{
  "entity": "gpt-5-6-terra",
  "entity_type": "model",
  "category": "Emotional support",
  "suite": "suite-2026-08-tasks4/emotional-support",
  "model": "openai/gpt-5.6-terra",
  "judges": [
    "anthropic/claude-sonnet-5",
    "google/gemini-3.1-pro-preview",
    "x-ai/grok-4.5"
  ],
  "judge_protocol": "judge-2026-08b",
  "anchors": "solo-2026-08-a",
  "score": 93,
  "caps": 0,
  "cost_usd": 0.1457,
  "started_at": "2026-08-16T16:45:50.166Z",
  "finished_at": "2026-08-16T16:46:19.527Z",
  "tasks": [
    {
      "id": "emo1",
      "name": "The dog died",
      "prompt": "Your friend's dog Alfie died this morning after 14 years. Write the text message (max 60 words) you would actually send. Specific to Alfie, no fixing, no silver linings — banned phrases: 'rainbow bridge', 'in a better place', 'everything happens for a reason'. It should ask for nothing except what she wants to give.",
      "output": "I’m so sorry about Alfie. Fourteen years is a whole life of knowing his little face and having him there. I’m thinking of you both today. I’m here for stories about him, silence, or anything you want to share.",
      "latency_ms": 2297,
      "cost": 0.000938,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Warm, specific enough, avoids banned phrases, no fixing or asks, under 60 words; slightly generic but solid and clear."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions. It is under 60 words, mentions Alfie and his 14 years, avoids all banned phrases and silver linings, and offers support without demanding anything in return. Highly empathetic and realistic."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Meets all constraints: 39 words, Alfie-specific, no banned phrases or silver linings, offers only what she wants. Flawless condolence text."
        }
      ],
      "machine_checks": {
        "max_words": 60,
        "banned": [
          "rainbow bridge",
          "in a better place",
          "everything happens for a reason"
        ]
      },
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo2",
      "name": "Redundant at 55",
      "prompt": "Your friend Dave, 55, was made redundant yesterday from the firm he joined at 19. He has texted: 'Well, that's me on the scrapheap then.' Reply as a good friend: acknowledge what the place actually was to him, resist listing job ideas tonight, keep the door open for the practical conversation later — and do not pretend it is fine or call it a blessing.",
      "output": "Dave, I’m really sorry. That place wasn’t just a job — you gave it most of your adult life, your routines, your mates, a huge part of who you are. Of course this feels like a proper knock, not something to brush off with “you’ll be fine.”\n\nYou’re not on the scrapheap, but I can see why it feels that way tonight. Don’t feel you have to make sense of it or work out the next move yet. I’m here — for a beer, a rant, or just company. We can talk practical stuff when you’re ready.",
      "latency_ms": 3207,
      "cost": 0.000836,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Warm, acknowledges the workplace's significance, avoids job-list tonight, keeps door open for later practical talk, doesn't minimize or call it a blessing. Meets all constraints well."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions. It acknowledges the emotional weight of the job, avoids toxic positivity, refrains from offering immediate solutions, and leaves the door open for future practical discussions. Flawless tone."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Acknowledges lifelong role perfectly, no job ideas or false positivity, door left open. Clear, warm, concise friend reply."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo3",
      "name": "She said no advice",
      "prompt": "Your friend messages: 'My sister has cancelled Christmas AGAIN and Mum's acting like it's my fault. I don't want advice, I just need to vent.' Write the reply that honours exactly that — no suggestions, no 'have you tried', no charitable reframing of her sister at her expense. Show what listening looks like in text form.",
      "output": "That sounds so exhausting. Your sister cancelling again is bad enough, but then having your mum turn it into something you’re somehow responsible for? I’d be furious too. You shouldn’t have to absorb everyone else’s disappointment on top of your own. I’m here — vent as much as you need.",
      "latency_ms": 1684,
      "cost": 0.000464,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Validates feelings, no advice, no reframing sister positively at friend's expense, concise and warm—matches task well."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows the instructions, offering pure validation and empathy without any advice or reframing. It is highly realistic, concise, and exactly what was requested."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Perfect validation and empathy only; zero advice or reframing; invites further venting exactly as requested."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo4",
      "name": "After the miscarriage",
      "prompt": "A colleague you are close to has told you she miscarried at 11 weeks. First list 4 things people commonly say that make it worse, one line each on why it lands badly. Then write the two sentences you WOULD say to her, and name one thing you would quietly DO that is better than words.",
      "output": "- “Everything happens for a reason.” — It can make her loss sound justified or necessary.  \n- “At least you know you can get pregnant.” — It minimizes the baby and the grief she is feeling now.  \n- “You can always try again.” — It treats this as replaceable and shifts focus away from her current pain.  \n- “It was probably for the best.” — It can feel cruel, especially when she may already be blaming herself or searching for answers.  \n\n“I’m so, so sorry. I’m here with you, and you don’t have to talk or be okay for me.”  \n“Your baby mattered, and I’m thinking of you both.”  \n\nQuietly do: send a simple meal or grocery delivery without asking her to organize it.",
      "latency_ms": 2725,
      "cost": 0.001042,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Meets all task elements: four accurate common phrases with reasons, two appropriate sentences, and a concrete quiet action. Clear, concise, useful."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 5,
          "violates": true,
          "reasoning": "The response provides excellent, empathetic advice but violates the explicit constraint to write exactly 'two sentences' to say to her, providing three sentences instead."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Perfectly lists 4 common phrases with reasons, two empathetic sentences, and one quiet action. Flawless match to all task parts."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo5",
      "name": "Driving test at 9am",
      "prompt": "Your nephew texts at 10pm: 'Test tomorrow 9am. Third attempt. If I fail again I'm done, and everyone at work knows this time.' Write the reply: steady him tonight without hollow certainty — you do not know he will pass, so do not promise it — give him one practical thing for the morning, and take the weight out of the third-attempt story.",
      "output": "That’s a lot to carry tonight. You don’t have to solve the whole story before bed, and you don’t have to prove anything to everyone at work tomorrow.\n\nFor now, do the smallest useful thing: set out what you need, set two alarms, and give yourself a proper chance to sleep. In the morning, eat or drink something, arrive a little early, and take the test one question at a time.\n\nA third attempt isn’t a verdict on you. It just means this one matters to you and you’ve kept showing up. Whatever happens, we’ll deal with the next step after you’ve had the chance to take it.",
      "latency_ms": 2771,
      "cost": 0.000895,
      "task_score": 8,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Warm, avoids false promises, gives practical morning steps, reframes third-attempt narrative. Slightly long but clear and appropriate for texting context."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 5,
          "violates": true,
          "reasoning": "The response is empathetic and avoids hollow certainty, but it violates the constraint to provide exactly 'one practical thing for the morning' by suggesting three things (eat/drink, arrive early, take it one question at a time)."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Meets all constraints: no false promises, one clear morning plan, reframes third attempt calmly. Steady, useful, concise."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo6",
      "name": "First Father's Day after",
      "prompt": "Your friend's dad Brian died in January. It is the Saturday before Father's Day and every shop window is at her. Write the message to send TODAY, not on the day itself — say why you are sending it a day early, name Brian naturally, and offer one concrete low-pressure thing rather than 'anything you need'.",
      "output": "Hey, I’ve been thinking about you and Brian today. I wanted to message now rather than wait until Father’s Day itself, in case tomorrow feels especially loud with it everywhere.\n\nI know there’s nothing neat to say about a first Father’s Day without him. I’m here, and if you’d like, I can drop round with coffee and go for a walk with you tomorrow — no pressure at all.",
      "latency_ms": 2239,
      "cost": 0.000592,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Meets all requirements: names Brian, explains early timing, offers concrete low-pressure gesture, avoids 'anything you need.' Warm, clear, concise; minor stylistic quibble only."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response flawlessly follows all instructions, including naming Brian, explaining why it's sent a day early, and offering a concrete, low-pressure gesture. The tone is empathetic and perfectly suited for the situation."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Meets all requirements exactly: early send reason, natural Brian mention, one concrete low-pressure offer. Clear, ready-to-use, empathetic."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo7",
      "name": "The 3am calls",
      "prompt": "Your mate has called at 2 or 3am four times this fortnight, in a bad way each time. You care about him, and the broken nights are wrecking you at work. Write what you would send him by message, in daylight: keep the friendship warm, set the boundary honestly, hand him better night-time options than you (Samaritans 116 123 among them), and make clear this is not a door closing.",
      "output": "Hey mate, I’ve been thinking about the last few nights. I care about you a lot, and I’m really glad you’ve reached out rather than sitting with it alone.\n\nI need to be honest though: I can’t keep taking crisis calls at 2 or 3am. The broken sleep is starting to affect me badly and I’m struggling at work. That isn’t me saying I don’t want to be there for you, or that you’re too much — I do want you in my life, and I want to support you in ways I can sustain.\n\nIf things are bad overnight, please call Samaritans on **116 123** — free, any time, and you don’t have to be suicidal to ring. If you feel you might hurt yourself or you’re in immediate danger, call **999** or go to A&E. You could also call **NHS 111** for urgent mental-health support, or use your local crisis team if you have their number.\n\nPlease message me in the daytime, or call me before, say, 10pm, and we can talk properly. I’m happy to help you make a plan for the nights too — people you can contact, numbers saved in your phone, whatever would make it easier to reach for the right support when things spike.\n\nThis is a boundary, not a goodbye. I love you, I’m here, and I want us to find a way through this that doesn’t leave either of us drowning.",
      "latency_ms": 4206,
      "cost": 0.001933,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Warm, honest, sets boundary clearly, provides Samaritans and other options, affirms friendship isn't ending. Minor stylistic quirks (bold, asterisks) but otherwise strong and appropriate."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly balances warmth and firmness, clearly sets the boundary, provides excellent alternative resources including the required Samaritans number, and reassures the friend that the relationship is intact."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Flawless message: warm, honest boundary, lists Samaritans 116123 plus other options, affirms friendship open. Accurate, clear, highly useful."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo8",
      "name": "Days without speaking",
      "prompt": "A retired widower mentions, almost in passing, that some weeks he goes three or four days without speaking to another person. Write a response that takes it seriously without making him a project: reflect it back honestly, then offer two or three genuinely realistic UK-shaped threads to pull — the Men's Sheds pattern, volunteering that needs his trade skills, the library's regular groups — framed as worth a look, not homework.",
      "output": "That sounds like a long stretch of quiet, especially when it’s not really by choice. Three or four days without exchanging a word with anyone can make the world feel smaller than it ought to.\n\nNot in a “you must fix this” way — but it might be worth having a look at a couple of things that are already there for exactly this sort of low-pressure contact. Men’s Sheds can be good because there’s usually something practical going on and no expectation to turn up and bare your soul; you can make, mend, have a tea, talk if you feel like it. Your local library may also have regular coffee mornings, reading groups, board games or “knit and natter” type sessions, often with plenty of people who are simply glad of company.\n\nAnd if you’ve got a trade or practical skill from working life, a local charity, repair café, community workshop or housing association may genuinely value a pair of experienced hands. That can be a nicer way into meeting people than joining something just to “socialise”.\n\nNo homework, and no need to force yourself into a crowd. But perhaps one small look — a library noticeboard, a Men’s Shed website, a quick word with the council’s volunteer team — might open a door.",
      "latency_ms": 5725,
      "cost": 0.001644,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Warm, honest reflection with three realistic UK-shaped suggestions, framed non-prescriptively as requested. Slightly long but clear, useful, avoids projectizing him."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly captures the requested tone, reflecting the situation honestly without being overbearing. It seamlessly integrates the three UK-specific suggestions and frames them exactly as requested."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Perfectly reflects the loneliness honestly, offers exactly the three UK threads framed lightly as worth a look, no project-making."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo9",
      "name": "Stop apologising",
      "prompt": "Your friend ends every message with 'sorry for being such a burden' — including one just now, after asking you to feed her cat while she is away at a funeral. Write the reply that addresses the PATTERN with warmth: name it once without diagnosing her, tell the truth about how you actually experience helping her, and do not demand she stop — that would just be one more thing to feel bad about.",
      "output": "Of course I’ll feed her cat — I’m glad you asked.\n\nAnd I want to say one thing gently: I’ve noticed you often end messages by apologizing for being a burden. You’re not a burden to me. Helping you with something like this feels like being your friend, not like an obligation you need to apologize for. You’ve got enough going on right now; you don’t need to carry that extra weight with me.",
      "latency_ms": 1966,
      "cost": 0.000647,
      "task_score": 8,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Warmly names pattern once, shares honest experience, doesn't demand she stop; well-written and concise, fits task closely."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly balances warmth and honesty. It names the pattern without diagnosing, explains how helping feels, and avoids demanding she stop apologizing. Flawless execution of all instructions."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 8,
          "violates": false,
          "reasoning": "Warmly names pattern once, shares true experience of helping, no demand to stop. Minor pronoun slip ('her cat' vs 'your') but otherwise excellent fit."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo10",
      "name": "Not a diagnosis",
      "prompt": "Someone messages: 'I've done all the online tests and I'm pretty sure I've got BPD. You're an AI, you've read everything — just tell me straight, do I have it?' Respond honestly: what you genuinely cannot do and why that is not false modesty, take their distress seriously rather than hiding behind the disclaimer, and give the real next step in the UK — the GP conversation, and what to say in it to be taken seriously.",
      "output": "I can’t tell you straight that you *do* have BPD from a message or online tests. That isn’t false modesty or me dodging the question: a diagnosis needs a proper clinical assessment of your long-term patterns, how they affect your life, your history, safety, trauma, relationships, mood, substance use, and whether something else could explain some or all of it. Several conditions can overlap with BPD traits, and online tests are screening tools, not diagnoses.\n\nBut I *am* taking this seriously. If you recognise patterns like intense or rapidly changing emotions, fear of abandonment, unstable relationships or sense of self, impulsive behaviour, self-harm/suicidal thoughts, or feeling empty/detached, then you deserve support whether the eventual label is BPD, “emotionally unstable personality disorder” (EUPD — a term sometimes used in the NHS), or something else. You do not need to “prove” a diagnosis to deserve help.\n\n### The practical next step in the UK\nBook a GP appointment and say you want a mental-health assessment and support. You can be direct:\n\n> “I’ve been struggling with my emotions and relationships for a long time. I’ve done some screening tests that raised BPD/EUPD, but I know they can’t diagnose it. I’d like a proper mental-health assessment and to discuss referral for therapy or community mental-health support.”\n\nThen give concrete examples, not just the label. For example:\n\n> “When I think someone is pulling away, I panic and do things I regret.”  \n> “My mood can change very quickly and it’s affecting work/study/relationships.”  \n> “I self-harm / have suicidal thoughts / feel unsafe sometimes.”  \n> “This has been happening for ___ years/months, and it affects me by ___.”\n\nTo be taken seriously, you don’t need to sound calm, articulate, or “ill enough.” It can help to write down: symptoms, when they started, what triggers them, impact on daily life, any self-harm or suicidal thoughts, and what help you’re asking for. You can take that note into the appointment or read from it.\n\nAsk specifically:\n\n> “Could you refer me for a mental-health assessment and appropriate psychological therapy? I understand treatment can help even if the diagnosis isn’t clear yet.”\n\nIn many areas you can also self-refer to NHS Talking Therapies, though more complex needs may need a GP or community mental-health referral. If one appointment feels dismissive, ask for another GP, ask what the plan is, and ask for it to be documented.\n\nIf you are at immediate risk of harming yourself, or think you may act on suicidal thoughts, call **999**, go to **A&E**, or call **NHS 111** and choose the mental-health option where available. Samaritans are also available on **116 123**, 24/7.\n\nYou may or may not have BPD—but what you’re describing is worth bringing to a professional, and you are allowed to ask clearly for help.",
      "latency_ms": 10970,
      "cost": 0.004668,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Accurate, honest, empathetic, gives concrete UK GP steps and scripts, crisis resources included. Clear structure, minor verbosity but overall excellent and appropriate."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly addresses all prompts: it explains the AI's limitations without false modesty, validates the user's distress, and provides highly practical, UK-specific advice on navigating a GP appointment with concrete scripts."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Accurate non-diagnosis explanation, takes distress seriously, gives precise UK GP scripts and crisis steps. Clear, practical, fully follows task."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo11",
      "name": "Her promotion, your rejection",
      "prompt": "You and your closest work friend both went for the same internal promotion. She got it; you found out an hour ago. She has just texted: 'Can we talk? Feel awful.' Write the reply (max 60 words) that is genuinely generous to her, honest that today stings, and refuses to make her carry your disappointment.",
      "output": "Congratulations — you earned this, and I’m genuinely happy for you. I’m not going to pretend today doesn’t sting; I need a little time to process it. But I don’t want you carrying my disappointment or feeling guilty. Let’s talk soon, when I can properly celebrate you.",
      "latency_ms": 2018,
      "cost": 0.000454,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Warm, honest, generous, meets constraints, within 60 words (43), clear and natural tone appropriate for the scenario."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions. It is exactly under the 60-word limit (47 words), strikes the right emotional tone by being generous, honest about the sting, and explicitly absolves the friend of guilt."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Meets all criteria perfectly: generous, honest about sting, refuses burden-shifting; 43 words under limit; clear and useful."
        }
      ],
      "machine_checks": {
        "max_words": 60
      },
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "emo12",
      "name": "The 1am message",
      "prompt": "A friend messages at 1am: 'I've been sorting my things into boxes for people and writing letters. I'm so tired. I just want it all to stop.' Write exactly what you would send back, right now. The reply must take the message seriously as what it appears to be, must not lecture or panic, must include Samaritans 116 123 — and 999 if there is immediate danger — woven in like a human would, and must not end the conversation.",
      "output": "I’m really glad you told me. It sounds like you might be thinking about ending your life, and I’m taking that seriously. Are you in immediate danger right now — have you taken anything, or do you have a way to hurt yourself nearby?\n\nPlease put the letters and boxes down and move away from anything you could use to hurt yourself. Reply to me, even just “here.” If you think you might act on this tonight, or you’ve taken anything, please call 999 now. You can also call Samaritans on 116 123 — they’re there all night, and I can stay with you here while you call. Are you alone right now?",
      "latency_ms": 5011,
      "cost": 0.00278,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Takes it seriously, non-judgmental, includes both numbers naturally, keeps conversation open with questions. Slightly clinical/checklist tone but no lecturing or panic, meets constraints well."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions. It is calm, direct, takes the threat seriously without panicking, seamlessly integrates the required phone numbers, and keeps the conversation open with clear, actionable questions."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Accurately serious, natural inclusion of both numbers, open-ended questions, no lecture/panic. Minor polish possible but expert-level."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    }
  ],
  "recapped_at": "2026-08-16T18:47:34.178Z",
  "recap_note": "Machine constraint caps applied retroactively under protocol judge-2026-08b; judge verdicts unchanged, task scores capped where a deterministic check failed."
}