PDM LLM transcription eval

PDM, sample 25 (offset 0), generated 2026-07-16T13:09:27.971Z. Prompt versions: transcribe v3, expand v3, transliterate v2.

Each entry shows the scanned group, then the researcher's version and the LLM's version of each step, with what the LLM missed struck out on the researcher line and what it added or changed highlighted on the LLM line. Where the LLM was unsure it wrote [a|b] (either reading) or ⁇ (illegible) instead of guessing — those count as right if either reading matches. Two scores: STRICT (every character) and LENIENT (punctuation and letter-case ignored - closer to substance; apostrophes still count, they are part of the Listuguj orthography). Both are lower bounds: a different-but-valid synonym or an error in the researcher row still scores as a difference. The JUDGE row under each step classifies every residual difference (punctuation-only / valid alternative / LLM error / researcher error / unclear) and gives an equivalence-in-substance score - the judge consults the scanned image where the versions disagree. Judge classifications are themselves model output: treat them as pre-structured questions for review, not verdicts.

Summary

step	scored	strict similarity	lenient similarity	judged equivalence	mean confidence
Transcription	24	80%	82%	77	51
Expanded	7	78%	78%	66	60
Transliteration	24	60%	62%	66	63

What the differences actually are (every judged difference, all steps): 27 punctuation/format only 31 unclear 21 researcher errors 52 LLM errors 21 valid alternatives

Confidence calibration (does the model's self-reported confidence predict how good the result actually is?):

confidence	results	mean similarity
0–39	6	53%
40–59	27	69%
60–79	17	77%
80–100	5	82%

The rules (the instructions given to the model)

Transcription (v3) — click to expand

You are reading ONE ENTRY cropped from Father Pacifique's handwritten
Mi'gmaq-French dictionary manuscript (early 1900s).  The text mixes
Mi'gmaq words (in Pacifique's own orthography) with French glosses, and
uses heavy abbreviation.

IMPORTANT: neighbouring entries' text has been MASKED OUT (blanked to
white) - only this entry's marked fragments are visible.  The visible
fragments, read left-to-right then top-to-bottom, ARE the whole entry;
white gaps between them are masking, not missing text.  Never return an
empty result while any handwriting is visible.

Transcribe the handwriting LETTER-BY-LETTER, exactly as written:
- Keep every abbreviation, punctuation mark, diacritic (ā, ŏ, ê, ...),
  parenthesis and reference citation (e.g. "(Pi. Met.)", "(Jo. 368)")
  exactly as it appears.
- Do NOT expand abbreviations, translate, or normalize spelling.
- Preserve the reading order; keep the punctuation actually written,
  attached to its run.
- Return the text as an ordered list of RUNS, each tagged with its
  language: "mm" for Mi'gmaq words (Pacifique's orthography), "fr" for
  French, "cit" for citations/references/section marks (e.g.
  "(Pi. Met.)", "(Jo. 368)").  Split runs exactly at language changes.
If you are genuinely unsure between readings, DO NOT silently pick one:
write the alternatives in square brackets separated by | (e.g. "pi[l|i]ei"
means the letter could be l or i).  Use ⁇ for a truly illegible character.
Use these sparingly - only where you are actually unsure.

Also return "confidence": an integer 0-100, your overall confidence that
your text is correct (100 = certain).  Be honest - this number is used to
decide which results need human review.

Expanded (v3) — click to expand

The image is ONE ENTRY from Father Pacifique's handwritten Mi'gmaq-French
dictionary manuscript (neighbouring entries are masked to white - the
visible fragments are the whole entry).  Below is a letter-by-letter
transcription of it.
Rewrite the transcription with the abbreviations EXPANDED, changing
nothing else:
- Expand French abbreviations in place (e.g. "milieu d. l. n." ->
  "milieu de la nuit", "p.ê." -> "peut être").
- Pacifique often writes a family of related Mi'gmaq words and elides the
  repeated stem, giving only the changed ending (e.g. "aptepogoei, goet"
  means the second word is "aptepogoet").  Restore the full word for each
  elided form, taking the stem from the neighbouring full words.
- Keep the original spelling/orthography of the Mi'gmaq words; keep
  citations like "(Pi. Met.)" unchanged; do NOT translate.
- The transcription may contain [a|b] ambiguity markers or ⁇ for
  illegible characters.  Consult the image: where the context or the ink
  resolves the ambiguity, write the resolved reading; where it does not,
  keep the marker.
- Keep the language-tagged RUN structure ("mm" Mi'gmaq / "fr" French /
  "cit" citation, in reading order); correct a run's tag if the
  transcription mis-tagged it.
If you are genuinely unsure between readings, DO NOT silently pick one:
write the alternatives in square brackets separated by | (e.g. "pi[l|i]ei"
means the letter could be l or i).  Use ⁇ for a truly illegible character.
Use these sparingly - only where you are actually unsure.

Also return "confidence": an integer 0-100, your overall confidence that
your text is correct (100 = certain).  Be honest - this number is used to
decide which results need human review.


Transcription (tagged runs):
⟨the transcription runs go here⟩⟨mm⟩

Transliteration (v2) — click to expand

Below is the expanded transcription of one entry from Father Pacifique's
handwritten Mi'gmaq-French dictionary, as language-tagged runs
(word⟨mm⟩ = Mi'gmaq in Pacifique's orthography, word⟨fr⟩ = French,
word⟨cit⟩ = citation/reference).  Produce the modern working version as
ONE flat text, keeping the order and punctuation:
- ⟨fr⟩ runs: translate the French into English.
- ⟨mm⟩ runs: convert from Pacifique's orthography into the Listuguj
  orthography.  Corpus-attested correspondences (Pacifique -> Listuguj):
    aposgigen -> apusqi'gn;  pitoigatem -> pituigatm;
    agtatpag -> aqtatpa'q;  mosgoatem -> musgwatm;
    mosgŏmg -> musqomg;  aptepogoei -> apt'puguei;
    eoltjeoetji -> ewuljewe'ji;  pepgoitjetegei -> pepguijete'gei;
    nenatoigtjemsit -> nenatuigjemsit;  nāntemigtjemg -> na'ntemigjemg.
  Recurring patterns in those pairs: o before a consonant usually -> u;
  tj -> j; g before t/p often -> q; ā -> a'; oe -> we/ue; word-final
  -em -> -m (vowel dropped); an unstressed initial vowel may take an
  apostrophe (aptepogoei -> apt'puguei).
- ⟨cit⟩ runs: copy VERBATIM, unchanged (citations like "(Pi. Met.)",
  "(Jo. 368)" are references, never translated).
- If the input contains [a|b] ambiguity markers or ⁇, carry the
  uncertainty through (resolve it only if the target orthography makes
  one reading clearly right).
If you are genuinely unsure between readings, DO NOT silently pick one:
write the alternatives in square brackets separated by | (e.g. "pi[l|i]ei"
means the letter could be l or i).  Use ⁇ for a truly illegible character.
Use these sparingly - only where you are actually unsure.

Also return "confidence": an integer 0-100, your overall confidence that
your text is correct (100 = certain).  Be honest - this number is used to
decide which results need human review.


Expanded transcription (tagged runs):
⟨the expanded runs go here⟩⟨mm⟩

Entries

1. ref 70411755521 (group 221678)

Transcription similarity 88% / lenient 90% confidence 55
researcher	aposgigen, (P. Met.), clef
LLM	aposgigen (P. Mr.), clef aposgigen ⟨mm⟩ (P. Mr.), ⟨cit⟩ clef⟨fr⟩
judge	equivalence 92 punctuation “aposgigen,” vs “aposgigen” — Version A adds a comma after the headword; punctuation only. unclear “Met.” vs “Mr.” — The abbreviation is hard to read; could be 'Met.' or 'Mr.', cannot definitively tell.
Expanded similarity — confidence 82
researcher	(none - not yet done by hand)
LLM	aposgigen (P. Mr.), clef aposgigen⟨mm⟩ (P. Mr.),⟨cit⟩ clef⟨fr⟩
Transliteration similarity 92% / lenient 89% confidence 92
researcher	apusqi'gn (P. Met.), key
LLM	apusqi'gn (P. Mr.), key
judge	equivalence 90 researcher-error “P. Met.” vs “P. Mr.” — The image shows 'Mr.' (superscript r), matching version B; version A's 'Met.' misreads it.

2. ref 2479001625590 (group 219703)

Transcription similarity 94% / lenient 97% confidence 55
researcher	eoltjeoetji, pauvre petit, chétif.
LLM	eoltjeoêtji , pauvre petit, chétif. eoltjeoêtji⟨mm⟩ , pauvre petit, chétif.⟨fr⟩
judge	equivalence 90 llm-error “eoltjeoetji” vs “eoltjeoêtji” — The handwriting shows a plain 'e', not a circumflexed 'ê', so B introduces an incorrect diacritic in the Mi'gmaq word. punctuation “eoltjeoetji,” vs “eoltjeoêtji ,” — B adds a space before the comma, a spacing-only difference.
Expanded similarity — confidence 82
researcher	(none - not yet done by hand)
LLM	eoltjeoêtji , pauvre petit, chétif. eoltjeoêtji⟨mm⟩ , pauvre petit, chétif.⟨fr⟩
Transliteration similarity 72% / lenient 76% confidence 88
researcher	ewuljewe'ji, poor little frail one
LLM	ewuljewe'ji , poor little one, puny.
judge	equivalence 95 punctuation “ewuljewe'ji” vs “ewuljewe'ji” — Same Mi'gmaq word; only spacing before comma differs. valid-alternative “poor little frail one” vs “poor little one, puny” — French 'chétif' rendered as 'frail' vs 'puny'; both are defensible translations.

3. ref 2588291230184 (group 220132)

Transcription similarity 62% / lenient 61% confidence 32
researcher	(7 pr 101) aolamsenoigtog
LLM	( ) pu 101 aŏlŏm renŏigtŏg ( ) ⟨cit⟩ pu 101⟨cit⟩ aŏlŏm ⟨mm⟩ renŏigtŏg⟨mm⟩
judge	equivalence 80 unclear “(7” vs “( )” — The bracketed mark is ambiguous; A reads a '7', B reads an empty paren. valid-alternative “pr” vs “pu” — Abbreviation reading pr vs pu is ambiguous in the script. valid-alternative “aolam” vs “aŏlŏm” — B uses breve diacritics over vowels, a plausible rendering of the same vowels. unclear “senoigtog” vs “renŏigtŏg” — Initial letter read as s vs r; the cursive loop is ambiguous.
Expanded similarity — confidence 42
researcher	(none - not yet done by hand)
LLM	( ) pu 101 aŏlŏm renŏigtŏg ( )⟨cit⟩ pu 101⟨cit⟩ aŏlŏm⟨mm⟩ renŏigtŏg⟨mm⟩
Transliteration similarity 35% / lenient 41% confidence 45
researcher	awlams'nuigtug
LLM	( ) pu 101 awlum renuigtoq
judge	equivalence 30 llm-error “awlams'nuigtug” vs “awlum” — Version B splits and misrenders the word; the header 'aŏlŏm' corresponds to A's 'awlams'-' beginning, but B drops the connection to the following morpheme. punctuation (absent) vs “( ) pu 101” — Version B retains the reference/page marker from the evidence; not part of the translated entry, no substantive content difference. llm-error “awlams'nuigtug” vs “renuigtoq” — Evidence 'renŏigtŏg' should convert to '-nuigtug'; B reads leading 'ren' and ends in 'toq' instead of proper orthography 'tug'.

4. ref 4714485949303 (group 218558)

Transcription similarity 76% / lenient 75% confidence 42
researcher	presser/passer la nuit, mètenemai;
LLM	presse la nuit, mètenemai; presse⟨fr⟩ la nuit,⟨fr⟩ mètenemai;⟨mm⟩
judge	equivalence 85 llm-error “presser/passer” vs “presse” — The image shows 'presser' (or 'passer'); version B reads 'presse' missing the final letter.
Expanded similarity — confidence 72
researcher	(none - not yet done by hand)
LLM	presse la nuit, mètenemai; presse⟨fr⟩ la nuit,⟨fr⟩ mètenemai;⟨mm⟩
Transliteration similarity 44% / lenient 47% confidence 55
researcher	spend the night (by the fire), me'te'nmai
LLM	hurry the night, metenemai;
judge	equivalence 55 valid-alternative “spend the night (by the fire)” vs “hurry the night” — French 'presse la nuit' is ambiguous; both are defensible translations, though A adds an interpretive gloss. llm-error “me'te'nmai” vs “metenemai” — The Listuguj orthography requires apostrophes (me'te'nmai); B omitted them and altered the vowel structure. punctuation (absent) vs “;” — Trailing semicolon present in the source, A dropped it.

5. ref 19091442321809 (group 216655)

Transcription similarity 89% / lenient 97% confidence 72
researcher	pepgetji, être mince, léger -(32, 1)
LLM	pepgetsi, être mince, léger (32,1)- pepgetsi,⟨mm⟩ être mince, léger⟨fr⟩ (32,1)-⟨cit⟩
judge	equivalence 90 llm-error “pepgetji” vs “pepgetsi” — The handwriting shows 'pepgetji' with a j, not 'pepgetsi'. punctuation “-(32, 1)” vs “(32,1)-” — Placement of dash and spacing inside parentheses differ but no content change.
Expanded similarity — confidence 95
researcher	(none - not yet done by hand)
LLM	pepgetsi, être mince, léger (32,1)- pepgetsi,⟨mm⟩ être mince, léger⟨fr⟩ (32,1)-⟨cit⟩
Transliteration similarity 71% / lenient 77% confidence 70
researcher	pepge'ji, to be thin, light
LLM	pepgetsi, to be thin, light (32,1)-
judge	equivalence 95 valid-alternative “pepge'ji” vs “pepgetsi” — The Listuguj orthography 'pepge'ji' is a valid conversion of 'pepgetsi'; A applied the orthography conversion as the task required, B kept the original spelling. punctuation (absent) vs “(32,1)-” — B retained the reference number/marginal notation from the source, which is metadata not content.

6. ref 25998998179493 (group 218910)

Transcription similarity 97% / lenient 100% confidence 58
researcher	natji losōgŏoi, aller visiter ses beaux-parents? aller y demeurer Pi. Met.
LLM	natji losōgŏoi, aller visiter ses beaux-parents ?. aller y demeurer Pi. Met. natji losōgŏoi,⟨mm⟩ aller visiter ses beaux-parents ?. aller y demeurer⟨fr⟩ Pi. Met.⟨cit⟩
judge	equivalence 100 punctuation “beaux-parents ?” vs “beaux-parents ?.” — Version B adds an extra period after the question mark; punctuation only.
Expanded similarity — confidence 88
researcher	(none - not yet done by hand)
LLM	natji losōgŏoi, aller visiter ses beaux-parents ?. aller y demeurer Pi. Met. natji losōgŏoi,⟨mm⟩ aller visiter ses beaux-parents ?. aller y demeurer⟨fr⟩ Pi. Met.⟨cit⟩
Transliteration similarity 76% confidence 62
researcher	najilu'sugowi, to go visit your in-laws? to go live with? Pi. Met.
LLM	natji lusuqwei, to go visit his in-laws?. to go live there Pi. Met.
judge	equivalence 70 unclear “najilu'sugowi” vs “natji lusuqwei” — The two orthographic renderings of the source 'losōgŏoi' differ substantially; hard to determine which correctly reflects Listuguj orthography from the handwriting. valid-alternative “to go visit your in-laws?” vs “to go visit his in-laws?” — French 'ses' can be rendered as his/her/your; both are defensible translations of the possessive. punctuation “?” vs “?.” — The source has '?.'; A dropped the period, B kept it - punctuation only. llm-error “to go live with?” vs “to go live there” — Source 'aller y demeurer' means 'to go live there'; B is correct and A's 'live with?' misrenders 'y' (there) as 'with'.

7. ref 28294850603427 (group 219128)

Transcription similarity 94% confidence 80
researcher	espĕg, il est haut
LLM	espêg, il est haut espêg,⟨mm⟩ il est haut⟨fr⟩
judge	equivalence 95 unclear “espĕg” vs “espêg” — The diacritic on the e differs (breve vs circumflex); the handwriting shows a mark over the e but which diacritic is intended is hard to determine.
Expanded similarity — confidence 95
researcher	(none - not yet done by hand)
LLM	espêg, il est haut espêg,⟨mm⟩ il est haut⟨fr⟩
Transliteration similarity 57% / lenient 55% confidence 80
researcher	espe'g, he/she/it is high/tall
LLM	espeg, it is high
judge	equivalence 80 llm-error “espe'g” vs “espeg” — The evidence has 'espêg' with a circumflex; the Listuguj orthography apostrophe (espe'g) reflects this, so the missing apostrophe in B is an error. valid-alternative “he/she/it is high/tall” vs “it is high” — Mi'gmaq third person can be rendered he/she/it, and 'haut' can mean high or tall; both are defensible translations.

8. ref 28403885750253 (group 218510)

Transcription similarity 87% / lenient 88% confidence 55
researcher	gis nogoag, il prend (fièvre?);
LLM	gis nogoay, il prend (pière?) gis⟨mm⟩ nogoay,⟨mm⟩ il prend (pière?)⟨fr⟩
judge	equivalence 80 unclear “nogoag” vs “nogoay” — The final letter of the Mi'gmaq word is ambiguous between g and y in the script. llm-error “fièvre?” vs “pière?” — The parenthetical word reads 'fièvre' (French for fever), not the nonstandard 'pière'. punctuation “;” vs (absent) — Trailing semicolon versus no terminal punctuation.
Expanded similarity — confidence 70
researcher	(none - not yet done by hand)
LLM	gis nogoay, il prend (pière?) gis nogoay,⟨mm⟩ il prend (pière?)⟨fr⟩
Transliteration similarity 40% / lenient 37% confidence 55
researcher	gisnu’gwa’q, it/he got (to a high temperature?);
LLM	gis nugway, he takes (stone?)
judge	equivalence 30 unclear “gisnu'gwa'q” vs “gis nugway” — The French evidence reads 'gis nogoay' as two words; A merges and applies Mi'gmaq orthography with apostrophes, B keeps two words closer to source spelling - hard to say which orthographic rendering is correct. llm-error “it/he got (to a high temperature?)” vs “he takes (stone?)” — French 'il prend' means 'he takes', matching B; A's 'it/he got' with 'high temperature' does not correspond to the source. llm-error “(to a high temperature?)” vs “(stone?)” — French 'pière' (pierre = stone) matches B's 'stone'; A's 'high temperature' is a mistranslation.

9. ref 35750964190593 (group 218920)

Transcription similarity 83% / lenient 85% confidence 45
researcher	natji togoāleg, aller le reveiller
LLM	natgi'toyŏŭleg , aller le reveiller natgi'toyŏŭleg⟨mm⟩ , ⟨fr⟩ aller le reveiller⟨fr⟩
judge	equivalence 80 unclear “natji” vs “natgi'” — The Mi'gmaq word's letters are hard to read; A reads 'natji' while B reads 'natgi'' with an apostrophe, and it is difficult to confirm which is correct. unclear “togoāleg” vs “toyŏŭleg” — The middle characters are ambiguous; A uses 'go' with a macron, B uses 'yŏŭ' with breves, hard to determine the correct reading. punctuation “reveiller” vs “reveiller” — Same word; the image shows a spelling that both transcribed as 'reveiller'. punctuation “,” vs “ , ” — Only spacing around the comma differs.
Expanded similarity — confidence 88
researcher	(none - not yet done by hand)
LLM	natgi'toyŏŭleg , aller le reveiller natgi'toyŏŭleg⟨mm⟩ , aller le reveiller⟨fr⟩
Transliteration similarity 55% / lenient 56% confidence 55
researcher	najitugwa'l'g, to go wake him/her
LLM	natgi'tuuleg , go wake him up with uncertainty markers: natgi'tuule[g\|q] , go wake him up
judge	equivalence 80 unclear “najitugwa'l'g” vs “natgi'tuule[g\|q]” — Both attempt the Mi'gmaq form; the source reads 'natgi'toyŏŭleg' and neither matches exactly, so it's hard to say which orthographic conversion is correct. valid-alternative “to go wake him/her” vs “go wake him up” — Both render the French 'aller le reveiller' equivalently; 'him/her' vs 'him up' are minor phrasing choices.

10. ref 37295888929095 (group 218851)

Transcription similarity 85% / lenient 88% confidence 55
researcher	lŏgsigan, log house (?? De l’anglais?)
LLM	lŏgsīgan, log house (im. de l'anglais?) lŏgsīgan,⟨mm⟩ log house⟨fr⟩ (im. de l'anglais?)⟨fr⟩
judge	equivalence 90 unclear “lŏgsigan” vs “lŏgsīgan” — The image shows a diacritic over the i; whether it is a breve or macron is ambiguous. researcher-error “??” vs “im.” — The image reads 'im.' (abbreviation), which A marked as illegible. punctuation “De” vs “de” — Capitalization difference only; image shows lowercase 'de'. punctuation “l’anglais” vs “l'anglais” — Apostrophe style difference only in French word.
Expanded similarity — confidence 80
researcher	(none - not yet done by hand)
LLM	lŏgsīgan, log house (imité de l'anglais?) lŏgsīgan,⟨mm⟩ log house⟨fr⟩ (imité de l'anglais?)⟨fr⟩
Transliteration similarity 58% / lenient 56% confidence 72
researcher	lo’qsi’gan, log house (?? De l’anglais?)
LLM	lugsi'gan, log house (imitated from English?)
judge	equivalence 80 valid-alternative “lo’qsi’gan” vs “lugsi'gan” — Two defensible Listuguj renderings of lŏgsīgan; both plausible transcriptions of the same word. researcher-error “?? De l’anglais?” vs “imitated from English?” — Evidence reads 'imité de l'anglais?'; A left the first word as '??' while B correctly translated it as 'imitated from'.

11. ref 43496499262409 (group 218604)

Transcription similarity 38% / lenient 50% confidence 18
researcher	battre le briquet, sagtai;
LLM	; buetteu a buguet-tejtai; ; buetteu a⟨fr⟩ buguet-tejtai;⟨mm⟩
judge	equivalence 15 llm-error “battre” vs “buetteu” — Version B misreads the French word 'battre' as a nonsense word 'buetteu'. llm-error “le” vs “a” — Version B misreads the article 'le' as 'a'. llm-error “briquet” vs “buguet” — Version B misreads 'briquet' as the nonsense 'buguet'. punctuation “, ” vs “-” — Comma vs hyphen separating French phrase from Mi'gmaq word. llm-error “sagtai” vs “tejtai” — Version B misreads the Mi'gmaq word 'sagtai' as 'tejtai'. punctuation (absent) vs “;” — Version B adds a leading semicolon not in A.
Expanded similarity — confidence 32
researcher	(none - not yet done by hand)
LLM	; buetteu a buguet-tejtai; ; buetteu a⟨mm⟩ buguet-tejtai;⟨mm⟩
Transliteration similarity 18% / lenient 23% confidence 32
researcher	use (strike) a flint to start a fire, to light a lamp, sagtai
LLM	; buetteu a buguet-tejtai;
judge	equivalence 10 llm-error “use (strike) a flint to start a fire, to light a lamp” vs “buetteu a buguet-tejtai” — Version B copied the Mi'gmaq/raw text instead of providing an English translation as required by the task. llm-error “sagtai” vs “tejtai” — Version B misreads the headword; the image and researcher indicate 'sagtai' not 'tejtai'.

12. ref 44075355142415 (group 218568)

Transcription similarity 80% / lenient 81% confidence 44
researcher	5. oeleiatigo, attitude mutuelle (bonne)
LLM	5. veleiatega , attitude mutuelle (hom.) 5. ⟨cit⟩ veleiatega⟨mm⟩ , ⟨fr⟩ attitude mutuelle ⟨fr⟩ (hom.)⟨cit⟩
judge	equivalence 70 llm-error “oeleiatigo” vs “veleiatega” — The word begins with 'oe' and ends in 'igo'; B misreads initial letter and vowels in a Mi'gmaq word. punctuation “,” vs “ , ” — Only spacing around the comma differs. llm-error “(bonne)” vs “(hom.)” — The parenthetical reads 'bonne' (feminine of 'good'), not the abbreviation 'hom.'.
Expanded similarity — confidence 55
researcher	(none - not yet done by hand)
LLM	5. veleiatega , attitude mutuelle (homm.) 5. ⟨cit⟩ veleiatega⟨mm⟩ , ⟨fr⟩ attitude mutuelle ⟨fr⟩ (homm.)⟨cit⟩
Transliteration similarity 79% / lenient 82% confidence 55
researcher	5. weleiati’gw, mutual attitude (good)
LLM	5. weleiatega , mutual attitude (homm.)
judge	equivalence 70 unclear “weleiati’gw” vs “weleiatega” — The handwriting shows 'veleiatego'-like form; researcher applied Listuguj orthography with apostrophe while B kept raw form, hard to verify exact ending. llm-error “(good)” vs “(homm.)” — The parenthetical 'homm.' (an abbreviation) was not translated/converted by B, whereas A rendered it; but note the image reads 'homm.' so A's 'good' may itself be an interpretation.

13. ref 49592942488526 (group 219579)

Transcription similarity 91% / lenient 95% confidence 52
researcher	doubler l'intérieur, pitoigatem; p.ê. ajouter, grandir, grossir
LLM	double l'intérieur , pitŏigaten ; p. ê. ajouter, grandir, grossir double⟨fr⟩ l'intérieur⟨fr⟩ , ⟨fr⟩ pitŏigaten⟨mm⟩ ; p. ê. ⟨fr⟩ ajouter, grandir, grossir⟨fr⟩
judge	equivalence 88 llm-error “doubler” vs “double” — The image shows 'doubler' with a final r; version B omitted it. llm-error “pitoigatem” vs “pitŏigaten” — The final letter reads 'm' not 'n'; the breve on o is a plausible diacritic reading but the last consonant is wrong. punctuation “;” vs “;” — Spacing around punctuation differs only.
Expanded similarity 93% / lenient 95% confidence 72
researcher	doubler l'intérieur, pitoigatem; peut être ajouter, grandir, grossir
LLM	double l'intérieur , pitŏigaten ; peut être ajouter, grandir, grossir double⟨fr⟩ l'intérieur⟨fr⟩ , ⟨fr⟩ pitŏigaten⟨mm⟩ ; peut être ⟨fr⟩ ajouter, grandir, grossir⟨fr⟩
judge	equivalence 90 llm-error “doubler” vs “double” — The image shows 'doubler' (infinitive), matching the other infinitives in the list; B dropped the final 'r'. llm-error “pitoigatem” vs “pitŏigaten” — The Mi'gmaq word ends in 'm' as shown in the image; B read a final 'n' incorrectly, and the breve is a formatting choice.
Transliteration similarity 68% / lenient 73% confidence 72
researcher	double the interior, pituigatm; perhaps, make bigger, make heavier
LLM	double the interior , pituigatm ; perhaps add, grow, enlarge
judge	equivalence 70 punctuation “perhaps” vs “perhaps” — Both render 'peut être' as 'perhaps'; identical. researcher-error “make bigger” vs “add” — Evidence lists three verbs: ajouter, grandir, grossir; B's 'add' correctly renders 'ajouter' which A omitted. valid-alternative “make bigger” vs “grow” — 'make bigger' and 'grow' are both defensible renderings of 'grandir'. llm-error “make heavier” vs “enlarge” — 'grossir' means to make/get bigger; 'enlarge' is closer than 'make heavier', but 'make heavier' is a weaker translation - both imperfect, though B's is more accurate here.

14. ref 51997072825681 (group 216564)

Transcription similarity 87% / lenient 91% confidence 45
researcher	peut-être le même sens dans matamgetjoig, abrégé par les blancs en Kejwick (age)
LLM	; peut-être le même sens dans matamgetsig , abrégé par les blancs en Ketfoigk (2ge ; peut-être⟨fr⟩ le même sens dans ⟨fr⟩ matamgetsig⟨mm⟩ , abrégé par les blancs en ⟨fr⟩ Ketfoigk⟨mm⟩ (2ge⟨cit⟩
judge	equivalence 70 punctuation “peut-être” vs “; peut-être” — Version B adds a leading semicolon and extra spacing, punctuation only. llm-error “matamgetjoig” vs “matamgetsig” — The image shows 'matamgetjoig'; B misreads letters (tj vs ts, missing 'o'). llm-error “Kejwick” vs “Ketfoigk” — The Mi'gmaq word is 'Kejwick'; B's 'Ketfoigk' misreads the letters. llm-error “(age)” vs “(2ge” — The image shows '(age'; B misreads the 'a' as '2' and omits closing paren.
Expanded similarity — confidence 55
researcher	(none - not yet done by hand)
LLM	; peut-être le même sens dans matamgetsig , abrégé par les blancs en Ketfoigk (2ge ; peut-être le même sens dans⟨fr⟩ matamgetsig⟨mm⟩ , abrégé par les blancs en⟨fr⟩ Ketfoigk⟨mm⟩ (2ge⟨cit⟩
Transliteration similarity 44% / lenient 47% confidence 55
researcher	possibly in the same direction Matamgejuig. Abbreviated to Kejwick by the whites.
LLM	; perhaps the same meaning in matamgetsig, abbreviated by the whites to Ketfoigk (2ge
judge	equivalence 88 valid-alternative “possibly in the same direction” vs “perhaps the same meaning” — 'sens' in French can mean 'direction' or 'meaning'; both are defensible translations. valid-alternative “Matamgejuig” vs “matamgetsig” — A converts to Listuguj orthography (ts->j) and capitalizes; both trace to the source 'matamgetsig'. valid-alternative “Abbreviated to Kejwick by the whites” vs “abbreviated by the whites to Ketfoigk” — Word order differs but meaning identical; 'Kejwick' vs 'Ketfoigk' - A applies Listuguj orthography to the placename. punctuation (absent) vs “(2ge” — B retained a fragment '(2ge' from the source that A dropped as incomplete/marginal.

15. ref 54452220495151 (group 217607)

Transcription similarity 80% / lenient 83% confidence 42
researcher	agtatpag milieu d. l. n.
LLM	ergtatpieg milieu d. l. n ergtatpieg⟨mm⟩ milieu d. l.⟨fr⟩ n⟨cit⟩
judge	equivalence 70 llm-error “agtatpag” vs “ergtatpieg” — The Mi'gmaq headword differs in letters; version B misreads the initial and adds an 'i', giving a wrong orthography. punctuation “n.” vs “n” — Trailing abbreviation period present in A but omitted in B.
Expanded similarity 86% confidence 55
researcher	agtatpag milieu de la nuit
LLM	ergtatpieg milieu de la nuit ergtatpieg⟨mm⟩ milieu de la nuit⟨fr⟩
judge	equivalence 70 unclear “agtatpag” vs “ergtatpieg” — The Mi'gmaq headword reading differs; the image script is ambiguous but the evidence line supports 'ergtatpieg', while A reads 'agtatpag'.
Transliteration similarity 90% / lenient 93% confidence 70
researcher	aqtatpa'q, middle of the night
LLM	aqtatpe'g middle of the night
judge	equivalence 85 unclear “aqtatpa'q” vs “aqtatpe'g” — The two versions differ in the vowel (a vs e) and final consonant (q vs g); the handwriting and evidence 'ergtatpieg' don't clearly resolve which orthographic form is correct. punctuation “aqtatpa'q,” vs “aqtatpe'g” — Researcher includes a comma separating headword and gloss; formatting only.

16. ref 62191851148884 (group 219587)

Transcription similarity 92% confidence 72
researcher	papgog? le temps où l'on ramasse l'écorce, juin (31)
LLM	pŏpgŏy? le temps où l'on ramasse l'écore, juin (31) pŏpgŏy?⟨mm⟩ le temps où l'on ramasse l'écore, juin⟨fr⟩ (31)⟨cit⟩
judge	equivalence 80 unclear “papgog” vs “pŏpgŏy” — The headword is hard to read; A reads 'papgog' while B reads 'pŏpgŏy' with breve diacritics, and it's unclear which matches the faint handwriting. llm-error “l'écorce” vs “l'écore” — The French word for bark is 'écorce'; B drops the 'c', which is a spelling error.
Expanded similarity — confidence 88
researcher	(none - not yet done by hand)
LLM	pŏpgŏy? le temps où l'on ramasse l'écore, juin (31) pŏpgŏy?⟨mm⟩ le temps où l'on ramasse l'écore, juin⟨fr⟩ (31)⟨cit⟩
Transliteration similarity 58% / lenient 55% confidence 72
researcher	papgug?/papqu'g the time when the bark is collected, June (31)
LLM	pupguy? the time when one gathers bark, June (31)
judge	equivalence 90 valid-alternative “papgug?/papqu'g” vs “pupguy?” — Both attempt to render pŏpgŏy?; A offers a French-style and Listuguj form, B keeps closer to the evidence spelling - defensible variants. valid-alternative “the bark is collected” vs “one gathers bark” — Both faithfully translate 'l'on ramasse l'écore' with equivalent phrasings.

17. ref 62304771631303 (group 219974)

Transcription similarity 92% / lenient 91% confidence 72
researcher	pastogǒpetjit, morse, vache marine ?
LLM	pastŏgŏpetsit, morse, vache marine ? pastŏgŏpetsit,⟨mm⟩ morse,⟨fr⟩ vache marine ?⟨fr⟩
judge	equivalence 88 unclear “pastogǒpetjit” vs “pastŏgŏpetsit” — The two differ in the vowel diacritic on the first 'o' (g vs ŏ) and in 'tj' vs 'ts'; the handwriting is ambiguous and both are plausible readings. unclear “gǒ” vs “gŏ” — Version A reads plain 'g' with breve on following o, B places breve on both o's; hard to determine from script. unclear “tj” vs “ts” — The cluster before 'it' could be 'tj' or 'ts'; the handwriting does not clearly disambiguate.
Expanded similarity — confidence 88
researcher	(none - not yet done by hand)
LLM	pastŏgŏpetsit, morse, vache marine ? pastŏgŏpetsit,⟨mm⟩ morse, vache marine ?⟨fr⟩
Transliteration similarity 94% / lenient 96% confidence 80
researcher	pastugopejit, walrus, sea cow ?
LLM	pastugupejit, walrus, sea cow?
judge	equivalence 97 unclear “pastugopejit” vs “pastugupejit” — Difference is o vs u in the fourth syllable; the evidence 'pastŏgŏpetsit' uses ŏ which could support either transliteration, so it is hard to say which is correct. punctuation “sea cow ?” vs “sea cow?” — Only spacing before the question mark differs.

18. ref 92615040038436 (group 221471)

Transcription similarity 82% / lenient 80% confidence 55
researcher	aptepogoei, goet (uoli for Micmacs) noyé (Pi. Met.)
LLM	aptejŏgŏei, goet (note par Micmac) noyé (Pi. Met.) aptejŏgŏei,⟨mm⟩ goet (note par Micmac)⟨fr⟩ noyé⟨fr⟩ (Pi. Met.)⟨cit⟩
judge	equivalence 80 unclear “aptepogoei” vs “aptejŏgŏei” — The Mi'gmaq headword differs in letters (p vs j, o vs ŏ); handwriting is ambiguous but the diacritics vs plain letters are a real orthographic divergence. llm-error “uoli” vs “note” — The image reads 'uoli'/'noli'-like script, not the French 'note'; B likely normalized to a French word. unclear “for” vs “par” — The parenthetical word could be read either way from the cursive; both are plausible. llm-error “Micmacs” vs “Micmac” — The image shows a final flourish suggesting plural 'Micmacs' which B dropped.
Expanded similarity 79% / lenient 76% confidence 72
researcher	aptepogoei, aptepogoet (uoli for Micmacs) noyé (Pi. Met.)
LLM	aptejŏgŏei, aptejŏgŏet (note par Micmac) noyé (Pi. Met.) aptejŏgŏei, aptejŏgŏet⟨mm⟩ (note par Micmac)⟨fr⟩ noyé⟨fr⟩ (Pi. Met.)⟨cit⟩
judge	equivalence 70 llm-error “aptepogoei” vs “aptejŏgŏei” — The evidence and image show 'aptejŏgŏei' with a j; researcher's 'aptepogoei' misreads the letters, but wait—need to assess which matches image; the diacritics in B match the evidence. researcher-error “aptepogoet” vs “aptejŏgŏet” — Evidence shows the stem 'aptejŏgŏ-'; researcher's 'aptepogoet' lacks the j and diacritics found in the evidence. researcher-error “uoli” vs “note” — The evidence reads 'note'; researcher's 'uoli' misreads this word. researcher-error “for” vs “par” — Evidence reads French 'par'; researcher wrote English 'for', a misreading. unclear “Micmacs” vs “Micmac” — Evidence shows 'Micmac'; the image ending is ambiguous whether an 's' follows.
Transliteration similarity 88% / lenient 87% confidence 72
researcher	apt'puguei, apt'puguet (uoli for Micmacs) drowned (Pi. Met.)
LLM	apt'puguei, apt'puguet (note by Micmac) drowned (Pi. Met.)
judge	equivalence 75 researcher-error “uoli” vs “note” — The evidence and image show 'note par Micmac'; version A rendered it as 'uoli for' which misreads the French word 'note'. researcher-error “for” vs “by” — The parenthetical translates French 'par' as 'by' (version B) which is correct; 'for' is wrong. valid-alternative “Micmacs” vs “Micmac” — Evidence has 'Micmac' singular; both are minor and defensible readings of the source.

19. ref 108302894874177 (group 216946)

Transcription similarity 62% / lenient 60% confidence 42
researcher	pepgoitjetag, tambour
LLM	pépyoittelâg, tombow pépyoittelâg,⟨mm⟩ tombow⟨fr⟩
judge	equivalence 35 llm-error “pepgoitjetag” vs “pépyoittelâg” — Automated version misreads multiple letters (g→y, tj→tt, t→l) and adds diacritics not in the Mi'gmaq orthography. llm-error “tambour” vs “tombow” — The French translation 'tambour' (drum) is misread as the non-word 'tombow'.
Expanded similarity — confidence 72
researcher	(none - not yet done by hand)
LLM	pépyoittelâg, tombeau pépyoittelâg,⟨mm⟩ tombeau⟨fr⟩
Transliteration similarity 55% / lenient 53% confidence 55
researcher	pepgwijeta'q, drum
LLM	pepguitte'la'q, tomb
judge	equivalence 30 unclear “pepgwijeta'q” vs “pepguitte'la'q” — The French source 'tombeau' (tomb/grave) suggests B's orthography; hard to verify exact Mi'gmaq spelling from the handwriting, but the two differ in letters and apostrophe placement. llm-error “drum” vs “tomb” — The French evidence is 'tombeau' meaning tomb/grave, so B's 'tomb' is correct and A's 'drum' is the mistranslation.

20. ref 112778112369284 (group 218713)

Transcription similarity 91% / lenient 90% confidence 55
researcher	mosgoatem, mosgŏmg, gŏmit, léche
LLM	mosgoatemi, mosgŏmg, gŏmit, l'écho mosgoatemi, mosgŏmg, gŏmit,⟨mm⟩ l'écho⟨fr⟩
judge	equivalence 80 llm-error “mosgoatem” vs “mosgoatemi” — The image shows 'mosgoatem' with no trailing 'i'; version B added an extra letter. unclear “léche” vs “l'écho” — The handwriting reads 'léche'/'l'éche'; B's 'l'écho' differs in the final vowel and apostrophe, but the reading is ambiguous.
Expanded similarity 83% / lenient 81% confidence 72
researcher	mosgoatem, mosgŏmg, mosgŏmit, léche
LLM	mosgoatemi, mosgŏmg, gŏmit, l'écho mosgoatemi, mosgŏmg, gŏmit,⟨mm⟩ l'écho⟨fr⟩
judge	equivalence 70 researcher-error “mosgoatem” vs “mosgoatemi” — Image shows a final i on the word, matching the evidence and version B. llm-error “mosgŏmit” vs “gŏmit” — Image shows 'gŏmit'; version A appears to have added 'mos', while B matches the evidence, so B is closer though the leading char reading differs. unclear “léche” vs “l'écho” — Image reads 'l'écho'-like; A's 'léche' differs but the handwriting is ambiguous.
Transliteration similarity 71% / lenient 68% confidence 72
researcher	musgwatm, musqomg, musqomit, lick
LLM	musgwatm, musqomg, qomit, the echo
judge	equivalence 55 llm-error “musqomit” vs “qomit” — The evidence reads 'gŏmit'; version A's 'musqomit' follows the preceding 'musqomg' pattern but the source shows 'gŏmit', so B's 'qomit' is closer though missing initial; A appears to have added 'mus' not in source, yet B drops the 'mus' too - the source word is 'gŏmit' = qomit, so B is correct here. llm-error “lick” vs “the echo” — French evidence 'l'écho' means 'the echo'; version A's 'lick' is a misreading/mistranslation, so B is correct.

21. ref 113170399083509 (group 216765)

Transcription similarity 92% confidence 61
researcher	pepgoitjetegei, tegetitjig, batter du tambou - pepgetjetegetieg (Jo. 368)
LLM	pepgoitetegei, tegetitjig, battre du tambour pepgettetegetieg (Jo. 368) pepgoitetegei,⟨mm⟩ tegetitjig,⟨mm⟩ battre du tambour⟨fr⟩ pepgettetegetieg⟨mm⟩ (Jo. 368)⟨cit⟩
judge	equivalence 80 llm-error “pepgoitjetegei” vs “pepgoitetegei” — Version B omits the 'j' letter (pepgoitjetegei vs pepgoitetegei), a missing letter in the Mi'gmaq word. researcher-error “batter” vs “battre” — The French verb is 'battre' (to beat); Version A's 'batter' is a misspelling. researcher-error “tambou” vs “tambour” — The French word is 'tambour' (drum); Version A dropped the final 'r'. llm-error “- pepgetjetegetieg” vs “pepgettetegetieg” — Version B reads 'tt' where A reads 'tj'; the Mi'gmaq form 'pepgetjetegetieg' is consistent with the headword's 'tj'.
Expanded similarity 85% confidence 72
researcher	pepgoitjetegei, pepgoitegetitjig, batter du tambou - pepgetjetegetieg (Jo. 368)
LLM	pepgoitetegei, tegetitjig, battre du tambour pepgettetegetieg (Jo. 368) pepgoitetegei, tegetitjig,⟨mm⟩ battre du tambour⟨fr⟩ pepgettetegetieg⟨mm⟩ (Jo. 368)⟨cit⟩
judge	equivalence 65 researcher-error “pepgoitjetegei” vs “pepgoitetegei” — The evidence and image support 'pepgoitetegei'; A inserted an extra 'j' not in the source. researcher-error “pepgoitegetitjig” vs “tegetitjig” — Evidence shows 'tegetitjig'; A appears to have expanded/altered it to 'pepgoitegetitjig'. researcher-error “batter du tambou” vs “battre du tambour” — French 'battre du tambour' (evidence) is correct; A's 'batter du tambou' has spelling errors. researcher-error “pepgetjetegetieg” vs “pepgettetegetieg” — Evidence reads 'pepgettetegetieg'; A altered the letters.
Transliteration similarity 80% / lenient 81% confidence 55
researcher	pepguijete'gei, pepguiteg'tijig, beat the drum - pepgejete'g'tieg (Jo. 368)
LLM	pepguijete'gei, tegetijig, to beat the drum pepgettete'getieg (Jo. 368)
judge	equivalence 75 llm-error “pepguiteg'tijig” vs “tegetijig” — The evidence reads 'tegetitjig'; B dropped a 't' giving 'tegetijig', but A's 'pepguiteg'tijig' also does not match the source's 'tegetitjig' well. valid-alternative “beat the drum” vs “to beat the drum” — French 'battre du tambour' can be rendered with or without the infinitive 'to'. unclear “pepgejete'g'tieg” vs “pepgettete'getieg” — Source has 'pepgettetegetieg'; B's form is closer to it while A restructured the syllables, hard to judge without clearer image.

22. ref 117863942057171 (group 220776)

Transcription similarity 44% / lenient 51% confidence 42
researcher	nemphae, plant aquatique; sasnogoatigen
LLM	plante aquatique; neneghas, -sasnogoategen ?) plante ⟨fr⟩ aquatique; ⟨fr⟩ neneghas, ⟨mm⟩ -sasnogoategen ?)⟨mm⟩
judge	equivalence 80 llm-error “nemphae” vs “neneghas” — The image shows 'nemghas' (or 'nemphas'); B's 'neneghas' adds an extra syllable not present. llm-error “plant” vs “plante” — The image reads 'plante' in French, so the researcher's 'plant' drops the final e. unclear “sasnogoatigen” vs “-sasnogoategen ?)” — The two differ on 'a' vs 'e' in the fourth syllable; the handwriting is ambiguous between 'atigen' and 'ategen'. punctuation (absent) vs “-” — B includes the leading dash and trailing ?) visible in the image, which are formatting marks.
Expanded similarity — confidence 58
researcher	(none - not yet done by hand)
LLM	plante aquatique; neneghas, sasnogoategen ?) plante aquatique;⟨fr⟩ neneghas, sasnogoategen ?)⟨mm⟩
Transliteration similarity 39% / lenient 44% confidence 55
researcher	nemphae, aquatic plant, sasnugwatigen
LLM	aquatic plant; nenega's, sasnugwate'gn ?)
judge	equivalence 80 unclear “nemphae” vs “nenega's” — The French evidence reads 'neneghas'; A's 'nemphae' misreads it while B's 'nenega's' is a plausible orthographic rendering, but both diverge from the source spelling. punctuation “aquatic plant, ... ” vs “aquatic plant; ...” — Order and punctuation differ but the translation 'aquatic plant' is identical in substance. valid-alternative “sasnugwatigen” vs “sasnugwate'gn” — Both are defensible orthographic conversions of 'sasnogoategen'; vowel/apostrophe placement differs but neither is clearly wrong. punctuation (absent) vs “?)” — B preserves the question mark and paren from the source; A omits them (punctuation only).

23. ref 120360640585459 (group 221496)

Transcription similarity 79% / lenient 78% confidence 32
researcher	oaga tjeimgeoei, impuissance
LLM	oagat Heimgeoei, impuissam oagat⟨mm⟩ Heimgeoei, impuissam⟨fr⟩
judge	equivalence 80 unclear “oaga tjeimgeoei” vs “oagat Heimgeoei” — The word split differs (oaga tj- vs oagat H-); the initial capital in the second fragment reads as a capital letter in the image, making B's 'H' plausible, but the exact split is ambiguous. llm-error “impuissance” vs “impuissam” — The French word is 'impuissance'; B's 'impuissam' is an incomplete/incorrect rendering.
Expanded similarity 74% / lenient 73% confidence 32
researcher	wagatjeimgeoei, impuissance
LLM	oagat Heimqeoei, impuissam oagat⟨mm⟩ Heimqeoei, impuissam⟨fr⟩
judge	equivalence 55 llm-error “wagatjeimgeoei” vs “oagat Heimqeoei” — Version B misreads the Mi'gmaq word: 'oagat Heimqeoei' vs 'wagatjeimgeoei', with wrong letters and spacing/capitalization. llm-error “impuissance” vs “impuissam” — The French translation should be 'impuissance'; version B truncates it to 'impuissam'.
Transliteration similarity 32% confidence 42
researcher	waqajeimgewei, helplessness/powerlessness
LLM	wagat powerless, impotent
judge	equivalence 55 llm-error “waqajeimgewei” vs “wagat” — The evidence shows 'oagat Heimqeoei' (two elements); B only rendered the first part 'wagat' and dropped the second, plus the researcher's fuller Mi'gmaq form reflects the whole headword. valid-alternative “helplessness/powerlessness” vs “powerless, impotent” — Both translate the French 'impuissance/impuissant'; noun vs adjective phrasing is a defensible variation.

24. ref 120953219354961 (group 218961)

Transcription similarity 48% / lenient 50% confidence 32
researcher	esnagsaong, gig, grappes
LLM	esmagueoeg, ge⁇, (⁇⁇⁇) greffes. with uncertainty markers: esmagueoeg, ge⁇, (⁇⁇⁇) greffes. esmagueoeg, ge⁇,⟨mm⟩ (⁇⁇⁇)⟨cit⟩ greffes.⟨fr⟩
judge	equivalence 55 llm-error “esnagsaong” vs “esmagueoeg” — The Mi'gmaq headword differs in several letters; B misreads the middle consonants and ending. unclear “gig” vs “ge⁇” — The abbreviation is hard to read; both offer different letters and B marks part illegible. valid-alternative (absent) vs “(⁇⁇⁇)” — B notes an illegible parenthetical fragment visible at lower left that A did not transcribe. llm-error “grappes” vs “greffes.” — The circled French word reads 'grappes' (clusters/bunches), not 'greffes'.
Expanded similarity 45% / lenient 51% confidence 42
researcher	esnagsaong, esnagsaongig, grappes
LLM	esmagueoeg, esmagueoegi, (Pi. Met.) greffes. esmagueoeg, esmagueoegi,⟨mm⟩ (Pi. Met.)⟨cit⟩ greffes.⟨fr⟩
judge	equivalence 45 researcher-error “esnagsaong” vs “esmagueoeg” — The image shows 'esmagueoeg' matching version B and the evidence; version A misread the letters. unclear “esnagsaongig” vs “esmagueoegi” — The headword form differs consistently with the first word; the ending (ig vs i) is hard to determine and the plural form is truncated in evidence. researcher-error (absent) vs “(Pi. Met.)” — A parenthetical abbreviation appears in the evidence and image which version A omitted entirely. llm-error “grappes” vs “greffes.” — The evidence reads 'greffes' matching B; version A's 'grappes' is a misreading.
Transliteration similarity 32% / lenient 33% confidence 60
researcher	esnaqsawng, esnaqsawngig, bunch/cluster/group/band
LLM	esmaguewe'g, esmaguewe'gi, (Pi. Met.) grafts.
judge	equivalence 10 researcher-error “esnaqsawng” vs “esmaguewe'g” — Evidence reads 'esmagueoeg'; the image shows 'esmague...' so B's reading is closer to the source while A's 'esnaqsawng' departs substantially. researcher-error “esnaqsawngig” vs “esmaguewe'gi” — Evidence is 'esmagueoegi'; A's inflected form diverges from the source like the headword. researcher-error “bunch/cluster/group/band” vs “grafts.” — French 'greffes' means grafts (B correct); A's translation is unrelated. valid-alternative (absent) vs “(Pi. Met.)” — B retains the abbreviation notation from the source, which A omitted.