{
  "schema": "sim-proof/1",
  "pr": 2591,
  "issue": 2528,
  "commit": "82c676608",
  "tier": "T3",
  "criteria": [
    {
      "id": "A1",
      "text": "The oracle from PR 2587 still shows 0 differences, with cases where an asked id equals an offered thing walk's label, offered and not offered.",
      "status": "met",
      "evidence": [
        "`node --test apps/server/src/edge/boot/world-places-oracle.test.ts` -> pass, 0 differences: Eden 2026, about 2,200 cases at arrival and again after one walk, plus 117 and 105 asked ids that are a known place's id and an offered thing walk's label (exact, after 'the', other case) and 40 that are the label of a thing walk offered at arrival and not offered after the walk"
      ]
    },
    {
      "id": "A2",
      "text": "explicitPlaceCapability ms per act at 50 Souls down at least 50 % against base, with no overlap between repeats.",
      "status": "unmet",
      "evidence": [
        "https://github.com/soulsimulation/souls/actions/runs/43848768342",
        "https://github.com/soulsimulation/souls/actions/runs/43848772778",
        "`LOAD.md Under explicitPlaceCapability, 50 Souls, ms per act` -> base 47.9, 34.3 (mean 41.1); candidate 52.7, 74.3 (mean 63.5): +55 %, repeats overlap; no gain",
        "offeredWalkObject at 50 Souls: base 25.1, 20.8 -> candidate 28.6, 33.8; at 24 Souls 13.9, 11.5 -> 9.3, 6.2 (explicitPlaceCapability at 24 Souls 24.8, 23.7 -> 22.7, 19.4, reported only)",
        "finding: skipping the per-walk standing points and route checks (walkKeep) leaves offeredWalkObject's cost under load at 50 Souls; what remains is the focused walk build's fixed part (the native capability read and the nav ranking every walk build runs), not the walks it skips"
      ]
    },
    {
      "id": "A3",
      "text": "eval-scripted --battery m4-v1 --seat scripted: same verdicts and receipts as base.",
      "status": "met",
      "evidence": [
        "https://github.com/soulsimulation/souls/actions/runs/43848777197",
        "https://github.com/soulsimulation/souls/actions/runs/43848782001",
        "`result.json per episode: succeeded, why, endedBy, card state/outcome/closed/acked` -> 12/12 the same, base bcbac140b against candidate b3b153469",
        "receipts: eat:eat.e_2accdf4c started and play-music:play-music.e_18fdb81d started on both; the other differences are admission retries only (world.busy, soul.busy), which write no receipt"
      ]
    }
  ],
  "media": [],
  "files": [
    {
      "file": "LOAD-base.md",
      "caption": "eval-load base, run 43848768342",
      "url": "https://assets.eden.art/proofs/pr-2591/LOAD-base.md"
    },
    {
      "file": "LOAD-cand.md",
      "caption": "eval-load candidate, run 43848772778",
      "url": "https://assets.eden.art/proofs/pr-2591/LOAD-cand.md"
    }
  ],
  "checks": [
    {
      "cmd": "node --test apps/server/src/edge/boot/world-places-oracle.test.ts",
      "result": "pass",
      "note": "0 differences"
    },
    {
      "cmd": "node --test apps/server/src/w1-legacy-compatibility.test.ts apps/server/src/edge-resolve2-resend.test.ts apps/server/src/god/walk-verb-offers.test.ts",
      "result": "pass",
      "note": "6/6"
    },
    {
      "cmd": "tsc --noEmit -p apps/server",
      "result": "pass",
      "note": ""
    },
    {
      "cmd": "node .github/scripts/guards.mjs scrub --base origin/main",
      "result": "pass",
      "note": "0 problems"
    }
  ],
  "failedAttempts": [
    {
      "what": "design step 3 by a narrower walk focus (walkKeep)",
      "why": "A2 unmet: no gain at 50 Souls; offeredWalkObject's cost stays",
      "evidence": "https://github.com/soulsimulation/souls/actions/runs/43848772778"
    }
  ],
  "externalities": [
    {
      "what": "each act on a far place by id adds one narrow focus build to the Soul's focus cache (at most 8 per Soul)",
      "result": "can evict other kept focuses sooner",
      "ok": true
    }
  ],
  "limits": [
    "Even arm only: eval-load has no shape flag on timed-run's allowlist.",
    "Every arm queued about 19.4 s per act (p50) and achieved 19 to 27 % of intended acts at 50 Souls, with both V0 fixes merged in.",
    "Candidate repeat 2 at 50 Souls answered only 71 acts in the window and entered red once."
  ],
  "playtest": {
    "scenario": "eval-load --souls 24,50 --act-sec 10 --duration-sec 120 --profile --menu-stats, and eval-scripted m4-v1",
    "arms": [
      {
        "name": "base eval-load",
        "repeats": 2,
        "passed": true,
        "url": "https://github.com/soulsimulation/souls/actions/runs/43848768342"
      },
      {
        "name": "candidate eval-load",
        "repeats": 2,
        "passed": true,
        "url": "https://github.com/soulsimulation/souls/actions/runs/43848772778"
      },
      {
        "name": "base m4-v1",
        "repeats": 1,
        "passed": true,
        "url": "https://github.com/soulsimulation/souls/actions/runs/43848777197"
      },
      {
        "name": "candidate m4-v1",
        "repeats": 1,
        "passed": true,
        "url": "https://github.com/soulsimulation/souls/actions/runs/43848782001"
      }
    ]
  }
}
