The tower you have not climbed yet.
Five drills. You leave with a printed maintenance plan, and five things for the section wall.
Three things you write after every patrol. Watch what the model gets right, and what it invents.
A number between 0 and 100 arrives against a transformer. Before you can use it you have to know how it got there — so watch one assemble, input by input. Then look at the one that scored 82.
Nine AI inspection findings from one route. Decide which you would act on. Read all nine before you decide — the list is not sorted by what matters, and it is not the whole of what the flight saw.
A clean drone report is not an inspected line. The model reports what it found in the frames it could read. Where the frame was poor it returns nothing, and nothing looks exactly like a tower with no defect. Coverage is a question about the flight; confidence is a question about the image. Neither is a question about the asset.
Eight things that come off a drone flight or a substation floor. For each, decide whether it can go into an AI tool at all.
A draft letter that leaks is an embarrassment. A folder of geo-tagged photographs of EHT towers and substation yards is imagery of critical national infrastructure, with coordinates attached, uploaded to a service that keeps what it is given. That is a different order of problem, and this is the one track where the everyday work product is the sensitive material.
This is not a new imposition. TGTRANSCO already runs Cyber Suraksha as a standing channel for employee IT-security awareness — the organisation long ago accepted that an employee's everyday computing habits are a security matter. This module extends that instinct by exactly one step: the camera roll is now also an exit route for asset data.
This module teaches the instinct, not the policy. Where any specific item sits under TGTRANSCO's own classification is for the IT wing to say. What is being drilled here is the habit of asking the question before the upload, so that when the policy does land you already know why it says what it says.
From findings to a plan. Your call, your reason, your name. Tap a bucket for each item. Anything you mark urgent needs a reason before the plan will print.
Six route segments with a risk score against each. The exercise is not "clear the highest score first." Read the whole table — access, species, growth rate, permission — then decide what each segment's date is actually set by.
AI flags and ranks. A named engineer plans and despatches.
Everything you assigned today, bucketed, with your reasons. Print it and take it to the section meeting. Nothing is stored in the browser — print before you close the tab.
A score is a ranking, not a prediction. 36 means look here first. It does not mean 36 days or 36 percent.
Five grades. Four of them the model can guess at; the fifth is yours alone.
| Grade | Meaning | Action | Timeline |
|---|---|---|---|
| S1 — Immediate | Failure imminent, or safety of persons at risk | Act on the next available outage window; inform the control room | Same-day decision |
| S2 — Urgent | Will degrade to S1 within the cycle | Schedule within the month | Named in the plan |
| S3 — Routine | Real defect, stable, no near-term progression | Next planned maintenance | Cycle |
| S4 — Monitor | Observed, not yet a defect | Re-observe at next patrol | Next patrol |
| R — Re-check | The finding cannot be graded from the evidence available | Manual or close-range verification before grading | Before the plan is signed |
The score is the start of the decision, not the decision.
Five instructions. Number two is the one that keeps an invented specification out of a store issue.