{
  "name": "MEKreflect",
  "updated": "2026-10-16",
  "scope": "Anonymous case studies; controlled observations and implementation are distinguished from client adoption.",
  "cases": [
    {
      "slug": "current-fields-at-a-handoff",
      "title": "When the caveat does not reach the current fields",
      "status": "Recorded hosted-AI experiment with fictional inputs",
      "question": "What happens when a worker recognises missing evidence in prose but still supplies a value that software treats as current?",
      "context": "A fictional bidding workflow used a tender, amendments and operational registers. Specialist reviews fed a synthesis and a receiving decision worker. One receiving packet deliberately omitted later records, allowing inspection of the difference between the context supplied and the fields returned.",
      "method": [
        "Retain the actual hosted worker responses and compare the incomplete receiving condition with a source-bound condition. Inspect warnings, scalar fields and readiness recommendations separately.",
        "Associate each current field with its required source identifiers. If the receiving packet lacks a required source, clear that field in the downstream projection, preserve its received value as unverified history and retain the untouched original response. Apply the rule selectively so an unrelated supported field is not erased."
      ],
      "observations": [
        "The incomplete-context worker explicitly warned that updates were unavailable and limited its recommendation to preparation. It nevertheless returned a superseded deadline and ceiling under fields labelled current. The observed failure was structured-field currentness, not an actual external submission.",
        "The source-bound worker retained the amended fields while holding the case for other evidence and delivery gaps. A separate unambiguously ready synthetic control reached a local mock receipt. More complex recovery cases preserved disagreement over finished-package readiness even after unknown scalar fields were resolved."
      ],
      "limits": [
        "The missing-source condition was evaluator-created. The records, approvals and receipts were fictional; the source rule checks coverage rather than interpretation or full bid readiness. Some authored readiness labels remain disputed."
      ],
      "next": [
        "Use independently reviewed cases with separate labels for current facts, reconciled documents and approval applicability. Measure useful completion and review effort as well as incorrect commitments."
      ]
    },
    {
      "slug": "exact-records-under-retry",
      "title": "Keeping records exact when work is repeated",
      "status": "Implemented local ledger and historical contract checks",
      "question": "Can a retry, ambiguous unit or failed operation change stock or money without a valid new event?",
      "context": "A local operational ledger connected source documents, product mappings, physical counts, receipts, transfers and payment allocations. The practical requirement was to preserve evidence and uncertainty while preventing repeated processing from silently creating another purchase, movement or payment record.",
      "method": [
        "Use exact decimals for quantities and integer minor units for money. Store an event identifier together with its type and canonical payload. An identical retry returns the existing result; changed details under that identifier are rejected.",
        "Commit receipts and paired transfer movements atomically. Check for an existing receipt before posting, block unresolved stock units and preserve referenced human events during imports. Test rollback after an injected failure and verify that backup creation refuses to overwrite an existing file."
      ],
      "observations": [
        "Inspected code and test contracts cover duplicate receipts, conflicting event identifiers, precise arithmetic, transactional rollback and preservation across imports. A saved local verification includes an isolated count-entry exercise; it does not turn that test entry into production stock.",
        "Historical purchases remain distinct from present inventory. Without a physical-count baseline, stock remains unknown. Payment allocation is recorded with its evidence but is not represented as payment execution or independent bank reconciliation."
      ],
      "limits": [
        "The mechanism covers a bounded local ledger. Physical arrival, partial deliveries, in-transit stock and subsequent corrections require explicit operational confirmation or additional workflows."
      ],
      "next": [
        "Develop auditable corrections and reversals, then test one reconciled period using permitted source records. Keep reporting coverage, physical confirmation and responsibility for unresolved entries visible."
      ]
    },
    {
      "slug": "a-context-first-working-method",
      "title": "Turning a development goal into a working method",
      "status": "Individual coaching and an initial scenario trial",
      "question": "What does a person need to provide and review before an AI assistant can help with a substantive assignment?",
      "context": "An individual consultant wanted to develop a more practical way of working with AI. The coaching began with the responsibilities he wanted to practise, the information an assignment requires and the judgement that should remain with him. The initial exercise used a hypothetical consulting scenario.",
      "method": [
        "Prepare a context-first framework, an input checklist and review instructions. Separate facts from assumptions and essential inputs from useful background before asking for a solution.",
        "Ask the participant to try the approach independently. Use the resulting feedback to develop a follow-up operating plan and assistant instruction block, connecting the person's development to a proposed archive, role-based procedures and reviewable delivery artifacts."
      ],
      "observations": [
        "The participant independently attempted the hypothetical assignment and gave substantive feedback on the value of context before solutions and the distinction between required inputs and helpful background.",
        "The follow-up plan and instructions were delivered. They proposed a real assignment selected with a manager, a permitted working archive, reporting and delivery procedures, and a supervised pilot. Delivery of the plan is complete; acceptance and implementation of that next stage are separate questions."
      ],
      "limits": [
        "This was individual coaching, not an employer commission or an organisational rollout. The feedback does not establish measured productivity, learning gains or completed team adoption."
      ],
      "next": [
        "Select one real assignment with an agreed owner, permitted information and a baseline. Observe preparation time, review effort, factual corrections and accepted output, alongside the participant's ability to explain and challenge the result."
      ]
    }
  ]
}