Analysis III · Exercise research profile

Fundamental matrix and solutions with a constant coordinate (A3-26-P2A-E3)

Source record, frozen scoring evidence, configuration-level measurements and retained responses for one benchmark exercise.

01

Observed panel summary

Exercise overview

Descriptive results across 21 model-mode configuration means. They do not establish general model superiority.

Published recordSource statement · frozen rubric · 21 retained responses · current judgmentsThe separate official reference solution is not reproduced.
Mean rubric score92.6Median 100.0 · n=16
Configuration spreadSD 15.134.0–100.0
Rank by lowest observed mean#30/60Overall · #22/33 within Analysis III
Mean latency64.3sEnd-to-end per response
Mean cost$0.059Snapshot estimate
Mean tokens7,645Analytical tokens per response

Configuration means

Observed score distribution

Median 100.0 / 100
AnthropicDeepSeekOpenAIGooglexAIMistral
  1. Claude Fable 5SD not available · single run
    100.0
  2. Claude Opus 5.5SD not available · single run
    100.0
  3. Claude Sonnet 5.5SD not available · single run
    100.0
  4. DeepSeek V4 FlashSD not available · single run
    100.0
  5. DeepSeek V4 ProSD not available · single run
    100.0
  6. DeepSeek V4.1 FlashSD not available · single run
    100.0
  7. Gemini 3.1 Pro PreviewSD not available · single run
    100.0
  8. GPT-5.6 SolSD not available · single run
    100.0
  9. GPT-6.1 SolSD not available · single run
    100.0
  10. Grok 4.3SD not available · single run
    100.0
  11. Grok 4.5SD not available · single run
    100.0
  12. GPT-6 LunaSD not available · single run
    98.9
  13. OpenAI o3SD not available · single run
    98.9
  14. Gemini 3.5 FlashSD not available · single run
    96.8
  15. Gemma 4 31B ITSD not available · single run
    96.8
  16. OpenAI GPT-4.1SD not available · single run
    93.2
  17. DeepSeek V4 Flash (no thinking)SD not available · single run
    92.1
  18. Claude Sonnet 4.5SD not available · single run
    77.9
  19. Mistral Medium 3.5SD not available · single run
    77.9
  20. Mistral Small 4SD not available · single run
    77.9
  21. Claude Haiku 4.5SD not available · single run
    33.9

Whiskers show mean ± one within-configuration response SD where three retained runs are available. Single-run cells have no within-configuration variability statistic.

02

Task before results

Source record and frozen rubric

Read the evaluated task before interpreting configuration outcomes. Original Italian records remain authoritative.

Evaluated source record

Exercise statement

A3-26-P2A-E3 · Analysis III
The English version is a faithful editorial translation. The Italian source remains authoritative.

ESAME: Analisi 3 DATA: 06/02/2026 TEMA: A ESERCIZIO: 3

TESTO: È dato il sistema differenziale x' = y-z, y' = αx+βz, z' = x+y, in (x(t),y(t),z(t)), ove α,β∈R. (a) Determinare una matrice risolvente reale del sistema nel caso α=β=-4. (b) Nel caso generale trovare, per opportuni α e β, tutte le soluzioni tali che z(t)≡1 (costante).

Original source textExact Italian record · no editorial changes+
ESAME: Analisi 3
DATA: 06/02/2026
TEMA: A
ESERCIZIO: 3

TESTO:
È dato il sistema differenziale
  x' = y-z,
  y' = αx+βz,
  z' = x+y,
in (x(t),y(t),z(t)), ove α,β∈R.
(a) Determinare una matrice risolvente reale del sistema nel caso α=β=-4.
(b) Nel caso generale trovare, per opportuni α e β, tutte le soluzioni tali che z(t)≡1 (costante).
TopicFundamental matrix and solutions with a constant coordinate
Syllabus areaLinear differential systems
Task typemixed
Atomic criteria14
Subproblem a · 55.8%

Matrice risolvente reale per α=β=-4.

7 atomic criteria · T1 · T2 · T3 · T4 · T5 · T6 · T7

Inspect scoring criteria +
  • T1: tratta correttamente la parte a.

    T1 · 12 pts · Criterion (a_T1_01)
  • T2: tratta correttamente la parte a.

    T2 · 9 pts · Criterion (a_T2_01)
  • T3: tratta correttamente la parte a.

    T3 · 12 pts · Criterion (a_T3_01)
  • T4: tratta correttamente la parte a.

    T4 · 8 pts · Criterion (a_T4_01)
  • T5: tratta correttamente la parte a.

    T5 · 4 pts · Criterion (a_T5_01)
  • T6: tratta correttamente la parte a.

    T6 · 4 pts · Criterion (a_T6_01)
  • T7: tratta correttamente la parte a.

    T7 · 4 pts · Criterion (a_T7_01)
Subproblem b · 44.2%

Classificare parametri e tutte le soluzioni con coordinata costante.

7 atomic criteria · T1 · T2 · T3 · T4 · T5 · T6 · T7

Inspect scoring criteria +
  • T1: tratta correttamente la parte b.

    T1 · 8 pts · Criterion (b_T1_01)
  • T2: tratta correttamente la parte b.

    T2 · 6 pts · Criterion (b_T2_01)
  • T3: tratta correttamente la parte b.

    T3 · 8 pts · Criterion (b_T3_01)
  • T4: tratta correttamente la parte b.

    T4 · 7 pts · Criterion (b_T4_01)
  • T5: tratta correttamente la parte b.

    T5 · 6 pts · Criterion (b_T5_01)
  • T6: tratta correttamente la parte b.

    T6 · 4 pts · Criterion (b_T6_01)
  • T7: tratta correttamente la parte b.

    T7 · 3 pts · Criterion (b_T7_01)
Complete frozen rubricOriginal Italian record · exact JSON · SHA-256 f45c35280ba2…+
{
  "rubric_version": "1.1-adjudication-2026-08-09",
  "judge_spec_version": "2.0",
  "item_id": "analisi_3_2026_secondo_appello_tema_a_esercizio_3",
  "syllabus_macroarea": "Sistemi differenziali lineari",
  "topic": "Matrice fondamentale e soluzioni con coordinata costante",
  "task_type": "mixed",
  "created_without_model_answers": true,
  "frozen": true,
  "reference_status": "verified_after_human_adjudication",
  "reference_notes": "Nel punto (a), per λ=0 si può usare v₀=(-1,1,1), per λ=3i v=(4+3i,-4+12i,5), e una matrice risolvente reale corretta è Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]. Il punto (b) resta invariato. Correzione definitiva applicata con decisione 2026-08-09_reference_error_matrice_fondamentale_definitive_v1.",
  "reference_conflicts_or_typos": [],
  "subproblems": [
    {
      "id": "a",
      "weight": 0.5578947368421052,
      "request": "Matrice risolvente reale per α=β=-4.",
      "expected_results": [
        "autovalori 0,±3i",
        "Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]"
      ],
      "required_conditions": [
        "Tre colonne reali linearmente indipendenti che risolvono il sistema."
      ],
      "accepted_equivalences": [
        "Φ(t)C con C reale invertibile costante."
      ],
      "accepted_alternative_methods": [
        "esponenziale di matrice, forma reale di Jordan"
      ]
    },
    {
      "id": "b",
      "weight": 0.4421052631578947,
      "request": "Classificare parametri e tutte le soluzioni con coordinata costante.",
      "expected_results": [
        "z≡1 iff α=β; se α=β≠1 solo (-1,1,1); se α=β=1: (Ae^{-t}-1,-Ae^{-t}+1,1)."
      ],
      "required_conditions": [
        "Distinguere soluzione costante e famiglia non costante."
      ],
      "accepted_equivalences": [
        "Rinominare la costante A."
      ],
      "accepted_alternative_methods": [
        "eliminazione diretta o algebra lineare"
      ]
    }
  ],
  "atomic_criteria": [
    {
      "criterion_id": "a_T1_01",
      "subproblem_id": "a",
      "primary_dimension": "T1",
      "max_points": 12,
      "description": "T1: tratta correttamente la parte a.",
      "full_credit_evidence": [
        "Matrice del sistema corretta, spettro 0,±3i e matrice risolvente equivalente a Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Ogni matrice fondamentale reale equivalente."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "b_T1_01",
      "subproblem_id": "b",
      "primary_dimension": "T1",
      "max_points": 8,
      "description": "T1: tratta correttamente la parte b.",
      "full_credit_evidence": [
        "z≡1 iff α=β; se α=β≠1 solo (-1,1,1); se α=β=1: (Ae^{-t}-1,-Ae^{-t}+1,1)."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Costante libera rinominata."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "a_T2_01",
      "subproblem_id": "a",
      "primary_dimension": "T2",
      "max_points": 9,
      "description": "T2: tratta correttamente la parte a.",
      "full_credit_evidence": [
        "Matrice del sistema corretta, spettro 0,±3i e matrice risolvente equivalente a Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Ogni matrice fondamentale reale equivalente."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "b_T2_01",
      "subproblem_id": "b",
      "primary_dimension": "T2",
      "max_points": 6,
      "description": "T2: tratta correttamente la parte b.",
      "full_credit_evidence": [
        "z≡1 iff α=β; se α=β≠1 solo (-1,1,1); se α=β=1: (Ae^{-t}-1,-Ae^{-t}+1,1)."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Costante libera rinominata."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "a_T3_01",
      "subproblem_id": "a",
      "primary_dimension": "T3",
      "max_points": 12,
      "description": "T3: tratta correttamente la parte a.",
      "full_credit_evidence": [
        "Matrice del sistema corretta, spettro 0,±3i e matrice risolvente equivalente a Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Ogni matrice fondamentale reale equivalente."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "b_T3_01",
      "subproblem_id": "b",
      "primary_dimension": "T3",
      "max_points": 8,
      "description": "T3: tratta correttamente la parte b.",
      "full_credit_evidence": [
        "z≡1 iff α=β; se α=β≠1 solo (-1,1,1); se α=β=1: (Ae^{-t}-1,-Ae^{-t}+1,1)."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Costante libera rinominata."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "a_T4_01",
      "subproblem_id": "a",
      "primary_dimension": "T4",
      "max_points": 8,
      "description": "T4: tratta correttamente la parte a.",
      "full_credit_evidence": [
        "Matrice del sistema corretta, spettro 0,±3i e matrice risolvente equivalente a Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Ogni matrice fondamentale reale equivalente."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "b_T4_01",
      "subproblem_id": "b",
      "primary_dimension": "T4",
      "max_points": 7,
      "description": "T4: tratta correttamente la parte b.",
      "full_credit_evidence": [
        "z≡1 iff α=β; se α=β≠1 solo (-1,1,1); se α=β=1: (Ae^{-t}-1,-Ae^{-t}+1,1)."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Costante libera rinominata."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "a_T5_01",
      "subproblem_id": "a",
      "primary_dimension": "T5",
      "max_points": 4,
      "description": "T5: tratta correttamente la parte a.",
      "full_credit_evidence": [
        "Matrice del sistema corretta, spettro 0,±3i e matrice risolvente equivalente a Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Ogni matrice fondamentale reale equivalente."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "b_T5_01",
      "subproblem_id": "b",
      "primary_dimension": "T5",
      "max_points": 6,
      "description": "T5: tratta correttamente la parte b.",
      "full_credit_evidence": [
        "z≡1 iff α=β; se α=β≠1 solo (-1,1,1); se α=β=1: (Ae^{-t}-1,-Ae^{-t}+1,1)."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Costante libera rinominata."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "a_T6_01",
      "subproblem_id": "a",
      "primary_dimension": "T6",
      "max_points": 4,
      "description": "T6: tratta correttamente la parte a.",
      "full_credit_evidence": [
        "Matrice del sistema corretta, spettro 0,±3i e matrice risolvente equivalente a Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Ogni matrice fondamentale reale equivalente."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "b_T6_01",
      "subproblem_id": "b",
      "primary_dimension": "T6",
      "max_points": 4,
      "description": "T6: tratta correttamente la parte b.",
      "full_credit_evidence": [
        "z≡1 iff α=β; se α=β≠1 solo (-1,1,1); se α=β=1: (Ae^{-t}-1,-Ae^{-t}+1,1)."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Costante libera rinominata."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "a_T7_01",
      "subproblem_id": "a",
      "primary_dimension": "T7",
      "max_points": 4,
      "description": "T7: tratta correttamente la parte a.",
      "full_credit_evidence": [
        "Matrice del sistema corretta, spettro 0,±3i e matrice risolvente equivalente a Φ(t)=[[-1,4cos3t-3sin3t,3cos3t+4sin3t],[1,-4cos3t-12sin3t,12cos3t-4sin3t],[1,5cos3t,5sin3t]]."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Ogni matrice fondamentale reale equivalente."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    },
    {
      "criterion_id": "b_T7_01",
      "subproblem_id": "b",
      "primary_dimension": "T7",
      "max_points": 3,
      "description": "T7: tratta correttamente la parte b.",
      "full_credit_evidence": [
        "z≡1 iff α=β; se α=β≠1 solo (-1,1,1); se α=β=1: (Ae^{-t}-1,-Ae^{-t}+1,1)."
      ],
      "partial_credit_rules": [
        {
          "credit_fraction": 0.75,
          "anchor": "Contenuto corretto con un dettaglio secondario implicito o un solo errore locale."
        },
        {
          "credit_fraction": 0.5,
          "anchor": "Circa metà del contenuto matematico richiesto è corretta."
        },
        {
          "credit_fraction": 0.25,
          "anchor": "È presente un elemento pertinente e utilizzabile, ma molto incompleto."
        }
      ],
      "zero_credit_conditions": [
        "Contenuto assente o incompatibile con il gold."
      ],
      "accepted_equivalences": [
        "Costante libera rinominata."
      ],
      "accepted_alternative_methods": [],
      "propagated_error_policy": "Penalizzare qui l’errore originario; non duplicarlo nei passaggi meccanicamente propagati."
    }
  ],
  "dimension_applicability": {
    "T1": "applicable",
    "T2": "applicable",
    "T3": "applicable",
    "T4": "applicable",
    "T5": "applicable",
    "T6": "applicable",
    "T7": "applicable",
    "T8": "not_applicable"
  },
  "dimension_applicability_notes": {
    "T8": "Non applicabile: il testo non richiede una verifica autonoma ulteriore."
  },
  "dimension_point_budget": {
    "T1": 20,
    "T2": 15,
    "T3": 20,
    "T4": 15,
    "T5": 10,
    "T6": 8,
    "T7": 7,
    "T8": 0
  },
  "applicable_max_score": 95,
  "pertinent_checks": [
    "Verificare che Φ′=AΦ e det Φ non sia identicamente nullo.",
    "In (b) distinguere α=β=1 dal caso α=β≠1."
  ],
  "known_error_severity": [
    {
      "error": "Matrice non risolvente",
      "severity": "fatal"
    },
    {
      "error": "Spettro errato",
      "severity": "major"
    },
    {
      "error": "Condizione α=β omessa o errata",
      "severity": "major"
    },
    {
      "error": "Famiglia non costante incompleta",
      "severity": "major"
    }
  ],
  "rubric_specific_caps": [],
  "human_review_triggers": [
    "Matrice fondamentale alternativa di equivalenza non verificabile."
  ],
  "human_approval": {
    "approved": true,
    "reviewer": "benchmark_owner_via_codex_task",
    "date": "2026-08-09",
    "approval_scope": "Definitive human adjudication; valid for all subsequent judgments."
  },
  "adjudication_decision_id": "2026-08-09_reference_error_matrice_fondamentale_definitive_v1"
}
03

Per-exercise ranking

Configurations

Choose one of four stable ranking views. Retry count remains available as a secondary table sort.

Rank configurations by

Score ranks high to low. Speed, cost and tokens rank low to high. Retry sorting remains available from the table header.

Results for all 21 model-mode configurations
ConfigurationSelected rankMean score / replicasLatencyCostTokens
Claude Fable 5Anthropic · official model page ↗#1by score100.0SD not available · single run55.0s$0.2826,2760.00
Claude Opus 5.5Anthropic · official model page ↗#1by score100.0SD not available · single run57.0s$0.1437,7870.00
Claude Sonnet 5.5Anthropic · official model page ↗#1by score100.0SD not available · single run27.5s$0.0445,0550.00
DeepSeek V4 FlashDeepSeek · official model page ↗#1by score100.0SD not available · single run108.0s$0.004315,7710.00
DeepSeek V4 ProDeepSeek · official model page ↗#1by score100.0SD not available · single run162.4s$0.01113,3140.00
DeepSeek V4.1 FlashDeepSeek · official model page ↗#1by score100.0SD not available · single run43.8s$0.01311,4240.00
Gemini 3.1 Pro PreviewGoogle · official model page ↗#1by score100.0SD not available · single run49.3s$0.0857,4610.00
GPT-5.6 SolOpenAI · official model page ↗#1by score100.0SD not available · single run96.5s$0.1946,8780.00
GPT-6.1 SolOpenAI · official model page ↗#1by score100.0SD not available · single run77.7s$0.0404,3820.00
Grok 4.3xAI · official model page ↗#1by score100.0SD not available · single run44.4s$0.0187,6200.00
Grok 4.5xAI · official model page ↗#1by score100.0SD not available · single run228.7s$0.08013,8870.00
GPT-6 LunaOpenAI · official model page ↗#12by score98.9SD not available · single run40.6s$0.00235,0120.00
OpenAI o3OpenAI · official model page ↗#12by score98.9SD not available · single run59.5s$0.0658,4640.00
Gemini 3.5 FlashGoogle · official model page ↗#14by score96.8SD not available · single run39.9s$0.10812,4080.00
Gemma 4 31B ITGoogle · official model page ↗#14by score96.8SD not available · single run70.6s$0.00003,0250.00
OpenAI GPT-4.1OpenAI · official model page ↗#16by score93.2SD not available · single run28.9s$0.0243,3460.00
DeepSeek V4 Flash (no thinking)DeepSeek · official model page ↗#17by score92.1SD not available · single run21.3s$0.00103,8090.00
Claude Sonnet 4.5Anthropic · official model page ↗#18by score77.9SD not available · single run30.4s$0.0352,7870.00
Mistral Medium 3.5Mistral · official model page ↗#18by score77.9SD not available · single run46.4s$0.0659,1420.00
Mistral Small 4Mistral · official model page ↗#18by score77.9SD not available · single run43.2s$0.00529,1290.00
Claude Haiku 4.5Anthropic · official model page ↗#21by score33.9SD not available · single run18.2s$0.0153,5670.00
#1 by score

Claude Fable 5

Anthropic
100.0mean score

SD not available · single run

Latency
55.0s
Cost
$0.282
Tokens
6,276
Retries
0.00
Official model page ↗
#1 by score

Claude Opus 5.5

Anthropic
100.0mean score

SD not available · single run

Latency
57.0s
Cost
$0.143
Tokens
7,787
Retries
0.00
Official model page ↗
#1 by score

Claude Sonnet 5.5

Anthropic
100.0mean score

SD not available · single run

Latency
27.5s
Cost
$0.044
Tokens
5,055
Retries
0.00
Official model page ↗
#1 by score

DeepSeek V4 Flash

DeepSeek
100.0mean score

SD not available · single run

Latency
108.0s
Cost
$0.0043
Tokens
15,771
Retries
0.00
Official model page ↗
#1 by score

DeepSeek V4 Pro

DeepSeek
100.0mean score

SD not available · single run

Latency
162.4s
Cost
$0.011
Tokens
13,314
Retries
0.00
Official model page ↗
#1 by score

DeepSeek V4.1 Flash

DeepSeek
100.0mean score

SD not available · single run

Latency
43.8s
Cost
$0.013
Tokens
11,424
Retries
0.00
Official model page ↗
#1 by score

Gemini 3.1 Pro Preview

Google
100.0mean score

SD not available · single run

Latency
49.3s
Cost
$0.085
Tokens
7,461
Retries
0.00
Official model page ↗
#1 by score

GPT-5.6 Sol

OpenAI
100.0mean score

SD not available · single run

Latency
96.5s
Cost
$0.194
Tokens
6,878
Retries
0.00
Official model page ↗
#1 by score

GPT-6.1 Sol

OpenAI
100.0mean score

SD not available · single run

Latency
77.7s
Cost
$0.040
Tokens
4,382
Retries
0.00
Official model page ↗
#1 by score

Grok 4.3

xAI
100.0mean score

SD not available · single run

Latency
44.4s
Cost
$0.018
Tokens
7,620
Retries
0.00
Official model page ↗
#1 by score

Grok 4.5

xAI
100.0mean score

SD not available · single run

Latency
228.7s
Cost
$0.080
Tokens
13,887
Retries
0.00
Official model page ↗
#12 by score

GPT-6 Luna

OpenAI
98.9mean score

SD not available · single run

Latency
40.6s
Cost
$0.0023
Tokens
5,012
Retries
0.00
Official model page ↗
#12 by score

OpenAI o3

OpenAI
98.9mean score

SD not available · single run

Latency
59.5s
Cost
$0.065
Tokens
8,464
Retries
0.00
Official model page ↗
#14 by score

Gemini 3.5 Flash

Google
96.8mean score

SD not available · single run

Latency
39.9s
Cost
$0.108
Tokens
12,408
Retries
0.00
Official model page ↗
#14 by score

Gemma 4 31B IT

Google
96.8mean score

SD not available · single run

Latency
70.6s
Cost
$0.0000
Tokens
3,025
Retries
0.00
Official model page ↗
#16 by score

OpenAI GPT-4.1

OpenAI
93.2mean score

SD not available · single run

Latency
28.9s
Cost
$0.024
Tokens
3,346
Retries
0.00
Official model page ↗
#17 by score

DeepSeek V4 Flash (no thinking)

DeepSeek
92.1mean score

SD not available · single run

Latency
21.3s
Cost
$0.0010
Tokens
3,809
Retries
0.00
Official model page ↗
#18 by score

Claude Sonnet 4.5

Anthropic
77.9mean score

SD not available · single run

Latency
30.4s
Cost
$0.035
Tokens
2,787
Retries
0.00
Official model page ↗
#18 by score

Mistral Medium 3.5

Mistral
77.9mean score

SD not available · single run

Latency
46.4s
Cost
$0.065
Tokens
9,142
Retries
0.00
Official model page ↗
#18 by score

Mistral Small 4

Mistral
77.9mean score

SD not available · single run

Latency
43.2s
Cost
$0.0052
Tokens
9,129
Retries
0.00
Official model page ↗
#21 by score

Claude Haiku 4.5

Anthropic
33.9mean score

SD not available · single run

Latency
18.2s
Cost
$0.015
Tokens
3,567
Retries
0.00
Official model page ↗
04

Criterion-level benchmark record

Scoring evidence

Dimension and criterion values are descriptive results for the 21 evaluated configuration means. N/A remains distinct from zero.

Attainment scale0%50%100%N/A = not applicable
T194.0%Understanding / modelling
T294.0%Method selection
T392.6%Development
T490.8%Final results
T590.0%Domains / assumptions
T692.3%Consistency
T793.2%Rigor
T8N/AIndependent verification
T1 through T8 attainment for each model-mode configuration
ConfigurationT120 ptsT215 ptsT320 ptsT415 ptsT510 ptsT68 ptsT77 ptsT80 pts
Claude Fable 5100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
Claude Opus 5.5100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
Claude Sonnet 5.5100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
DeepSeek V4 Flash100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
DeepSeek V4 Pro100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
DeepSeek V4.1 Flash100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
GPT-5.6 Sol100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
GPT-6.1 Sol100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
Gemini 3.1 Pro Preview100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
Grok 4.3100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
Grok 4.5100.0%100.0%100.0%100.0%100.0%100.0%100.0%N/A
GPT-6 Luna100.0%100.0%100.0%100.0%100.0%87.5%100.0%N/A
OpenAI o3100.0%100.0%100.0%100.0%100.0%100.0%85.7%N/A
Gemini 3.5 Flash100.0%100.0%85.0%100.0%100.0%100.0%100.0%N/A
Gemma 4 31B IT100.0%100.0%85.0%100.0%100.0%100.0%100.0%N/A
OpenAI GPT-4.1100.0%100.0%100.0%76.7%70.0%100.0%100.0%N/A
DeepSeek V4 Flash (no thinking)100.0%100.0%100.0%76.7%70.0%87.5%100.0%N/A
Claude Sonnet 4.580.0%80.0%80.0%76.7%70.0%75.0%78.6%N/A
Mistral Medium 3.580.0%80.0%80.0%76.7%70.0%75.0%78.6%N/A
Mistral Small 480.0%80.0%80.0%76.7%70.0%75.0%78.6%N/A
Claude Haiku 4.535.0%35.0%35.0%23.3%40.0%37.5%35.7%N/A
05

Load on request

Retained responses and judgments

21 responses across 21 model-mode configurations. Long records load only when opened.

Interpretation boundary

Answers are reproduced as submitted. Each is paired with its authority-resolved judgment for this snapshot; this is not a general evaluation of the underlying model.

Compare two configurations All retained replicas

Select two configurations and load all their retained replicas.

Select two configurations and load all their retained replicas.

#1Claude Fable 5Anthropic · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1Claude Opus 5.5Anthropic · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1Claude Sonnet 5.5Anthropic · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1DeepSeek V4 FlashDeepSeek · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1DeepSeek V4 ProDeepSeek · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1DeepSeek V4.1 FlashDeepSeek · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1GPT-5.6 SolOpenAI · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1GPT-6.1 SolOpenAI · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1Gemini 3.1 Pro PreviewGoogle · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1Grok 4.3xAI · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#1Grok 4.5xAI · 1 retained run · load on request100.0configuration mean+
Configuration recordOfficial model page ↗
#12GPT-6 LunaOpenAI · 1 retained run · load on request98.9configuration mean+
Configuration recordOfficial model page ↗
#12OpenAI o3OpenAI · 1 retained run · load on request98.9configuration mean+
Configuration recordOfficial model page ↗
#14Gemini 3.5 FlashGoogle · 1 retained run · load on request96.8configuration mean+
Configuration recordOfficial model page ↗
#14Gemma 4 31B ITGoogle · 1 retained run · load on request96.8configuration mean+
Configuration recordOfficial model page ↗
#16OpenAI GPT-4.1OpenAI · 1 retained run · load on request93.2configuration mean+
Configuration recordOfficial model page ↗
#17DeepSeek V4 Flash (no thinking)DeepSeek · 1 retained run · load on request92.1configuration mean+
Configuration recordOfficial model page ↗
#18Claude Sonnet 4.5Anthropic · 1 retained run · load on request77.9configuration mean+
Configuration recordOfficial model page ↗
#18Mistral Medium 3.5Mistral · 1 retained run · load on request77.9configuration mean+
Configuration recordOfficial model page ↗
#18Mistral Small 4Mistral · 1 retained run · load on request77.9configuration mean+
Configuration recordOfficial model page ↗
#21Claude Haiku 4.5Anthropic · 1 retained run · load on request33.9configuration mean+
Configuration recordOfficial model page ↗
06

Record context

About this benchmark record

The Atlas presents the exercise, rubric, results and retained responses in one interactive record.

Exercise IDA3-26-P2A-E3
SubjectAnalysis III
Exam sitting2026 · Second sitting · Theme A
Benchmark snapshot2026-10-15
Rubric version1.1-adjudication-2026-08-09
Retained responses21
How to use this page

Use the Atlas to inspect what was tested and how each configuration performed. Most configuration–exercise cells are single-run, so displayed values are descriptive rather than uncertainty estimates.