Certification
    June 9, 2026

    CCA · Page 6 — Domain 5: Context Management & Reliability

    Exam-ready notes for CCA Foundations Domain 5 (15%): context-window management, escalation, error propagation, confidence calibration, and provenance.

    Share

    Domain 5 — Context Management & Reliability

    CCA Foundations course · Page 6 of 8 · ← Back to all courses · Weight: 15%. Mark complete at the bottom to advance.


    5.1 — Context Window Management · Core

    • Case-facts block at the TOP, verbatim — and it must survive /compact (don't let summarization drop exact amounts/dates).
    • Lost-in-the-middle effect: put key info at the beginning/end of long inputs.
    • Trim verbose tool output to only relevant fields (e.g. 40 fields → 5) before it accumulates.
    • Use scratchpad files to persist findings during long exploration.

    5.2 — Escalation Patterns · Core

    • Explicit human request = immediate escalation, no exceptions (don't attempt resolution first).
    • Policy gap / silence on the request = escalate.
    • Sentiment-based or self-reported-confidence escalation = anti-pattern (poor proxies for complexity).

    5.3 — Error Propagation · Core

    • Silent suppression AND full abort are both wrong.
    • Return a structured error with partialResults — preserve successful work and annotate the gaps.

    ⚠ Often-missed — Human Review & Confidence Calibration · Gap

    • Raw confidence ≠ accuracy. Calibrate against a labeled validation set (e.g. via ECE), then route on the calibrated probability.
    • Blocking review for irreversible actions; async review for reversible ones.

    5.6 — Provenance & Conflicting Sources · Core

    • Mandatory claim-source mappings (source URL + date) preserved through synthesis.
    • Conflicting statistics → annotate BOTH with attribution; NEVER pick one or average them.
    • A temporal difference is not a contradiction (require publication/collection dates).

    Exam reflexes for Domain 5

    • "Context filling up with exact values" → case-facts block at TOP of every prompt.
    • "Explicit human request" → escalate immediately, no resolution attempt.
    • "Conflicting statistics" → annotate both with source + date; never pick or average.
    • "Timeout returns empty success" → structured error with isError / partialResults.
    • "Confidence score for routing" → calibrate first (ECE), route on calibrated probability.

    Ask about this article

    Get answers grounded in this post. AI-generated — based on this article, and may be imperfect.

    Scaled AI Weekly

    Enjoyed this? Get more like it every Monday.

    Real architecture decisions, LLMOps patterns that survive production, and engineering leadership advice — from 12+ years of building at enterprise scale. Free. No spam. Unsubscribe anytime.

    Join engineers building production AI systems