Audit in five minutes · run TRANSCRIPTION-001 · part of transcription
The reference, the transcript and the verdict, side by side
Operated by Tempo AI LLC, accountable for the calculation, publication and correction of every figure published here. No vendor pays to be measured; TORNEO buys what it tests.
Each case below is one clip of this run: the reference frozen before the first call, the text the tool returned, both as the metric sees them once normalized, and the journal entry the result was computed from. The word errors are recomputed on this page and printed next to the ones the journal recorded, so a reader can find them disagreeing. Nothing is summarized: a score without its raw lines would be marketing.
Schema torneo_audit_sample.transcription.v1; five clips per arm, selected by the hash of schema, arm and clip, alternating a clip returned with errors and one returned perfectly, and covering the three languages of the corpus; every verdict is checked against the journal before a case is written. Raw file: AUDIT_SAMPLE.jsonl; step-by-step commands: AUDIT_README.md.
Case 1 · arm gemini-2-5-flash · clip de-4681637981934807169 · block B1
German (de) · 16.68 s · CC-BY-4.0 · selected by 009b5cf9baffffb8
Reference, frozen before the first call
Zu dieser Schlussfolgerung kam sie aufgrund der zahlreichen positiven Kommentare und der Ermutigung, die sie sowohl von Männern als auch von Frauen erhielt, die darauf drängten, dass Verhütungsmittel als medizinische Notwendigkeit angesehen werden sollten.
sha256 of the compared form: c3c0894b9f9ee26f800219d8b9ead931b5003af151ca0563c90fb7dc80c0dd53
Journal entry
seq
90
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
Zu dieser Schlussfolgerung kam sie aufgrund der zahlreichen positiven Kommentare und der Ermutigung, die sie sowohl von Männern als auch von Frauen erhielt, die darauf drängten, das Verhütungsmittel als medizinische Notwendigkeit angesehen werden sollten.
sha256 computed here f3ed54f36a6b099f60eece1f98815826bfa3c7ae8ff6ab4320eb3a8b8615f7a6, sha256 in the journal 1d4c37e0beba7a762bec2e4f8fe0ba73577026604e681a97b2a3b26fd244aaab.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
zu dieser schlussfolgerung kam sie aufgrund der zahlreichen positiven kommentare und der ermutigung die sie sowohl von männern als auch von frauen erhielt die darauf drängten dass verhütungsmittel als medizinische notwendigkeit angesehen werden sollten
zu dieser schlussfolgerung kam sie aufgrund der zahlreichen positiven kommentare und der ermutigung die sie sowohl von männern als auch von frauen erhielt die darauf drängten das verhütungsmittel als medizinische notwendigkeit angesehen werden sollten
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
34
34
Words returned
34
not recorded
Word errors
1
1
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-flash/de-4681637981934807169.json.
Case 2 · arm gemini-2-5-flash · clip en-1329002535244170064 · block B1
English (en) · 13.18 s · CC-BY-4.0 · selected by 02bf6c4ad20eac04
Reference, frozen before the first call
In many cases, enrolling on a gap-year course abroad can actually improve your chances of moving into higher education back in your home country.
sha256 of the compared form: 4ce2ed8c84cc665dc63cfbeaed03deae97ca887f5b786ad8c834b075678727f7
Journal entry
seq
113
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
In many cases, enrolling on a gap year course abroad can actually improve your chances on moving into higher education back in your home country.
sha256 computed here 2b8aee63378bcea5ed83266cfe6872c772f45649c1403492a6255c4dfe0e9f2a, sha256 in the journal 2deb01bfebb4608f3fde7f3e63cb80ba50f2cb2c6d8dfeb8257cfcc631d62d16.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
in many cases enrolling on a gap year course abroad can actually improve your chances of moving into higher education back in your home country
in many cases enrolling on a gap year course abroad can actually improve your chances on moving into higher education back in your home country
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
25
25
Words returned
25
not recorded
Word errors
1
1
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-flash/en-1329002535244170064.json.
Case 3 · arm gemini-2-5-flash · clip en-10035998729701048400 · block B1
English (en) · 16.38 s · CC-BY-4.0 · selected by 02eaab9b0680aef4
Reference, frozen before the first call
Now widely available throughout the archipelago, Javanese cuisine features an array of simply seasoned dishes, the predominant flavorings the Javanese favor being peanuts, chillies, sugar (especially Javanese coconut sugar) and various aromatic spices.
sha256 of the compared form: 2e56873a7dacfbddd6d4c16c886fa904884f7d04815bd00b9b3149cb3ef6d085
Journal entry
seq
106
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
Now widely available throughout the archipelago, Japanese cuisine features an array of simply seasoned dishes, the predominant flavorings, the Japanese flavor being peanuts, chilies, sugar, especially Japanese coconut sugar and various aromatic spices.
sha256 computed here 72e11288c294e24760e265827e7cecf43102f3083f47bfbe6c6852d87711be48, sha256 in the journal 21aa910b0095916a7f620b32f43db6aa9e155c13b2fdffef722d0751ed5f82b0.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
now widely available throughout the archipelago javanese cuisine features an array of simply seasoned dishes the predominant flavorings the javanese favor being peanuts chillies sugar especially javanese coconut sugar and various aromatic spices
now widely available throughout the archipelago japanese cuisine features an array of simply seasoned dishes the predominant flavorings the japanese flavor being peanuts chilies sugar especially japanese coconut sugar and various aromatic spices
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
33
33
Words returned
33
not recorded
Word errors
5
5
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-flash/en-10035998729701048400.json.
Case 4 · arm gemini-2-5-flash · clip fr-10283447552217995593 · block B1
French (fr) · 8.52 s · CC-BY-4.0 · selected by 033338ce74e49f92
Reference, frozen before the first call
La principale religion en Moldavie est le christianisme orthodoxe.
sha256 of the compared form: de79df65607232de260ad22b2dba753e997ea855fb52d42c7ffc05df92287771
Journal entry
seq
136
observed
success recorded
yes
verdict agrees with the journal
yes
What the tool returned
La principale religion en Moldavie est le christianisme orthodoxe.
sha256 computed here d7acd93adfe761448f87d13fafb62cf801e941ac1e3a606995ed4c00d02e1abc, sha256 in the journal 9c8017199932be64d9d874bd12250d4c220f6afe06d30aebf93ba06ef70024dc.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
la principale religion en moldavie est le christianisme orthodoxe
la principale religion en moldavie est le christianisme orthodoxe
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
9
9
Words returned
9
not recorded
Word errors
0
0
Transcribed with no error at all
yes
yes
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-flash/fr-10283447552217995593.json.
Case 5 · arm gemini-2-5-flash · clip fr-5025549189826598672 · block B2
French (fr) · 9 s · CC-BY-4.0 · selected by 070b2ebf23704328
Reference, frozen before the first call
Un plat au curry peut être « solide » ou « liquide », selon la quantité de sauce qu'il contient.
sha256 of the compared form: d3387ced9250951144078148d96b3c0f00ee236277e546386efa8d2a7bab76dc
Journal entry
seq
154
observed
success recorded
yes
verdict agrees with the journal
yes
What the tool returned
Un plat au curry peut être solide ou liquide, selon la quantité de sauce qu'il contient.
sha256 computed here 46e82d40231ffc0c5eb84cfd97a3002d7b0aa10d37681611c2db90df974b8349, sha256 in the journal 29989bf35059f6476878e6b90d262047efa240caa5686b373a9f544d5d7a0834.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
un plat au curry peut être solide ou liquide selon la quantité de sauce qu il contient
un plat au curry peut être solide ou liquide selon la quantité de sauce qu il contient
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
17
17
Words returned
17
not recorded
Word errors
0
0
Transcribed with no error at all
yes
yes
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-flash/fr-5025549189826598672.json.
Case 6 · arm gemini-2-5-pro · clip fr-4103141930996483169 · block B1
French (fr) · 7.68 s · CC-BY-4.0 · selected by 08140e32793bb69a
Reference, frozen before the first call
Les Arabes ont également apporté la pratique de l'Islam dans les terres et cette religion s'est largement répandue aux Comores et à Mayotte.
sha256 of the compared form: c2362e3938b01183a515bfbf34b93429622103031a0d85823d253e20843b9fe4
Journal entry
seq
243
observed
success recorded
yes
verdict agrees with the journal
yes
What the tool returned
Les Arabes ont également apporté la pratique de l'Islam dans les terres et cette religion s'est largement répandue aux Comores et à Mayotte.
sha256 computed here 5a4e4429e85194eacf0ac58c1d0f64e208f29d7944b5e82018488985a533fb09, sha256 in the journal ae0a32422bf7a9e3c0f541d125d5f0e3609cce1275a6d8238b6e3800d29b6ef3.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
les arabes ont également apporté la pratique de l islam dans les terres et cette religion s est largement répandue aux comores et à mayotte
les arabes ont également apporté la pratique de l islam dans les terres et cette religion s est largement répandue aux comores et à mayotte
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
25
25
Words returned
25
not recorded
Word errors
0
0
Transcribed with no error at all
yes
yes
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-pro/fr-4103141930996483169.json.
Case 7 · arm gemini-2-5-pro · clip de-13416209024459339640 · block B1
German (de) · 19.38 s · CC-BY-4.0 · selected by 08f07d2d4e08024f
Reference, frozen before the first call
Ihr erstes war der Slalom, bei dem sie in ihrem ersten Lauf ein „Did not Finish“ erhielt. 36 der 116 Teilnehmer erzielten in diesem Rennen das gleiche Ergebnis.
sha256 of the compared form: de94f80e0fc273a64e37e91356ec2d5c246b545d7d7ed553d344899a6ba2c28a
Journal entry
seq
172
observed
success recorded
yes
verdict agrees with the journal
yes
What the tool returned
Ihr erstes war der Slalom, bei dem sie in ihrem ersten Lauf ein Did Not Finish erhielt. 36 der 116 Teilnehmer erzielten in diesem Rennen das gleiche Ergebnis.
sha256 computed here 5701b87b06789cfe45d837038a197bc712718fd954bd52fbecad5b52f0ba2282, sha256 in the journal f99fe2980c86121d5c1553de968f02980651762ad77f8a3eecbf4878266c722f.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
ihr erstes war der slalom bei dem sie in ihrem ersten lauf ein did not finish erhielt 36 der 116 teilnehmer erzielten in diesem rennen das gleiche ergebnis
ihr erstes war der slalom bei dem sie in ihrem ersten lauf ein did not finish erhielt 36 der 116 teilnehmer erzielten in diesem rennen das gleiche ergebnis
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
28
28
Words returned
28
not recorded
Word errors
0
0
Transcribed with no error at all
yes
yes
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-pro/de-13416209024459339640.json.
Case 8 · arm gemini-2-5-pro · clip en-1329002535244170064 · block B1
English (en) · 13.18 s · CC-BY-4.0 · selected by 0d4dca5dce2e1761
Reference, frozen before the first call
In many cases, enrolling on a gap-year course abroad can actually improve your chances of moving into higher education back in your home country.
sha256 of the compared form: 4ce2ed8c84cc665dc63cfbeaed03deae97ca887f5b786ad8c834b075678727f7
Journal entry
seq
203
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
In many cases, enrolling on a gap year course abroad can actually improve your chances on moving into higher education back in your home country.
sha256 computed here 2b8aee63378bcea5ed83266cfe6872c772f45649c1403492a6255c4dfe0e9f2a, sha256 in the journal 8170ec6f6c165bb6c68f9ec1dac1ba7ed35906e1e20fdb279b0d07f93c0b120a.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
in many cases enrolling on a gap year course abroad can actually improve your chances of moving into higher education back in your home country
in many cases enrolling on a gap year course abroad can actually improve your chances on moving into higher education back in your home country
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
25
25
Words returned
25
not recorded
Word errors
1
1
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-pro/en-1329002535244170064.json.
Case 9 · arm gemini-2-5-pro · clip fr-13637937540811867898 · block B1
French (fr) · 9.36 s · CC-BY-4.0 · selected by 106bf2ba66ac32ba
Reference, frozen before the first call
Bien que trois personnes aient été présentes dans la maison quand la voiture l'a percutée, aucune n'a été blessée.
sha256 of the compared form: c4e1f7d52cd50762f391b42f6881d6c9660bfac3bbc747f5f29b407f18959821
Journal entry
seq
233
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
Bien que trois personnes aient été présentes dans la maison quand la voiture l'a percuté, aucune n'a été blessée.
sha256 computed here 15d20129aa66731450fd03a73d096c4a045c298ec824c453e474e6928865a8b7, sha256 in the journal 41d4f815967f6a56623a93a9a90295408769da0c6ec6f3b8eea7bcafda38f561.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
bien que trois personnes aient été présentes dans la maison quand la voiture l a percutée aucune n a été blessée
bien que trois personnes aient été présentes dans la maison quand la voiture l a percuté aucune n a été blessée
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
21
21
Words returned
21
not recorded
Word errors
1
1
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-pro/fr-13637937540811867898.json.
Case 10 · arm gemini-2-5-pro · clip de-9721551483652950762 · block B2
German (de) · 7.44 s · CC-BY-4.0 · selected by 24a40b28458d2e58
Reference, frozen before the first call
Geld kann in der einzigen Bank auf der Insel umgetauscht werden, welche sich in Stanley gegenüber vom FIC West Store befindet.
sha256 of the compared form: 301668025503ad1ee318a70ca3c1bd2613814c5a48131cfa1990edd611ce8f69
Journal entry
seq
194
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
Geld kann in der einzigen Bank auf der Insel umgetauscht werden, welche sich in Stanley gegenüber dem FIC West Store befindet.
sha256 computed here 2c56a6c41e5d7ae22a07d4e64dd80df700aecb1753dcdfc2a8dee27a3be9f0a9, sha256 in the journal 4f1b25a07b9f7b792261252faf7496190405c4ab58a9e75bfb509c78d794dd2d.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
geld kann in der einzigen bank auf der insel umgetauscht werden welche sich in stanley gegenüber vom fic west store befindet
geld kann in der einzigen bank auf der insel umgetauscht werden welche sich in stanley gegenüber dem fic west store befindet
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
21
21
Words returned
21
not recorded
Word errors
1
1
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-2-5-pro/de-9721551483652950762.json.
Case 11 · arm gemini-3-5-flash · clip en-5097652267684488736 · block B1
English (en) · 10.98 s · CC-BY-4.0 · selected by 05a3fbb19443ee28
Reference, frozen before the first call
108 plates of Chhappan Bhog (in Hinduism, 56 different edible items, like, sweets, fruits, nuts, dishes etc. which are offered to deity) were served to Baba Shyam.
sha256 of the compared form: 04bae5139e7314baec8b892eafe08aed516f6a9e852c41ad729b6815e20acfff
Journal entry
seq
307
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
108 plates of chappan bhog in Hinduism 56 different edible items like sweets, fruits, nuts, dishes, etc. which are offered to deity were served to Baba Shyam.
sha256 computed here 956a73317328af1e7a658373ac525529d3bb00ca20c0a969e2d67a909ebdec09, sha256 in the journal cb8e6c34aae7b768eeb763b0f71b94d0f55c9c9efa8348eeed1fd5cc1dae8f59.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
108 plates of chhappan bhog in hinduism 56 different edible items like sweets fruits nuts dishes etc which are offered to deity were served to baba shyam
108 plates of chappan bhog in hinduism 56 different edible items like sweets fruits nuts dishes etc which are offered to deity were served to baba shyam
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
27
27
Words returned
27
not recorded
Word errors
1
1
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-flash/en-5097652267684488736.json.
Case 12 · arm gemini-3-5-flash · clip en-3639051488396204031 · block B1
English (en) · 8.04 s · CC-BY-4.0 · selected by 075d82d9dfd63b5b
Reference, frozen before the first call
The U.N. also hopes to finalize a fund to help countries affected by global warming to cope with the impacts.
sha256 of the compared form: 15ed08aaeaf3a87cb2cf4fef2b8831c9b23600d786b4d6a9f2c6ecf04b90418b
Journal entry
seq
304
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
The UN also hopes to finalize a fund to help countries affected by global warming to cope with the impacts.
sha256 computed here 22348fd42247b2aeeb9c78454b759ee242ce524c246548e98c642a24a3f49399, sha256 in the journal 65d8d70dc7ffedbcc52a15a201d4f92c4b759ffe781f66e59d5ffc9ce483bae8.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
the u n also hopes to finalize a fund to help countries affected by global warming to cope with the impacts
the un also hopes to finalize a fund to help countries affected by global warming to cope with the impacts
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
21
21
Words returned
20
not recorded
Word errors
2
2
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-flash/en-3639051488396204031.json.
Case 13 · arm gemini-3-5-flash · clip fr-5992216071413521711 · block B1
French (fr) · 4.8 s · CC-BY-4.0 · selected by 0d7c8e3d8e6804fd
Reference, frozen before the first call
Des chanteurs célèbres dans tout le pays ont présenté des bhajans, ou chants de dévotion, aux pieds de Shri Shyam.
sha256 of the compared form: 8e64202f26ce557877860062a398c38703ae93043a5d714c73d64bae9c922e06
Journal entry
seq
336
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
Des chanteurs célèbres dans tout le pays ont présenté des bhajans ou chants de dévotion aux pieds de Sri Shyam.
sha256 computed here cefac7bdb63d4882c4913202b8f41df45c30c3df5ccc68bacf57d805f5dbdb98, sha256 in the journal 75c5938591498763629ee7ff8ad4aa120e812c48f833842c13f44c1f21b23dcd.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
des chanteurs célèbres dans tout le pays ont présenté des bhajans ou chants de dévotion aux pieds de shri shyam
des chanteurs célèbres dans tout le pays ont présenté des bhajans ou chants de dévotion aux pieds de sri shyam
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
20
20
Words returned
20
not recorded
Word errors
1
1
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-flash/fr-5992216071413521711.json.
Case 14 · arm gemini-3-5-flash · clip de-7111175006318241186 · block B2
German (de) · 12 s · CC-BY-4.0 · selected by 0fd4cbf525e91982
Reference, frozen before the first call
Einige Patienten haben sich möglicherweise im Krankenhaus mit dem Erreger angesteckt, meint Dr. Moll, und mindestens zwei waren medizinische Krankenhausangestellte.
sha256 of the compared form: a6c5e994a3b583355d7c04b11c4587aff4a424f7d44be07d1ce774384f9efcec
Journal entry
seq
278
observed
success recorded
yes
verdict agrees with the journal
yes
What the tool returned
Einige Patienten haben sich möglicherweise im Krankenhaus mit dem Erreger angesteckt, meint Dr. Moll, und mindestens zwei waren medizinische Krankenhausangestellte.
sha256 computed here 0494910654b72d7e9a3628079a8f6eff5ccc8610f8af3a870469660f1030449b, sha256 in the journal df3e04f0a8481a8cae6c166e1656fbcd1320912cfd390b295a1efceb95a74216.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
einige patienten haben sich möglicherweise im krankenhaus mit dem erreger angesteckt meint dr moll und mindestens zwei waren medizinische krankenhausangestellte
einige patienten haben sich möglicherweise im krankenhaus mit dem erreger angesteckt meint dr moll und mindestens zwei waren medizinische krankenhausangestellte
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
20
20
Words returned
20
not recorded
Word errors
0
0
Transcribed with no error at all
yes
yes
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-flash/de-7111175006318241186.json.
Case 15 · arm gemini-3-5-flash · clip en-11527336559557410281 · block B2
English (en) · 10.92 s · CC-BY-4.0 · selected by 12815ec15f5778ab
Reference, frozen before the first call
Courier companies are well paid for delivering things quickly. Frequently, time is very important with business documents, merchandise or spare parts for an urgent repair.
sha256 of the compared form: 02a9c2a8493aa951f73f98ff2fcd42a1bff9eef7c12632230d89aa7facdb496a
Journal entry
seq
289
observed
success recorded
yes
verdict agrees with the journal
yes
What the tool returned
Courier companies are well paid for delivering things quickly. Frequently, time is very important with business documents, merchandise, or spare parts for an urgent repair.
sha256 computed here c24799f51a936ea94e996431c20b40e1530545fd3a5395e462e2dd788ba9af0b, sha256 in the journal 3ccac4cd6e3ef7ea90deda9cf1f06f3ec11c4b0917a3865ecfb3a49f1be9351e.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
courier companies are well paid for delivering things quickly frequently time is very important with business documents merchandise or spare parts for an urgent repair
courier companies are well paid for delivering things quickly frequently time is very important with business documents merchandise or spare parts for an urgent repair
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
25
25
Words returned
25
not recorded
Word errors
0
0
Transcribed with no error at all
yes
yes
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-flash/en-11527336559557410281.json.
Case 16 · arm gemini-3-5-transcribe · clip en-16256278689854059757 · block B1
English (en) · 7.08 s · CC-BY-4.0 · selected by 00885fd0362097fb
Reference, frozen before the first call
Money can be exchanged at the only bank in the islands which is located in Stanley across from the FIC West store.
sha256 of the compared form: 7ed1b8f70173f63a955e1c6d54995972432c4257d72c095b5f6e39ecd9b486a8
Journal entry
seq
45
observed
success recorded
yes
verdict agrees with the journal
yes
What the tool returned
Money can be exchanged at the only bank in the islands, which is located in Stanley across from the FIC West Store.
sha256 computed here b87491fae5b4cf357b8eaf94ff09d43848a48f326cd26d5930a83b460ec052db, sha256 in the journal 244448dc3f2c91a8b3c8c5dedb9ba567c947add8a892e3e8e48baeed64026512.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
money can be exchanged at the only bank in the islands which is located in stanley across from the fic west store
money can be exchanged at the only bank in the islands which is located in stanley across from the fic west store
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
22
22
Words returned
22
not recorded
Word errors
0
0
Transcribed with no error at all
yes
yes
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-transcribe/en-16256278689854059757.json.
Case 17 · arm gemini-3-5-transcribe · clip de-14245076574529164779 · block B2
German (de) · 10.8 s · CC-BY-4.0 · selected by 02831127d6518985
Reference, frozen before the first call
Ein Kurs deckt alle hier diskutierten Probleme in der Regel viel detaillierter ab, normalerweise in Kombination mit praktischer Erfahrung.
sha256 of the compared form: 38fefce9922a74a31f2498ee34465ed377517da645c7b7a978eab6e5b18e6bbe
Journal entry
seq
8
observed
success recorded
yes
verdict agrees with the journal
yes
What the tool returned
Ein Kurs deckt alle hier diskutierten Probleme in der Regel viel detaillierter ab, normalerweise in Kombination mit praktischer Erfahrung.
sha256 computed here 6c92fee1231bacfe8aaa57362c96add0b7949466df85f9e75009a8b6547fc79a, sha256 in the journal 0f8ebe5b387c4e16ac999b77616133c9e3b6d6571731f2b9d4c8d6833102c5bd.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
ein kurs deckt alle hier diskutierten probleme in der regel viel detaillierter ab normalerweise in kombination mit praktischer erfahrung
ein kurs deckt alle hier diskutierten probleme in der regel viel detaillierter ab normalerweise in kombination mit praktischer erfahrung
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
19
19
Words returned
19
not recorded
Word errors
0
0
Transcribed with no error at all
yes
yes
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-transcribe/de-14245076574529164779.json.
Case 18 · arm gemini-3-5-transcribe · clip de-9721551483652950762 · block B2
German (de) · 7.44 s · CC-BY-4.0 · selected by 06dc3d38586d8461
Reference, frozen before the first call
Geld kann in der einzigen Bank auf der Insel umgetauscht werden, welche sich in Stanley gegenüber vom FIC West Store befindet.
sha256 of the compared form: 301668025503ad1ee318a70ca3c1bd2613814c5a48131cfa1990edd611ce8f69
Journal entry
seq
29
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
Geld kann in der einzigen Bank auf der Insel umgetauscht werden, welche sich in Stanley gegenüber dem FIC West Store befindet.
sha256 computed here 2c56a6c41e5d7ae22a07d4e64dd80df700aecb1753dcdfc2a8dee27a3be9f0a9, sha256 in the journal ea1f4b506c133682c82623947402a56a19854d18eb61217664c00cb0fd7cdc33.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
geld kann in der einzigen bank auf der insel umgetauscht werden welche sich in stanley gegenüber vom fic west store befindet
geld kann in der einzigen bank auf der insel umgetauscht werden welche sich in stanley gegenüber dem fic west store befindet
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
21
21
Words returned
21
not recorded
Word errors
1
1
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-transcribe/de-9721551483652950762.json.
Case 19 · arm gemini-3-5-transcribe · clip de-8144480992448526 · block B2
German (de) · 29.22 s · CC-BY-4.0 · selected by 119badd67c76dbea
Reference, frozen before the first call
"„Ich habe meine Schwester und ihre Freundin verloren, und auf meinem Weg waren zwei Behinderte im Rollstuhl, über die Leute einfach drübergesprungen sind und sie geschubst haben"", sagte Armand Versace."
sha256 of the compared form: 37aab5a63721fabee97b238858e7e9e675075eb7ae44d76d2fa4b65d230882be
Journal entry
seq
25
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
Ich habe meine Schwester und ihre Freundin verloren und auf meinem Weg waren zwei Behinderte im Rollstuhl, über die Leute einfach drüber gesprungen sind und sie geschubst haben, sagte Armand Wersach.
sha256 computed here 17938915b7098d094a4777683ad3079458d3d909c9b764f1088424fd04b4beb8, sha256 in the journal aefb1a8d08e4b0ce37926780b3470769be6ff281875d05605d2d09787aa732f0.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
ich habe meine schwester und ihre freundin verloren und auf meinem weg waren zwei behinderte im rollstuhl über die leute einfach drübergesprungen sind und sie geschubst haben sagte armand versace
ich habe meine schwester und ihre freundin verloren und auf meinem weg waren zwei behinderte im rollstuhl über die leute einfach drüber gesprungen sind und sie geschubst haben sagte armand wersach
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
30
30
Words returned
31
not recorded
Word errors
3
3
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-transcribe/de-8144480992448526.json.
Case 20 · arm gemini-3-5-transcribe · clip fr-10828476775853126892 · block B2
French (fr) · 11.76 s · CC-BY-4.0 · selected by 13204266bb8d0a0f
Reference, frozen before the first call
Les femmes doivent réaliser que certaines différences culturelles peuvent entraîner des comportements qu'elles considèrent comme du harcèlement. Il est en effet courant qu'elles soient suivies, attrapées par le bras, etc.
sha256 of the compared form: 79651b025d207224260291eab7c8d61f2c931edfe90c6b5fc0fe11004aa0a2df
Journal entry
seq
64
observed
success recorded
no
verdict agrees with the journal
yes
What the tool returned
Les femmes doivent réaliser des réaliser que certaines différences culturelles peuvent entraîner des comportements qu'elles considèrent comme du harcèlement. Il est en effet courant qu'elles soient suivies, attrapées par le bras, etc.
sha256 computed here 0ceac5468ce12cfbb7cd7fc8d527ecc5e0040e6cd3a8f87dda6573783cb52511, sha256 in the journal 687854dbd5229384b059f667bcd928995fbab25253282bb5135ae85934eb7daf.
The two texts as the metric sees them (tnorm.v1)
The verdict is computed on these two lines and on nothing else. Reading the raw text alone cannot tell a model error from a normalization one.
les femmes doivent réaliser que certaines différences culturelles peuvent entraîner des comportements qu elles considèrent comme du harcèlement il est en effet courant qu elles soient suivies attrapées par le bras etc
les femmes doivent réaliser des réaliser que certaines différences culturelles peuvent entraîner des comportements qu elles considèrent comme du harcèlement il est en effet courant qu elles soient suivies attrapées par le bras etc
Verdict, recomputed here and read from the journal
Word counts and errors, recomputed and as journalled
Quantity
Recomputed on this page
In the journal
Words in the reference
32
32
Words returned
34
not recorded
Word errors
2
2
Transcribed with no error at all
no
no
This case counts toward perfect_transcription_rate, the metric the result is built from: the share of clips returned with no word error at all. An API failure counts as an unsuccessful attempt, never as a missing observation. Source file: transcripts/gemini-3-5-transcribe/fr-10828476775853126892.json.