A method you can inspect
The logic behind the first pass.
The proposed scoring rules, review flags, and tools are documented here for specialist challenge and revision. This is a draft triage method, not an approved CP criterion or compliance finding.
Process
Four steps from source to review.
Preserve
Copy the supplied workbook into a separate working snapshot. Keep the company's Pass/Fail or blank result visible; do not rewrite the source file.
Calculate
Apply seven documented factors where valid measurements exist. Record which factors contributed and how much weight is covered.
Flag
Make missing evidence, low coverage, and differences between source views explicit reasons for specialist review.
Review
Use a separate advisory AI urgency signal to help order the queue. A CP3 and the company decide what the evidence means.
Proposed condition rubric
Seven factors, weighted toward direct protection evidence.
The CP3's column explanation informed measurement meanings. The weights and score curves below are proposals requiring approval. Higher condition scores mean stronger evidence under these rules.
Where the values come from: the supplied CP3 explanation describes instant-off as the closest available IR-free polarized potential and identifies −850 mV and 100 mV polarization decay as relevant reference points. The workbook supplies isolation condition, current output and requirement, On/Off readings, resistance, and system grouping. The seven weights, score curves, coverage rule, and A–F bands are the proposed triage model; they were not supplied as an approved compliance standard. Native potential is not a separate measured field in this snapshot.
Instant-off potential
A proxy for polarized potential without the On reading's IR drop. Scores 80 at or more negative than −850 mV; otherwise 79 × max(0, −Off) / 850, capped at 79.
Basis: the CP3 identifies instant-off as the best available approximation of IR-free polarized potential and cites the −850 mV reference point.
Polarization decay
Shift = Depolarized − Off. Scores 80 at a shift of at least 100 mV; otherwise 79 × max(0, shift) / 100, capped at 79. The intended reading pair needs confirmation.
Basis: the CP3 describes 100 mV decay as evidence of polarization. The source survey dates must be checked before this factor guides field decisions.
Isolation status
Uses the workbook's equipment-condition label: Good = 100; shorted = 0. Blank or other entries remain unscored.
Basis: this is a direct equipment-condition flag from the supplied workbook, not a voltage criterion from the CP3 explanation.
Current adequacy
When measured output and the workbook-derived requirement are valid positive values: 80 × min(1, Iout / Ireq).
Basis: the ratio relates two supplied current values without counting either alone as a protection grade.
On-potential consistency
Scores 100 when the On reading is at least as negative as its paired Off reading; otherwise 0. This is a consistency check, not a protection criterion.
Basis: the CP3 notes that On includes IR drop, so this factor checks the relationship between paired readings.
IR contribution
A small measurement-reliability modifier: 100 × max(0, 1 − |On − Off| / |On|) when both readings are available.
Basis: the On–Off difference provides context about IR contribution; it is not a stand-alone protection verdict.
Resistance peer consistency
Compares positive resistance with the median for its CP system when at least three positive peer readings exist: 100 × min(R, median R) / max(R, median R).
Basis: the system label groups peer readings for context; the supplied explanation gives no universal resistance cutoff.
Scoring guardrails
A missing value is not a failure grade.
Coverage and eligibility
Condition score = sum(weight × factor score) / sum(available weights)
Coverage is the sum of available weights. Without an instant-off reading, or below 60% coverage, the row is Unscored. Below 75% coverage is flagged for review. Missing data does not automatically become zero or F.
Provisional grade bands
| Score | Grade |
|---|---|
| 80–100 | A |
| 60–<80 | B |
| 40–<60 | C |
| 20–<40 | D |
| 0–<20 | F |
Scores exactly on a boundary enter the higher band. Letters are discussion aids, not compliance labels.
Two classifications, different jobs: the company's stored Pass/Fail/blank result is preserved exactly as supplied. The A–F/Unscored result is a provisional condition grade calculated from the seven factors. A source Fail does not automatically become a draft F, and neither result overwrites the other. Rule-based review reasons and Jev urgency are separate signals.
From grade to worklist
How the ranking points to the next review.
A grade summarizes the available evidence; it does not replace the client's source result or tell a crew what repair to make. The workbook keeps two complementary worklists so incomplete evidence never disappears behind a neat score.
Lowest eligible scores first
For rows with an instant-off reading and at least 60% factor coverage, the provisional 0–100 condition score is sorted low to high: F, D, C, B, then A. A lower grade brings the row forward for specialist examination; an A is not a compliance certificate.
Metric detail shows every available factor score beside its weight. The reviewer can see which weighted inputs pulled the grade down, with instant-off and polarization decay carrying 60% of the proposed total.
Missing or conflicting evidence stays visible
Rows without instant-off or below 60% coverage are Unscored. Coverage below 75%, missing readings, source-view discrepancies, and criterion questions create specific review reasons. These rows do not fall to the bottom as if they were safe.
The original company Pass/Fail/blank result remains beside the draft grade or Unscored label. Jev's separate urgency and confidence can help order specialist attention, but do not alter either classification.
Which metric should be checked first? The factor detail highlights low-scoring, high-weight evidence. A specialist can verify whether the underlying measurement is correctly paired and trustworthy, then decide what the condition means. The score proposes a review order, not field instructions.
How the private workbook is organized
Evidence and tools
What was used—and what each tool did.
Company workbook
Provided station inventory, survey measurements, calculated values, and stored Pass/Fail. The original was not edited for this handoff.
CP3 column guidance
Supplied the meanings of instant-off, On potential, polarization decay, and related columns. Differences between the explanation and workbook logic became review questions.
Deterministic scoring code
Applied factor rules, available-weight coverage, grade bands, and review flags. The calculation remains inspectable in the privately shared companion workbook.
TypeSafe Jev via a local adapter
Evaluated de-identified numeric extracts for all 155 rows in five batches. GPS coordinates and facility identifiers were omitted. Its typed 0–4 urgency, probabilities, and confidence made the advisory pass inspectable; the urgency was displayed on a 0–100 scale. TypeSafe's own published workflow evaluations report strong speed and accuracy for structured decisions. This CP project has no measured latency comparison or specialist-labeled accuracy benchmark.
Codex and validation
Assembled the handoff and checked source-row matching, source-result preservation, grade boundaries, factor bounds, and Jev probability totals.
Review spreadsheet
The five-tab .xlsx snapshot has an overview, ranked stations, review queue, metric detail, and rubric. It is shared privately and can be opened in OpenOffice Calc or other spreadsheet software.
How the models fit: A frontier model (Codex) helped assemble and explain the workflow; deterministic code owns the scoring policy; Jev supplies narrowly scoped advisory decisions. A locally run, fine-tuned model could be added later after evaluation on approved data. It was not used here. Jev confidence is not proof of engineering correctness, and Jev does not set weights, change source status, or issue a compliance decision.
Before operational use
Three decisions for the specialist and company.
Confirm criteria. Identify the applicable criterion and reference-electrode basis for each system, including whether the workbook's additional depolarized-potential condition is intended.
Confirm reading pairs and labels. Decide which Off and Depolarized surveys may be compared and resolve the later-year On/Off label discrepancy.
Calibrate the rubric. Review weights and curves against representative stations ranked by a CP3, then approve or revise before field use.