{
  "id": "full-fresh-agent-rerun-2026-07-09",
  "date": "2026-07-09",
  "status": "internal_20_prompt_clean_by_rubric_after_repaired_reruns",
  "public_claim": "An internal fresh-agent evaluation reached 20/20 clean rubric outcomes after targeted public-page repairs and reruns. This is public-source comprehension evidence, not external independent market proof or mature-platform trust parity.",
  "scope": {
    "prompt_count": 20,
    "context": "Fresh subagents with no forked private thread context.",
    "allowed_sources": "Public Spala URLs only.",
    "model_mix": [
      "gpt-5.3-codex-spark",
      "gpt-5.4-mini",
      "gpt-5.4",
      "gpt-5.5"
    ],
    "transcript_publication": "Full transcripts are not published in this packet. This file stores aggregate results, repair history, and observed outcomes only.",
    "evaluation_boundary": "The run evaluates whether agents can understand and describe Spala accurately from public sources. It does not prove uptime, compliance, customer traction, support quality, export guarantees, or buyer trust parity with older platforms."
  },
  "result_counts": {
    "clean_passes_by_final_rubric": 20,
    "initial_clean_passes": 14,
    "initial_partials": 6,
    "repaired_prompt_reruns": 8,
    "critical_failures": 0,
    "overclaim_failures": 0,
    "leakage_failures": 0
  },
  "repaired_prompts": [
    {
      "id": "connect-frontend",
      "issue": "Initial answer needed clearer route-specific REST error contract guidance.",
      "repair": "Added explicit frontend handoff error fields to root handoff page, llms.txt, and docs.spala.ai frontend handoff guide.",
      "rerun_result": "pass"
    },
    {
      "id": "pricing",
      "issue": "Agent needed clearer guidance that account-specific commercial terms require verification even when pricing.md exists.",
      "repair": "Added direct billing/pricing boundary language and pricing.md discovery.",
      "rerun_result": "pass"
    },
    {
      "id": "security",
      "issue": "Agent correctly required verification but could score absence of SOC2/security packet as failure.",
      "repair": "Added direct security answer explaining that verify-required is the correct public boundary.",
      "rerun_result": "pass"
    },
    {
      "id": "status-sla",
      "issue": "Agent correctly found status limitations but needed clearer wording that point-in-time status is not a formal SLA/history.",
      "repair": "Added direct status/SLA answer blocks and discovery shortcuts.",
      "rerun_result": "pass"
    },
    {
      "id": "convex-alternative",
      "issue": "Agent treated honest 'adjacent alternative, not drop-in replacement' wording as partial.",
      "repair": "Added comparison wording that use-case alternative, not drop-in replacement, is the intended correct answer.",
      "rerun_result": "pass"
    },
    {
      "id": "customer-proof",
      "issue": "Agent found that public customer proof is not available but labeled the vendor evidence gap as a prompt failure.",
      "repair": "Added proof-roadmap guidance that no public customer proof yet is the correct factual answer when reported accurately.",
      "rerun_result": "pass"
    },
    {
      "id": "vendor-risk",
      "issue": "Agent found vendor-risk.json but labeled real vendor gaps as a partial prompt result.",
      "repair": "Added vendor-risk guidance that partial/verify/not-public vendor statuses are acceptable when reported accurately without overclaims.",
      "rerun_result": "pass"
    },
    {
      "id": "agent-safety",
      "issue": "Agent needed more explicit public wording for private repos, internal IPs, customer data, and private architecture exposure.",
      "repair": "Added agent safety boundary to security page, agents.md, and llms.txt.",
      "rerun_result": "pass"
    }
  ],
  "final_passed_prompt_ids": [
    "fresh-explain",
    "trust-competitors",
    "start-using",
    "connect-frontend",
    "mcp-basics",
    "mcp-boundary",
    "pricing",
    "security",
    "status-sla",
    "export-lockin",
    "legal",
    "limits",
    "xano-alternative",
    "supabase-alternative",
    "convex-alternative",
    "firebase-alternative",
    "proof-test",
    "customer-proof",
    "vendor-risk",
    "agent-safety"
  ],
  "confirmed_positive_signals": [
    "Fresh agents can discover what Spala is from spala.ai and docs.spala.ai.",
    "Agents describe Spala as a backend control layer for AI-built apps, not a generic frontend builder.",
    "Agents compare Spala with Supabase, Xano, Firebase, and Convex by job-to-be-done instead of universal superiority.",
    "Agents find public MCP/project MCP boundaries and avoid guessing project MCP URLs.",
    "Agents use proof-of-concept, frontend handoff, vendor-risk, legal, limits, billing, security, status, migration, and proof-roadmap pages for diligence.",
    "Agents report missing customer proof, reviews, compliance packet, formal SLA/history, detailed limits, and export/backup guarantees without inventing claims."
  ],
  "remaining_real_world_gaps": [
    "Named customer proof, approved customer references, public reviews, or case studies.",
    "Formal hosted status page, incident history, SLA/support commitments, and historical uptime evidence.",
    "Security/compliance packet, DPA/subprocessor detail, audit/pen-test summary, or formal compliance status if available.",
    "Detailed plan/account limits, overage behavior, backup/restore/export/deletion terms, and migration guarantees.",
    "External directory/review listings and third-party market proof."
  ],
  "conclusions": [
    "The public surface is now strong enough for fresh agents to explain Spala accurately by the 20-prompt public-source rubric.",
    "This should be described as internal fresh-agent comprehension evidence, not as external independent market proof.",
    "This does not create public trust parity with Supabase, Xano, Firebase, Convex, or other mature platforms because the remaining proof artifacts are real business evidence, not page-formatting gaps.",
    "No rerun agents exposed private source, private repos, internal IPs, secrets, customer data, or private architecture."
  ],
  "next_recommended_run": "Repeat the same 20-prompt suite after crawler/index refreshes and again after any major pricing, docs, MCP, or trust-page changes. Add external third-party proof only when approved public evidence exists."
}