5 calls it a release blocker). Contrast the `rock 'n' roll` case, which is a MEDIAL elision between two SQ marks, not this rule's leading/trailing single-mark elision shape — see en-us-quotes-rock-n-roll for how spec 1.1.0's universal medial-n veto handles that different shape. History: before the per-kind stack split landed (pre-0.3.0), this exact case's canonical fixture asserted the single-shared-stack damage shown above (the stranded-quote output) as its expected `out` — a previous implementation defect, not a documented limitation; the two-stack design fixed it, and the `out` actually asserted above (the correct, two-stack form) has been correct ever since."
     },
     {
       "id": "en-us-quotes-adjacent-to-punctuation",
@@ -1653,47 +1653,47 @@
       "mode": "text",
       "in": "rock ’n’ roll",
       "out": "rock ’n’ roll",
-      "note": "quotes.md §3.2/§5, §6 row P4: U+2019 is NARROW, so the veto fires identically on already-correct input — a fixed point. Discharges the idempotency argument's second-pipeline-pass case for this construction."
+      "note": "quotes.md §3.2/§5: U+2019 is NARROW, so the veto fires identically on already-correct input — a fixed point. This is the witness for spec 1.1.0's widening of the predicate from straight ASCII to the whole NARROW class, which is an idempotency obligation rather than a preference: the veto now lets apostrophe CONVERT both marks, so a straight-ASCII-only predicate would not recognise its own output and pass 2 would pair rock ’n’ roll as an ordinary NARROW quotation on the next pipeline run. See ru-quotes-rock-n-roll-already-curly and fi-quotes-rock-n-roll-already-curly, which fail loudly under a straight-only predicate where this en-US case would not."
     },
     {
-      "id": "en-us-quotes-letter-n-not-elided",
+      "id": "en-us-quotes-letter-n-elided",
       "rule": "quotes",
       "mode": "text",
       "in": "The letter 'n' is common.",
-      "out": "The letter 'n' is common.",
-      "note": "quotes.md §3.2: no `left`/`right` context matches (\"letter\" is not \"rock\"), so the listed-idiom veto does not fire — but 'n' (1 letter, space-flanked) is preserved by the general ambiguous-medial-span veto regardless (quotes.md §3.2), which does not require a citable idiom, only the structural shape. Pinned because this exact input was falsely elided by the first, context-free `elisionForms` design (spec 0.4.0, withdrawn before release) — see quotes.md §7 item 8. Contrast en-us-quotes-rock-n-roll, where the full left/elided/right context does match a cited idiom and apostrophe independently converts both marks."
+      "out": "The letter ’n’ is common.",
+      "note": "quotes.md §3.2, §6 row N1: no left/right context matches (\"letter\" is not \"rock\"), so the listed-idiom veto does not fire — but the universal medial-n veto does, and apostrophe converts both marks. **This is spec 1.1.0's accepted false positive**: a genuine quotation of the letter n comes out as an elision. It is pinned precisely because it is the input that sank the context-free `elisionForms` design in 0.4.0; 1.1.0 accepts it deliberately, against the measured alternative recorded in quotes.md §7 item 8 — spec 0.5.0 avoided this one case by declining every short quotation in every locale, a far larger error class. Through spec 0.5.0 this case asserted the unconverted input and was named en-us-quotes-letter-n-not-elided."
     },
     {
-      "id": "en-us-quotes-nested-quote-n-not-elided",
+      "id": "en-us-quotes-nested-quote-n-elided",
       "rule": "quotes",
       "mode": "text",
       "in": "He said \"press 'n' now\".",
-      "out": "He said “press 'n' now”.",
-      "note": "quotes.md §3.2: same reason as en-us-quotes-letter-n-not-elided (\"press\" is not \"rock\"), nested one level deeper inside a resolved outer quotation. The inner 'n' (1 letter, space-flanked) is preserved by the general ambiguous-medial-span veto — quotes.md §3.2's predicate is local to the mark's own immediate neighbours, not aware of enclosing pairs, so nesting depth is irrelevant to it. Also pinned as a regression witness for the withdrawn context-free `elisionForms` design (quotes.md §7 item 8)."
+      "out": "He said “press ’n’ now”.",
+      "note": "quotes.md §3.2, §6 row N2: the same shape as en-us-quotes-letter-n-elided, one level deeper inside a resolved outer quotation. §3.2's predicate reads a mark's immediate neighbours and knows nothing of enclosing pairs, so nesting depth never reaches it and the inner marks are converted while the outer pair resolves normally. The second half of spec 1.1.0's accepted false positive, pinned for the same reason. Through spec 0.5.0 this case asserted the unconverted inner marks and was named en-us-quotes-nested-quote-n-not-elided."
     },
     {
-      "id": "en-us-quotes-rock-n-pop-not-elided",
+      "id": "en-us-quotes-rock-n-pop-elided",
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' pop",
-      "out": "rock 'n' pop",
-      "note": "quotes.md §3.2, §6 row N3: `right` is \"pop\", not \"roll\" — no configured idiom matches. 'n' (1 letter, space-flanked) is preserved by the general ambiguous-medial-span veto instead (quotes.md §3.2), which does not require a citable idiom to fire, only the structural shape — a documented false negative rather than a pairing."
+      "out": "rock ’n’ pop",
+      "note": "quotes.md §3.2, §6 row N3: `right` is \"pop\", not \"roll\", so no configured idiom matches — and since spec 1.1.0 none is needed. The universal medial-n veto fires on the bare medial n and apostrophe converts both marks. Named en-us-quotes-rock-n-pop-not-elided through spec 0.5.0, when the same input was preserved unconverted."
     },
     {
-      "id": "en-us-quotes-fish-n-chips-not-elided",
+      "id": "en-us-quotes-fish-n-chips-elided",
       "rule": "quotes",
       "mode": "text",
       "in": "fish 'n' chips",
-      "out": "fish 'n' chips",
-      "note": "quotes.md §3.2, §6 row N4: not sourced or listed. en-US.json's AHD citation attests only rock/n/roll; no { left: \"fish\", elided: \"n\", right: \"chips\" } entry exists. 'n' is preserved by the general ambiguous-medial-span veto instead (quotes.md §3.2), the same mechanism as en-us-quotes-rock-n-pop-not-elided — no idiom is required for it to fire."
+      "out": "fish ’n’ chips",
+      "note": "quotes.md §3.2, §6 row N4: en-US.json's AHD citation attests only rock/n/roll, and no { left: \"fish\", elided: \"n\", right: \"chips\" } entry exists — the universal medial-n veto covers it anyway, which is the practical reason spec 1.1.0 stopped requiring a citation per phrase: the long tail of X 'n' Y borrowings cannot be enumerated one citation at a time. Named en-us-quotes-fish-n-chips-not-elided through spec 0.5.0."
     },
     {
-      "id": "en-us-quotes-rock-cap-n-roll-not-elided",
+      "id": "en-us-quotes-rock-cap-n-roll-elided",
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'N' roll",
-      "out": "rock 'N' roll",
-      "note": "quotes.md §3.2, §6 row N5: `elided` is matched exactly, with no case leniency at all — the first-code-point leniency applies only to `left`/`right`. `N` ≠ `n`, so no idiom matches (the en-US elisionIdioms entry requires exactly lower-case \"n\"). 'N' is preserved by the general ambiguous-medial-span veto instead (quotes.md §3.2), which is case-insensitive to the enclosed letter — this fixture is the witness that idiom-matching stays case-sensitive even though the general veto covering the same bytes is not."
+      "out": "rock ’N’ roll",
+      "note": "quotes.md §3.2, §6 row N5: `elided` is matched exactly by the LISTED mechanism, with no case leniency at all — the first-code-point leniency applies only to left/right — so `N` ≠ `n` still fails the cited en-US idiom. The universal medial-n veto matches U+004E as well as U+006E and converts both marks. This fixture is the witness that idiom-matching stays case-sensitive even though the veto covering the same bytes is not. Named en-us-quotes-rock-cap-n-roll-not-elided through spec 0.5.0."
     },
     {
       "id": "en-us-quotes-rock-n-roll-html",
@@ -1708,16 +1708,16 @@
       "rule": "quotes",
       "mode": "html",
       "in": "<p><em>rock</em> 'n' roll</p>",
-      "out": "<p><em>rock</em> 'n' roll</p>",
-      "note": "quotes.md §3.2, §6 row H1: modes.md §3.2's boundary marker is not INLINE-SPACE, so `left`'s word is unreachable across the <em> span boundary and the veto does not fire — matches text mode's own boundary behaviour on split input. spec 0.5.0: the general ambiguous-medial-span veto's left-context test only reads the single INLINE-SPACE code point immediately before the mark (quotes.md 3.2), which sits inside the same prose span as the mark itself — unlike idiom word-matching (wordEndsAt), it does not need to see past the span-boundary marker to \"rock\" in the preceding <em> span, so it fires here even though the idiom match (pre-0.5.0 behaviour, still correctly not firing) does not."
+      "out": "<p><em>rock</em> ’n’ roll</p>",
+      "note": "quotes.md §3.2, §6 row H1: modes.md §3.2's boundary marker is not INLINE-SPACE, so `left`'s word is unreachable across the <em> span boundary and the LISTED idiom does not match. The universal medial-n veto (spec 1.1.0) reads no context word at all — only the enclosed code point and the single space immediately outside each mark, both of which sit in the same prose span as the marks — so it fires here and apostrophe converts both marks, converging with the unsplit en-us-quotes-rock-n-roll-html. Through spec 0.5.0 this case asserted the unconverted input."
     },
     {
       "id": "en-us-quotes-rock-n-roll-html-right-split",
       "rule": "quotes",
       "mode": "html",
       "in": "<p>rock 'n' <em>roll</em></p>",
-      "out": "<p>rock 'n' <em>roll</em></p>",
-      "note": "quotes.md §3.2, §6 row H2: mirror case on `right`. spec 0.5.0: symmetric case to the left-split row above — the general veto's right-context test reads the INLINE-SPACE immediately after the mark, inside the same span, and fires regardless of \"roll\" sitting in a following <em> span."
+      "out": "<p>rock ’n’ <em>roll</em></p>",
+      "note": "quotes.md §3.2, §6 row H2: the mirror of en-us-quotes-rock-n-roll-html-left-split, with `right`'s word inside the following <em> span. modes.md §3.2's boundary marker is not INLINE-SPACE, so `right` is unreachable and the LISTED idiom does not match. The universal medial-n veto (spec 1.1.0) reads no context word at all — only the enclosed code point and the single space immediately outside each mark, both in the same prose span as the marks — so it fires and apostrophe converts both, converging with the unsplit en-us-quotes-rock-n-roll-html. Through spec 0.5.0 this case asserted the unconverted input."
     },
     {
       "id": "en-us-quotes-rock-n-roll-markdown-left-split",
@@ -1725,32 +1725,32 @@
       "mode": "markdown",
       "dialect": "commonmark",
       "in": "*rock* 'n' roll\n",
-      "out": "*rock* 'n' roll\n",
-      "note": "quotes.md §3.2, §6 row H3: `left`'s word sits inside an emphasis span; the boundary marker between spans is not INLINE-SPACE, so the veto does not cross it. spec 0.5.0: same mechanism as the HTML left-split row — Markdown emphasis creates a span boundary before \"rock\", but the general veto's left-context test only needs the immediately-preceding INLINE-SPACE, which is in the same span as the mark."
+      "out": "*rock* ’n’ roll\n",
+      "note": "quotes.md §3.2, §6 row H3: `left`'s word sits inside an emphasis span, so the LISTED idiom's word-boundary test cannot reach it across the span marker (modes.md §3.2) and no idiom matches. The universal medial-n veto (spec 1.1.0) needs no context word — the mark's own immediately-preceding INLINE-SPACE is in the same span as the mark — so it fires and apostrophe converts both marks, matching text mode on the same characters split the same way. Through spec 0.5.0 this case asserted the unconverted input."
     },
     {
       "id": "en-us-quotes-ambiguous-single-letter",
       "rule": "quotes",
       "mode": "text",
       "in": "She chose 'A' today",
-      "out": "She chose 'A' today",
-      "note": "spec 0.5.0, general ambiguous-medial-span veto (quotes.md 3.2): 1 LETTER enclosed, space-flanked on both sides, no matching elisionIdioms entry — preserved as literal U+0027 on both marks. A documented false negative: this reads as a plausible genuine quotation of the letter A, but the structural predicate cannot distinguish it from rock 'n' roll's shape without reasoning about meaning, which is out of scope (AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 3.1: false negatives are preferable to text damage)."
+      "out": "She chose “A” today",
+      "note": "spec 1.1.0, quotes.md §3.2: 'A' is a single quoted letter and pairs as an ordinary quotation. Under spec 0.5.0 the general ambiguous-medial-span veto preserved it as literal U+0027, because its shape test admitted any 1 to 3 LETTER code points and could not tell this from rock 'n' roll. This fixture flipped role in 1.1.0: it pinned that veto's reach, and now pins that the replacement veto does NOT reach a span whose enclosed code point is not n or N."
     },
     {
       "id": "en-us-quotes-ambiguous-two-letter",
       "rule": "quotes",
       "mode": "text",
       "in": "They said 'no' yesterday",
-      "out": "They said 'no' yesterday",
-      "note": "spec 0.5.0, general ambiguous-medial-span veto: 2 LETTERs enclosed, same reasoning as en-us-quotes-ambiguous-single-letter."
+      "out": "They said “no” yesterday",
+      "note": "spec 1.1.0, quotes.md §3.2: 2 letters enclosed — not a medial n, so the universal veto does not fire and the span pairs as an ordinary quotation. Same reasoning and same role reversal as en-us-quotes-ambiguous-single-letter."
     },
     {
       "id": "en-us-quotes-ambiguous-three-letter",
       "rule": "quotes",
       "mode": "text",
       "in": "say 'yes' now",
-      "out": "say 'yes' now",
-      "note": "spec 0.5.0, general ambiguous-medial-span veto: 3 LETTERs enclosed — the upper bound of the shape (quotes.md 3.2, MAX_ENCLOSED). Preserved."
+      "out": "say “yes” now",
+      "note": "spec 1.1.0, quotes.md §3.2: 3 letters enclosed — the upper bound of spec 0.5.0's withdrawn shape test, kept as a fixture precisely because it was that test's boundary case. The universal medial-n veto does not fire, and the span pairs as an ordinary quotation."
     },
     {
       "id": "en-us-quotes-ambiguous-four-letter-not-caught",
@@ -1765,8 +1765,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "say '𐐀' now",
-      "out": "say '𐐀' now",
-      "note": "spec 0.5.0: the enclosed code point is U+10400 (DESERET CAPITAL LETTER LONG A, astral plane) — general category Lu, a member of LETTER. The veto's LETTER test (engine/unicode.js isLetter) operates on full code points, not UTF-16 units, so this is treated identically to an ASCII letter and preserved."
+      "out": "say “𐐀” now",
+      "note": "spec 1.1.0, quotes.md §3.2: a single astral-plane LETTER (U+10400, two UTF-16 units, one code point). It pairs as an ordinary quotation because it is not n or N — the veto compares code points, so an astral letter is neither specially admitted nor specially excluded. Retained from spec 0.5.0, where it pinned that the withdrawn shape test counted code points rather than UTF-16 units; the code-point discipline it tests (ARCHITECTURE.md §4.2) is unchanged."
     },
     {
       "id": "en-us-quotes-ambiguous-digit-not-letter",
@@ -1774,7 +1774,7 @@
       "mode": "text",
       "in": "say '12' now",
       "out": "say “12” now",
-      "note": "spec 0.5.0: the enclosed code points are DIGIT, not LETTER, so the general ambiguous-medial-span veto does not fire — digits must not accidentally qualify as a LETTER sequence (quotes.md 3.2). Pairs as an ordinary quotation exactly as before 0.5.0."
+      "note": "spec 1.1.0, quotes.md §3.2: the enclosed code points are digits, not n or N, so the veto does not fire and the span pairs as an ordinary quotation. This output has been constant across 0.5.0 and 1.1.0 — under 0.5.0 because digits were not LETTER, now because they are not the one code point the veto matches."
     },
     {
       "id": "en-us-quotes-ambiguous-punctuation-not-letter",
@@ -1790,7 +1790,7 @@
       "mode": "text",
       "in": "She chose ’A’ today",
       "out": "She chose “A” today",
-      "note": "spec 0.5.0: the general ambiguous-medial-span veto matches straight ASCII U+0027 pairs only (quotes.md 3.2) — an already-curly U+2019 pair is out of its scope by construction, so this falls through to ordinary NARROW-pair quotation exactly as it did before 0.5.0. Documents the chosen contract deliberately, so the ASCII-only scope is not silently broadened later."
+      "note": "spec 1.1.0, quotes.md §3.2: 'A' is not n or N, so the veto does not fire and the already-curly pair is re-typeset as an ordinary quotation. **The reason this output is unchanged is not the reason it was unchanged in 0.5.0.** 0.5.0's veto matched straight ASCII U+0027 only, so a curly pair was out of scope by construction; 1.1.0's matches the whole NARROW class — an idempotency obligation, since the veto now produces U+2019 — and declines this span on its enclosed code point instead. See en-us-quotes-rock-n-roll-already-curly for a curly pair the veto does match."
     },
     {
       "id": "en-us-ranges-decimal-and-path-rejected",
@@ -1802,6 +1802,70 @@
         "ranges": true
       },
       "note": "ranges.md §3.2, G3 — not part of a decimal or a path. `before` is U+002E (.) for the first candidate and U+002F (/) for the second; `after` is U+002F (/) for the second. This is currently the only canonical fixture exercising G3 — previously untested by any canonical case even before the spec 0.5.0 ownership correction (dashes.test.ts's engine-level 'must not touch' list covered it, but no fixture did)."
+    },
+    {
+      "id": "en-us-spaces-dot-word-start-dotnet",
+      "rule": "spaces",
+      "mode": "text",
+      "in": "Use .NET, .NET Core or .NET Framework",
+      "out": "Use .NET, .NET Core or .NET Framework",
+      "note": "spaces.md §3.4 word-start clause (spec 1.2.0), §6 row 10f: each dot is followed by a letter, so it starts a token and the space before it is not deleted. Spec 1.1.0 produced `Use.NET,.NET Core or.NET Framework`."
+    },
+    {
+      "id": "en-us-spaces-dot-word-start-decimal",
+      "rule": "spaces",
+      "mode": "text",
+      "in": "the values .5 and .9 apply",
+      "out": "the values .5 and .9 apply",
+      "note": "spaces.md §3.4 word-start clause (spec 1.2.0), §6 row 10h: an ASCII digit after the dot counts the same as a letter."
+    },
+    {
+      "id": "en-us-spaces-dot-word-start-misplaced-full-stop",
+      "rule": "spaces",
+      "mode": "text",
+      "in": "the end .Next sentence",
+      "out": "the end .Next sentence",
+      "note": "spaces.md §3.4, §6 row 10i: the accepted cost of the word-start clause. A misplaced full stop followed directly by a letter keeps its stray space, because deleting it would glue two words that a later pass cannot separate again."
+    },
+    {
+      "id": "en-us-spaces-dot-terminal-still-strips",
+      "rule": "spaces",
+      "mode": "text",
+      "in": "See p. 12 . Next",
+      "out": "See p. 12. Next",
+      "note": "spaces.md §3.4, §6 row 10j: a dot followed by a space is terminal punctuation, so the space before it is still deleted under spec 1.2.0."
+    },
+    {
+      "id": "en-us-nbsp-span-boundary-no-punctuation-data",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "<strong>Note :</strong> read this",
+      "out": "<strong>Note:</strong> read this",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0): accepting the marker changes nothing in a locale whose punctuation lists are empty. `spaces` deletes the space and nothing is put back, the same result as in text mode."
+    },
+    {
+      "id": "en-us-apostrophe-elision-before-quote",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "the magazine l'\"Uomo\" today",
+      "out": "the magazine l’“Uomo” today",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0): `quotes` converts the straight double quotes to U+201C/U+201D, and the elision before U+201C converts to U+2019."
+    },
+    {
+      "id": "en-us-apostrophe-prime-before-bracket-untouched",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "f'(x) = 2x",
+      "out": "f'(x) = 2x",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0), §6 row 18: `(` is in OPENISH but not in OPENQUOTE, so a prime on a function name is left alone."
+    },
+    {
+      "id": "en-us-spaces-dot-before-span-boundary-is-terminal",
+      "rule": "spaces",
+      "mode": "html",
+      "in": "Use .<b>NET</b> today",
+      "out": "Use.<b>NET</b> today",
+      "note": "spaces.md §3.4 word-start clause (spec 1.2.0): the code point after the dot is the −1 span boundary marker, which is neither a LETTER nor an ASCII digit (modes.md §3.3), so the dot counts as terminal and the space before it is deleted."
     }
   ]
 }

data/lib/polytypo/data/fixtures/fi.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "fi",
   "cases": [
     {
@@ -752,8 +752,16 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "quotes.md §7.1, recorded as a known defect: the two elision marks pair at depth 1 and take the Finnish primary glyphs. spec 0.5.0: fi's own dictionary writes this loanword closed up (rock'n'roll, no spaces) — quotes.md §7 item 8 — so a citation for the spaced form would be evidenced-inert and none was added. The general ambiguous-medial-span veto now preserves the spaced ASCII input as-is rather than pairing it as an ordinary quotation, which is the pre-0.5.0 behaviour this fixture used to assert."
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs. fi's own dictionary writes this loanword closed up (rock'n'roll), so a citation for the spaced form would have been evidenced-inert and none was ever added — quotes.md §7 item 8. The universal veto covers the spaced form regardless, which is the point of moving the mechanism out of locale data. See fi-quotes-rock-n-roll-already-curly for the second-pass fixed point this output depends on."
+    },
+    {
+      "id": "fi-quotes-rock-n-roll-already-curly",
+      "rule": "quotes",
+      "mode": "text",
+      "in": "rock ’n’ roll",
+      "out": "rock ’n’ roll",
+      "note": "quotes.md §3.2, the sharpest regression witness for spec 1.1.0's widening of the veto to the whole NARROW class: fi's SECONDARY pair is U+2019 on both sides, so the marks this rule's own output leaves behind are indistinguishable from fi's own prescribed nested-quotation glyphs. A port that matches U+0027 only does not recognise them, pairs them, and turns the output of fi-quotes-rock-n-roll into `rock ”n” roll` — fi's primary pair at depth 1. Companion to ru-quotes-rock-n-roll-already-curly, which fails the same way in a locale whose primary is WIDE; both are measured, not hypothesised."
     },
     {
       "id": "fi-quotes-already-correct",
@@ -1301,6 +1309,14 @@
       "in": "'90s were fun,' he said",
       "out": "”90s were fun,” he said",
       "note": "quotes.md §6 row E3s: the straight-input control for E3."
+    },
+    {
+      "id": "fi-apostrophe-elision-before-shared-quote-glyph",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "lehti l'\"Humanité\"",
+      "out": "lehti l’”Humanité”",
+      "note": "apostrophe.md §3.3 cases 3 and 3a (spec 1.2.0): fi opens and closes with U+201D, which is in CLOSEISH, so case 3 already converted this before 1.2.0. Pinned so that case 3a is not read as changing a shared-glyph locale."
     }
   ]
 }

data/lib/polytypo/data/fixtures/fr-CA.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "fr-CA",
   "cases": [
     {
@@ -80,8 +80,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "\"Il a dit 'oui' hier.\"",
-      "out": "« Il a dit 'oui' hier. »",
-      "note": "spec 0.5.0: same as fr-quotes-nested — 'oui' is preserved by the general ambiguous-medial-span veto."
+      "out": "« Il a dit “oui” hier. »",
+      "note": "spec 1.1.0: same as fr-quotes-nested — the inner 'oui' pairs as a nested quotation again now that the veto matches only a medial n (quotes.md §3.2), where 0.5.0's shape test preserved every 1-to-3-letter span."
     },
     {
       "id": "fr-ca-apostrophe-elision",
@@ -250,8 +250,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "spec 0.5.0: fr-CA has no cited quotes.elisionIdioms entry for this idiom (quotes.md §7 item 8 — only en-US does, via Chicago Manual of Style Online and the American Heritage Dictionary). The general ambiguous-medial-span veto (quotes.md 3.2) preserves both marks as literal U+0027 rather than pairing them as an ordinary quotation, closing the false positive AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 3.1 named for \"equivalent cases in several locales.\""
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs."
     },
     {
       "id": "fr-ca-ranges-none",
@@ -263,6 +263,31 @@
         "ranges": true
       },
       "note": "spec 0.5.0: fr-CA's dash.range is \"none\" (no verified convention), so `ranges` substitutes nothing even when explicitly enabled (ranges.md §2, §3.3) — the same invariant the pre-0.5.0 `dashes`-tagged no-op fixtures for this locale already proved for the combined rule."
+    },
+    {
+      "id": "fr-ca-nbsp-span-boundary-strong-colon",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "<strong>Résistant au gel :</strong> il reste souple",
+      "out": "<strong>Résistant au gel :</strong> il reste souple",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0): same as the `fr` case; `fr-CA` lists `:` in its own punctuation data."
+    },
+    {
+      "id": "fr-ca-nbsp-span-boundary-markdown-strong-colon",
+      "rule": "nbsp",
+      "mode": "markdown",
+      "dialect": "commonmark",
+      "in": "**Note :** voir plus bas\n",
+      "out": "**Note :** voir plus bas\n",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0): the colon ends the strong span; the −1 marker after it is in CLOSEISH, so N1 restores the U+00A0 that `spaces` deleted. (fr-CA lists no punctuation for N2, so `?` would take nothing back in any mode.)"
+    },
+    {
+      "id": "fr-ca-apostrophe-elision-before-straight-quote",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "l'\"idée\" est simple",
+      "out": "l’« idée » est simple",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0): same as `fr`."
     }
   ]
 }

data/lib/polytypo/data/fixtures/fr.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "fr",
   "cases": [
     {
@@ -235,8 +235,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "\"Il a dit 'oui' hier.\"",
-      "out": "« Il a dit 'oui' hier. »",
-      "note": "Depth 1 primary « », depth 2 secondary “ ” whose innerSpace is none. spec 0.5.0: the inner 'oui' (3 letters, space-flanked) is preserved by the general ambiguous-medial-span veto — fr has no elisionIdioms entry for this shape."
+      "out": "« Il a dit “oui” hier. »",
+      "note": "Depth 1 primary « », depth 2 secondary “ ” whose innerSpace is none. spec 1.1.0: the inner 'oui' pairs again — 0.5.0's general ambiguous-medial-span veto had preserved it as literal U+0027 for being 3 letters and space-flanked, and the universal medial-n veto that replaced it matches only a single n (quotes.md §3.2)."
     },
     {
       "id": "fr-quotes-unbalanced",
@@ -585,8 +585,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "spec 0.5.0: fr has no cited quotes.elisionIdioms entry for this idiom (quotes.md §7 item 8 — only en-US does, via Chicago Manual of Style Online and the American Heritage Dictionary). The general ambiguous-medial-span veto (quotes.md 3.2) preserves both marks as literal U+0027 rather than pairing them as an ordinary quotation, closing the false positive AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 3.1 named for \"equivalent cases in several locales.\""
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs."
     },
     {
       "id": "fr-ranges-none",
@@ -598,6 +598,112 @@
         "ranges": true
       },
       "note": "spec 0.5.0: fr's dash.range is \"none\" (no verified convention), so `ranges` substitutes nothing even when explicitly enabled (ranges.md §2, §3.3) — the same invariant the pre-0.5.0 `dashes`-tagged no-op fixtures for this locale already proved for the combined rule."
+    },
+    {
+      "id": "fr-spaces-dot-word-start-dotfile",
+      "rule": "spaces",
+      "mode": "text",
+      "in": "Voir le fichier .htaccess .",
+      "out": "Voir le fichier .htaccess.",
+      "note": "spaces.md §3.4 word-start clause (spec 1.2.0): the space before `.htaccess` survives, the space before the terminal dot is deleted. `.` is in no `nbsp` list for `fr`, so nothing is put back."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-strong-colon",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "<strong>Résistant au gel :</strong> il reste souple",
+      "out": "<strong>Résistant au gel :</strong> il reste souple",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0), §6 row 6c; modes.md §3.3: `spaces` deletes the typed space before `:`, and the −1 span boundary marker after the mark is in `nbsp`'s CLOSEISH, so N1 inserts U+00A0 at the mark's index, inside the span. Spec 1.1.0 runtimes produced `gel:</strong>`."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-br-question",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "Et la réglementation ?<br>Oui",
+      "out": "Et la réglementation ?<br>Oui",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0), §6 row 6d: `<br>` leaves no line terminator in the gap, so the marker is −1; it is in CLOSEISH and N2 restores U+202F."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-link-question",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "<a href=\"/faq\">Comment ?</a> Oui",
+      "out": "<a href=\"/faq\">Comment ?</a> Oui",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0): the mark ends the link text, the marker follows it, and N2 restores U+202F instead of leaving `Comment?`."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-markdown-strong-colon",
+      "rule": "nbsp",
+      "mode": "markdown",
+      "dialect": "commonmark",
+      "in": "**Résistant au gel :** il reste souple\n",
+      "out": "**Résistant au gel :** il reste souple\n",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0): the Markdown form of the French definition pattern; the closing `**` is a skipped region with no line terminator, so the marker is −1 and N1 restores U+00A0."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-not-openish-em",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "Il dit <em>non</em> ! Oui",
+      "out": "Il dit <em>non</em> ! Oui",
+      "note": "nbsp.md §3.3 step 3 (spec 1.2.0), §6 row 6e; modes.md §3.3: the marker is not in `nbsp`'s OPENISH, so the quote-glyph guard does not decline and N2 converts the space. A port that applied the quote-glyph guard to the marker would leave a plain U+0020 here."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-not-openish-link",
+      "rule": "nbsp",
+      "mode": "markdown",
+      "dialect": "commonmark",
+      "in": "Voir [ceci](http://example.org) : oui\n",
+      "out": "Voir [ceci](http://example.org) : oui\n",
+      "note": "nbsp.md §3.3 step 3 (spec 1.2.0): the second witness that the quote-glyph guard does not decline at a span boundary. The link text ends two places left of `:`, and N1 still converts the space. The marker's absence from the rest of `nbsp`'s OPENISH is pinned by ru-nbsp-span-boundary-not-openish-short-word."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-mark-alone-in-span",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "mot<em>!</em>",
+      "out": "mot<em>!</em>",
+      "note": "nbsp.md §3.3 step 2; modes.md §3.4 edge-growth rule: the mark is alone in its span, so the insertion point is the span edge and the insertion is discarded, as it was before spec 1.2.0."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-colon-before-digit-span",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "Il est 12:<b>30</b>",
+      "out": "Il est 12 :<b>30</b>",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0), §6 row 6f: the stated cost of accepting the marker. Step 2 cannot see past the span boundary, so a colon that ends one span while a digit starts the next is treated as sentence punctuation."
+    },
+    {
+      "id": "fr-apostrophe-elision-before-guillemet",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "un mélange d'« urine diluée »",
+      "out": "un mélange d’« urine diluée »",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0), §6 row 15: letter left, opening quotation glyph right. Spec 1.1.0 left the apostrophe straight, because no case matched a letter followed by an OPENISH-only glyph."
+    },
+    {
+      "id": "fr-apostrophe-elision-before-straight-quote",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "l'\"idée\" est simple",
+      "out": "l’« idée » est simple",
+      "note": "apostrophe.md §3.2, §3.3 case 3a (spec 1.2.0): `quotes` (order 40) turns the straight `\"` into `«` first, so the mark reaches this rule next to an opening glyph and case 3a converts it."
+    },
+    {
+      "id": "fr-apostrophe-elision-qu-before-guillemet",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "Il faut qu'« il vienne »",
+      "out": "Il faut qu’« il vienne »",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0): `qu'` before a quotation."
+    },
+    {
+      "id": "fr-nbsp-span-boundary-colon-before-url-span",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "Voir http:<a href=\"//x.org\">//x.org</a>",
+      "out": "Voir http :<a href=\"//x.org\">//x.org</a>",
+      "note": "nbsp.md §3.3 step 2 (spec 1.2.0): the second stated cost of accepting the marker. Step 2 cannot see the `/` in the next span, so the colon is treated as sentence punctuation and N1 inserts U+00A0."
     }
   ]
 }

data/lib/polytypo/data/fixtures/locale-resolution.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "cases": [
     {
       "id": "exact-en-us",

data/lib/polytypo/data/fixtures/ru.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "ru",
   "cases": [
     {
@@ -323,8 +323,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "Он сказал: \"это 'моё' дело\".",
-      "out": "Он сказал: «это 'моё' дело».",
-      "note": "quotes.md §6 case 14: depth 1 takes the ёлочки « », depth 2 the лапки „ “ — Лопатин 2006. Note the secondary pair's closing glyph is U+201C, which is in OPENISH; the class says nothing about the role. spec 0.5.0: the inner 'моё' (3 Cyrillic letters, space-flanked) is preserved by the general ambiguous-medial-span veto, which reads Unicode LETTER generally, not only ASCII — ru has no elisionIdioms entry for this shape."
+      "out": "Он сказал: «это „моё“ дело».",
+      "note": "quotes.md §6 case 14: depth 1 takes the ёлочки « », depth 2 the лапки „ “ — Лопатин 2006. Note the secondary pair's closing glyph is U+201C, which is in OPENISH; the class says nothing about the role. spec 1.1.0: the inner 'моё' pairs as a nested quotation again. Under 0.5.0 the general ambiguous-medial-span veto preserved it as literal U+0027 for being 3 LETTER code points and space-flanked — a count that also depended on the input's normalization form, since LETTER includes Mn and the decomposed ё is two code points. The universal medial-n veto that replaced it compares one code point against a fixed pair (quotes.md §3.2) and has neither problem."
     },
     {
       "id": "ru-quotes-roundtrip",
@@ -681,8 +681,40 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "spec 0.5.0: ru has no cited quotes.elisionIdioms entry for this idiom (quotes.md §7 item 8 — only en-US does, via Chicago Manual of Style Online and the American Heritage Dictionary). The general ambiguous-medial-span veto (quotes.md 3.2) preserves both marks as literal U+0027 rather than pairing them as an ordinary quotation, closing the false positive AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 3.1 named for \"equivalent cases in several locales.\""
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs. See ru-quotes-rock-n-roll-already-curly for the second-pass fixed point this output depends on."
+    },
+    {
+      "id": "ru-quotes-rock-n-roll-already-curly",
+      "rule": "quotes",
+      "mode": "text",
+      "in": "rock ’n’ roll",
+      "out": "rock ’n’ roll",
+      "note": "quotes.md §3.2, the regression witness for spec 1.1.0's widening of the veto from straight ASCII to the whole NARROW class. This is the output of ru-quotes-rock-n-roll fed back in, so it is that case's idempotency obligation made explicit — and it is the case a naive implementation gets wrong. A port that matches U+0027 only (as spec 0.5.0's withdrawn predicate did, correctly for its own preserve-then-stop outcome) does not recognise the U+2019 marks it just produced: the pair falls through to ordinary NARROW quotation and this input becomes `rock «n» roll`, ru's primary pair at depth 1. Measured, not hypothesised. `apostrophe` emits U+2019 only in place of U+0027 (apostrophe.md §1), so once the veto matches, nothing edits these marks and the form is stable."
+    },
+    {
+      "id": "ru-spaces-dot-word-start-dotfile",
+      "rule": "spaces",
+      "mode": "text",
+      "in": "Проверьте файл .env сейчас",
+      "out": "Проверьте файл .env сейчас",
+      "note": "spaces.md §3.4 word-start clause (spec 1.2.0): a dot followed by a letter starts a token in every locale; `spaces` reads no locale data (§2)."
+    },
+    {
+      "id": "ru-apostrophe-elision-before-guillemet",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "газета l'\"Humanité\"",
+      "out": "газета l’«Humanité»",
+      "note": "apostrophe.md §3.3 case 3a (spec 1.2.0): a French title in Russian text; `quotes` gives « », and the elision before « converts."
+    },
+    {
+      "id": "ru-nbsp-span-boundary-not-openish-short-word",
+      "rule": "nbsp",
+      "mode": "html",
+      "in": "Он живёт в <em>Москве</em>",
+      "out": "Он живёт в <em>Москве</em>",
+      "note": "nbsp.md §3.1, §7 item 12 (spec 1.2.0); modes.md §3.3: the −1 span boundary marker is not in `nbsp`'s OPENISH, so N3's following-token guard does not accept it and the space after `в` stays U+0020. This pins the membership split itself: a port that also put the marker in OPENISH would bind here."
     }
   ]
 }

data/lib/polytypo/data/fixtures/sv.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "locale": "sv",
   "cases": [
     {
@@ -752,8 +752,8 @@
       "rule": "quotes",
       "mode": "text",
       "in": "rock 'n' roll",
-      "out": "rock 'n' roll",
-      "note": "quotes.md §7.1, a known defect: the elision marks pair at depth 1 and take the Swedish primary glyphs. spec 0.5.0: same reasoning as fi-quotes-rock-n-roll — sv also writes the loanword closed up, no citation exists for the spaced form, and the general ambiguous-medial-span veto now preserves the ASCII input."
+      "out": "rock ’n’ roll",
+      "note": "spec 1.1.0, quotes.md §3.2's universal medial-n elision veto: by operator decision the idiom is international, so `quotes` declines the pairing in every locale rather than only where a cited quotes.elisionIdioms entry exists, and `apostrophe` (order 50) converts each mark by its own case ladder — leading elision (apostrophe.md §3.3 case 4), then trailing elision (case 3). Same output and same two case decisions as en-us-quotes-rock-n-roll, which reaches them through the cited idiom instead. History: spec 0.5.0's general ambiguous-medial-span veto preserved both marks here as literal U+0027; before 0.5.0 `quotes` paired them as an ordinary depth-1 quotation in this locale's primary glyphs. sv writes the loanword closed up in its own dictionaries, exactly as fi does, so no spaced-form citation exists here either — see fi-quotes-rock-n-roll."
     },
     {
       "id": "sv-quotes-already-correct",
@@ -1285,6 +1285,14 @@
       "in": "'90s were fun,' he said",
       "out": "”90s were fun,” he said",
       "note": "quotes.md §6 row E4s: the straight-input control for E4."
+    },
+    {
+      "id": "sv-apostrophe-elision-before-shared-quote-glyph",
+      "rule": "apostrophe",
+      "mode": "text",
+      "in": "tidningen l'\"Humanité\"",
+      "out": "tidningen l’”Humanité”",
+      "note": "apostrophe.md §3.3 cases 3 and 3a (spec 1.2.0): same as fi; U+201D is in CLOSEISH and the output is unchanged from spec 1.1.0."
     }
   ]
 }

data/lib/polytypo/data/locales/registry.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "$comment": "Locale resolution input. Algorithm is specified in spec/rules/locale-resolution.md and is identical in every runtime — never delegate it to a platform locale-negotiation library. \"spec\" here must track spec/VERSION exactly — it is not itself the global version source; scripts/validate-spec.mjs enforces the match.",
   "locales": ["en-US", "en-GB", "de-DE", "de-CH", "fr", "fr-CA", "ru", "fi", "sv", "el"],
   "aliases": {

data/lib/polytypo/data/rules/order.json

@@ -1,5 +1,5 @@
 {
-  "spec": "1.0.0",
+  "spec": "1.2.0",
   "$comment": "Single source of truth for pipeline order. Rules run in ascending `order`. Disabling a rule removes it from the sequence and never reorders the rest. Rule ids are public API (see docs/ARCHITECTURE.md section 5). \"spec\" here must track spec/VERSION exactly — it is not itself the global version source; scripts/validate-spec.mjs enforces the match.",
   "rules": [
     {
@@ -48,15 +48,15 @@
       "default": "on",
       "modes": ["text", "html", "markdown"],
       "localeData": ["quotes"],
-      "summary": "Straight quotes to locale primary/secondary pairs with nesting resolution. Runs before `apostrophe` so that a straight apostrophe is still available as quote evidence. Declines to pair an ambiguous medial short single-quoted span (spec 0.5.0) unless a cited `quotes.elisionIdioms` entry matches."
+      "summary": "Straight quotes to locale primary/secondary pairs with nesting resolution. Runs before `apostrophe` so that a straight apostrophe is still available as quote evidence. Declines to pair a medial single-quoted `n` (spec 1.1.0, every locale) or a span matching a cited `quotes.elisionIdioms` entry, so that `apostrophe` converts both marks instead."
     },
     {
       "id": "apostrophe",
       "order": 50,
       "default": "on",
       "modes": ["text", "html", "markdown"],
-      "localeData": ["quotes"],
-      "summary": "Remaining straight apostrophes to U+2019 without corrupting contractions. As of spec 0.5.0, also reads `quotes.elisionIdioms` (indirectly, via the shared ambiguity predicate) to keep a straight-ASCII mark `quotes` deliberately preserved from being independently curled by this rule's own structural case ladder."
+      "localeData": [],
+      "summary": "Remaining straight apostrophes to U+2019 without corrupting contractions. Reads no locale data: spec 0.5.0's `quotes.elisionIdioms` lookup existed only for the preserve set withdrawn in spec 1.1.0, so every mark `quotes` declined to claim now reaches this rule's ordinary structural case ladder."
     },
     {
       "id": "symbols",

data/lib/polytypo/engine/rules/apostrophe.rb

@@ -9,7 +9,7 @@ require_relative "../registry"
 module Polytypo
   module Engine
     module Rules
-      # spec/rules/apostrophe.md (spec 0.5.0), order 50.
+      # spec/rules/apostrophe.md (spec 1.2.0), order 50.
       #
       # Converts a straight U+0027 to U+2019 where it is genuinely an apostrophe: a contraction,
       # an elision, a possessive, or a decade elision. Runs immediately after `quotes` (order 40)
@@ -17,9 +17,10 @@ module Polytypo
       # replacing one code point; the rule never inserts, never deletes, and never touches U+2019
       # itself.
       #
-      # As of spec 0.5.0 this rule additionally skips a small, precisely-defined set of positions
-      # entirely (apostrophe.md 3.4) -- the shared ambiguous-medial-span preserve set,
-      # QuoteAmbiguity.compute_preserve_indices -- rather than applying its case ladder to them.
+      # As of spec 1.1.0 this rule reads no locale data and skips no position (apostrophe.md
+      # 3.4). Spec 0.5.0's preserve set existed to stop the case ladder from converting the marks
+      # quotes had vetoed; conversion is now the specified outcome for exactly those marks --
+      # cases 4 and 3 are what turn `rock 'n' roll` into `rock ’n’ roll`, in every locale.
       module Apostrophe
         SQ = 0x27
         RIGHT_SINGLE = 0x2019
@@ -46,6 +47,10 @@ module Polytypo
           0x2D, 0x2011, 0x2013, 0x2014
         ].freeze
 
+        # OPENQUOTE is apostrophe.md 3.1 OPENQUOTE (spec 1.2.0): the quotation glyphs of OPENISH,
+        # without its brackets, its dashes and Engine::MARKER. Read only by case 3a.
+        OPENQUOTE = [0xAB, 0x2018, 0x201A, 0x201B, 0x201C, 0x201E, 0x201F, 0x2039].freeze
+
         BREAK = [0x0A, 0x0D, 0x0B, 0x0C, 0x85, 0x2028, 0x2029, Engine::LINE_MARKER].freeze
 
         # SPACELIKE, including Engine::LINE_MARKER as a member of BREAK for every rule everywhere
@@ -80,6 +85,11 @@ module Polytypo
             return true
           end
 
+          # Case 3a (spec 1.2.0) -- elision before a quotation: `d'« urine »`, `l'“idea”`. Case 3
+          # misses it because an opening quotation glyph is not in CLOSEISH; `f'(x)` stays straight
+          # because a bracket is not in OPENQUOTE.
+          return true if UnicodeUtil.letter?(left) && OPENQUOTE.include?(right)
+
           # Case 4 -- leading elision: `'90s`, `'tis`, `'em`, `'n'` (the leading mark). The
           # replacement is U+2019, never U+2018 -- a leading elision is a raised comma, not an
           # opening quotation mark, and `quotes` has already had its chance to claim the mark as a
@@ -92,20 +102,12 @@ module Polytypo
           false
         end
 
-        def self.scan(cp, locale_data, _ctx)
+        def self.scan(cp, _locale_data, _ctx)
           n = cp.length
-          # apostrophe.md 3.4, spec 0.5.0: positions in the shared ambiguous-medial-span preserve
-          # set (an ambiguous shape with no cited elisionIdioms match) are skipped entirely,
-          # before left/right are even read -- this rule's own case ladder would otherwise curl
-          # both marks of e.g. `rock 'n' roll` independently and silently defeat quotes'
-          # deliberate veto.
-          idioms = (locale_data["quotes"] || {})["elisionIdioms"] || []
-          preserve = QuoteAmbiguity.compute_preserve_indices(cp, idioms)
 
           edits = []
           (0...n).each do |i|
             next unless cp[i] == SQ
-            next if preserve.key?(i)
 
             left = QuoteAmbiguity.at(cp, i - 1)
             right = QuoteAmbiguity.at(cp, i + 1)

data/lib/polytypo/engine/rules/nbsp.rb

@@ -9,7 +9,7 @@ require_relative "../../errors"
 module Polytypo
   module Engine
     module Rules
-      # `nbsp` -- spec/rules/nbsp.md (spec 0.6.0), order 70 (last).
+      # `nbsp` -- spec/rules/nbsp.md (spec 1.2.0), order 70 (last).
       #
       # Ten sub-rules, N1 through N10, evaluated in that fixed order (3.2); each produces
       # candidate edits keyed by the index of the space (or insertion point) it claims, and the
@@ -54,16 +54,7 @@ module Polytypo
         end
 
         # BREAK (nbsp.md 3.1), including LINE_MARKER -- a member of BREAK for every rule,
-        # everywhere (modes.md 3.2).
-        #
-        # MARKER is deliberately NOT a member here, and openish?/closeish? below do not add it
-        # either. modes.md 3.3's table says nbsp's OPENISH/CLOSEISH include the span-boundary
-        # MARKER, the same way quotes' and apostrophe's do -- but the JS reference
-        # implementation's own isBreak/isOpenish/isCloseish never test for MARKER, only
-        # LINE_MARKER via isBreak. That is a documented, preserved discrepancy (tracked, not
-        # resolved, in the roadmap), carried forward here unchanged rather than "corrected"
-        # against the table, for cross-runtime consistency with the already-shipped JS, Python
-        # and Go ports.
+        # everywhere (modes.md 3.2). MARKER is not a member of BREAK; see closeish? below.
         def self.break?(cp)
           cp == 0x0A || cp == 0x0D || cp == 0x0B || cp == 0x0C || cp == 0x85 ||
             cp == 0x2028 || cp == 0x2029 || cp == LINE_MARKER
@@ -197,13 +188,19 @@ module Polytypo
         end
 
         # OPENISH / CLOSEISH (nbsp.md 3.1): the ASCII brackets plus every locale quote glyph.
-        # See break?'s comment for why MARKER is not a member here.
+        #
+        # Since spec 1.2.0 the span boundary MARKER is a member of CLOSEISH and not of OPENISH
+        # (nbsp.md 3.1 and 7 item 12, modes.md 3.3). CLOSEISH is read only by N1/N2's right-context
+        # guard, where membership lets French `<strong>gel :</strong>` keep its no-break space
+        # after `spaces` deletes the U+0020. OPENISH stays marker-free: with the marker in it,
+        # N1/N2's quote-glyph guard would decline `<em>non</em> ! Oui`, and the left-boundary
+        # tests of N3, N7, N9, N10 would widen at span edges.
         def self.openish?(prep, cp)
           prep.opens.include?(cp)
         end
 
         def self.closeish?(prep, cp)
-          prep.closes.include?(cp)
+          cp == MARKER || prep.closes.include?(cp)
         end
 
         def self.mark?(prep, cp)

data/lib/polytypo/engine/rules/quote_ambiguity.rb

@@ -26,14 +26,17 @@ module Polytypo
       # the right one as a trailing possessive/elision, apostrophe.md 3.3 cases 3/4) must not
       # convert them either.
       module QuoteAmbiguity
-        # STRAIGHT_APOSTROPHE is the GENERAL ambiguous-shape veto's own trigger glyph -- U+0027
-        # only. An already-curly U+2018/U+2019 pair is out of this predicate's scope by
-        # construction.
-        STRAIGHT_APOSTROPHE = 0x27
-
-        # NARROW -- quotes.md 3.1 NARROW -- every glyph an elision idiom's marks may appear as
-        # across pipeline passes (straight, or already curled by an earlier pass). Shared with
-        # quotes.rb so the two rules cannot define two slightly different NARROW sets.
+        # LOWER_N / UPPER_N -- quotes.md 3.2's one enclosed code point, in either case.
+        LOWER_N = 0x6E
+        UPPER_N = 0x4E
+
+        # NARROW -- quotes.md 3.1 NARROW -- every glyph an elision mark may appear as across
+        # pipeline passes (straight, or already curled by an earlier pass). Shared with quotes.rb
+        # so the two rules cannot define two slightly different NARROW sets. For the universal
+        # medial-n veto, matching the whole class is an IDEMPOTENCY obligation rather than a
+        # preference: its marks are converted to U+2019 by apostrophe, so a straight-ASCII-only
+        # predicate would not recognise its own output and pass 2 would pair `rock ’n’ roll`
+        # as an ordinary NARROW quotation on the next run.
         NARROW = [0x27, 0x2018, 0x2019, 0x201A, 0x201B, 0x2039, 0x203A].freeze
 
         # INLINE_SPACE -- quotes.md 3.1 INLINE-SPACE, deliberately excluding BREAK/MARKER/
@@ -41,9 +44,6 @@ module Polytypo
         # same anchor the elisionIdioms matcher already uses.
         INLINE_SPACE = [0x20, 0x09, 0xA0, 0x202F, 0x2007, 0x2009, 0x200A].freeze
 
-        MIN_ENCLOSED = 1
-        MAX_ENCLOSED = 3
-
         def self.narrow?(cp)
           NARROW.include?(cp)
         end
@@ -187,28 +187,27 @@ module Polytypo
 
         # compute_ambiguous_shape_indices is the general ambiguous-medial-span shape,
         # locale-independent (quotes.md 3.2, spec 0.5.0): a pair of straight ASCII single quotes
-        # (U+0027 only) enclosing 1-3 LETTER code points, with at least one INLINE-SPACE code
-        # point immediately outside each mark. Both mark positions are returned for every match.
-        # A superset of compute_idiom_matched_indices's output whenever an idiom's elided field
-        # is itself 1-3 letters (true of every idiom shipped so far), but computed independently
-        # rather than assumed, since a future idiom's elided field is not required to be that
-        # short.
+        # (spec 1.1.0) a pair of NARROW marks enclosing exactly one code point, U+006E or
+        # U+004E, with at least one INLINE-SPACE code point immediately outside each mark. Both
+        # mark positions are returned for every match. A superset of
+        # compute_idiom_matched_indices's output for every idiom whose elided field is a single
+        # n (true of every idiom shipped so far), but computed independently rather than assumed,
+        # since a future idiom's elided field is not required to be that short.
         def self.compute_ambiguous_shape_indices(cp)
           ambiguous = {}
           n = cp.length
 
           (0...n).each do |i|
-            next unless at(cp, i) == STRAIGHT_APOSTROPHE
+            next unless narrow?(at(cp, i))
 
             l_lit = at(cp, i - 1)
             next if l_lit == Engine::NONE || !inline_space?(l_lit)
 
-            k = 0
-            k += 1 while k < MAX_ENCLOSED && UnicodeUtil.letter?(at(cp, i + 1 + k))
-            next if k < MIN_ENCLOSED
+            enclosed = at(cp, i + 1)
+            next unless [LOWER_N, UPPER_N].include?(enclosed)
 
-            j = i + 1 + k
-            next unless at(cp, j) == STRAIGHT_APOSTROPHE
+            j = i + 2
+            next unless narrow?(at(cp, j))
 
             r_lit = at(cp, j + 1)
             next if r_lit == Engine::NONE || !inline_space?(r_lit)
@@ -219,22 +218,6 @@ module Polytypo
 
           ambiguous
         end
-
-        # compute_preserve_indices is the set of straight-ASCII-quote index positions
-        # apostrophe.md 3.4 requires `apostrophe` to skip byte-identically: ambiguous-shaped, but
-        # with no matching cited idiom. A position with a matching idiom is not in this set --
-        # apostrophe's ordinary case ladder still curls it, exactly as spec 0.4.0-0.4.1 did.
-        def self.compute_preserve_indices(cp, idioms)
-          ambiguous = compute_ambiguous_shape_indices(cp)
-          return ambiguous if ambiguous.empty?
-
-          idiom_matched = compute_idiom_matched_indices(cp, idioms)
-          return ambiguous if idiom_matched.empty?
-
-          ambiguous.each_with_object({}) do |(idx, _), preserve|
-            preserve[idx] = true unless idiom_matched.key?(idx)
-          end
-        end
       end
     end
   end

data/lib/polytypo/engine/rules/spaces.rb

@@ -107,10 +107,17 @@ module Polytypo
         # after any edit had been applied would break the Chicago spaced ellipsis
         # "Hello . . .", where every dot is a lone dot at decision time and all three spaces
         # must still strip.
+        #
+        # Spec 1.2.0's word-start clause: a single dot directly followed by a letter or an ASCII
+        # digit starts a token (".NET", ".env", ".5"), so "Use .NET" keeps its space. A span
+        # boundary marker after the dot is neither, so "a .</em>" still strips.
         def self.lone_dot?(cp, e)
           return true if at(cp, e) != FULL_STOP
 
-          !dotlike?(at(cp, e + 1))
+          after = at(cp, e + 1)
+          return false if UnicodeUtil.letter?(after) || digit_ascii?(after)
+
+          !dotlike?(after)
         end
 
         def self.digit_ascii?(value)

data/lib/polytypo/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Polytypo
-  VERSION = "1.0.1"
+  VERSION = "1.2.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: polytypo
 version: !ruby/object:Gem::Version
-  version: 1.0.1
+  version: 1.2.0
 platform: ruby
 authors:
 - Iurii Rogulia