NERC GADS reporting with every outage coded to the failed part and checked before it is filed
Outage events coded right for GADS and the board.
See one case, screen by screen ↓Each quarter, a NERC-registered generating fleet reports every outage, derate and reserve shutdown with its event type, its start and end, the MW lost and a cause code that names the failed equipment. The file is due 45 days after the quarter ends. The raw material was never built for reporting. The breaker and MW signals sit in the historian, the reason sits in the control-room shift log, and the failed part sits in a maintenance work order. Availability analysts join the three by hand, in spreadsheets, often days or weeks after the event.
The coding is rule-heavy. The event type depends on timing — immediate U1, delayed U2 within six hours, postponed U3, start-up failure SF. The cause code must name the part that failed, not the alarm that tripped the unit. An extension to a maintenance outage is allowed only for work in the original scope, and solar events count only at 20 MW or more. So steam codes land on gas-turbine trips, forced outages hide as maintenance extensions, fuel curtailments are charged to the plant, and the same failure is coded three ways, so the board never sees one problem. A wrong code moves the forced outage rate, which feeds capacity accreditation, regulator benchmarks and availability-linked payments.
Estimated split for one quarter of a six-unit fleet with about 60 events, in working days, by hand and with the solution.
Six agents build each event, read the log and the work order, propose the cause code and check it; the analyst confirms.
Lumina Utilities runs six units, 1,700 MW: Harrow Point 1–4 (steam), Kestrel Bay CC 1 (combined cycle) and Brannock Solar. On Sunday 20 September 2026 at 02:47, Harrow Point 3's generator breaker opens on a waterwall tube leak. Three weeks later, with the Q3 file due on Monday 16 November, here is how that outage — and the rest of the quarter — gets coded, checked and signed, screen by screen, in the working solution.
Ines codes the fleet's events. Each unit's 2026 is one strip of states — in service, planned, maintenance and forced outages, derates drawn by MW lost, reserve shutdown — with each row's equivalent availability and forced outage rate at the end. The Q3 bar reads 62 events, 55 coded, 3 checks to fix, 34 days left. The biggest uncoded block, first on the To code list, is Harrow Point 3.
“To code · Biggest lost energy first · every proposal waits for you”
The event builder turned the breaker and output signals into one event: Sun 20 Sep 02:47 to Wed 23 Sep 19:15, 88.5 h, 330 MW, 29,194 MWh. On the signal chart, feedwater minus steam flow climbs to +18 t/h — the leak shows 22 minutes before the breaker opens — and each shift log entry is pinned at its minute.
Limit: never invents a start or end time; signals that disagree go to the analyst.
The log reader picked out seven entries from 01:58. Sara Duarte at 02:31: “Feedwater flow 18 t/h above steam flow.” At 02:38: “Acoustic leak alarm ch 7, furnace front wall el. 38 m.” Joel Mbeki, shift manager, at 02:47: “Waterwall tube leak suspected, front wall.” The work order matcher linked WO-3-55812: tube 114 pinhole leak, tube 115 steam-cut, front wall, el. 38.2 m, radiography 100 % accepted.
Limit: the log reader quotes the log and never paraphrases a cause the log does not state.
The cause coder proposes the type from the timing and the code from the failed part. The unit came off at once on the shift manager's order, so it is an immediate forced outage, U1. The code is 1000, citing the log and the work order, and it notes this is the third front-wall leak on the unit this year. Two alternatives are kept: 1040 Waterwall, fireside slagging or fouling at 4 %, and 1999 Boiler, miscellaneous at 2 %. The record already passes 9 of 9 checks.
“Pick one, or any code in the library; your choice and the reason are kept with the record.”
Ines confirms U1 · 1000. The lens now shows the code on file, “Confirmed by you just now”. On the To code list, Harrow Point 3 is greyed out as coded by her, and the Q3 count falls to 6 left. Undo is beside the confirmation, and Next to code moves her to the next-biggest event.
“Filed into the Q3 record with its rationale. 6 left to code.”
Kestrel Bay CC 1 lost 160 MW for 8.8 h on Tue 4 Aug. The log reader found Farah Quist's 12:58 entry: curtailment notice CN-2026-118 from Corran Gas Transmission, offtake pressure held at 24 bar from 13:15. There is no plant work order, so the matcher attached the supplier's notice instead. The proposal is D1 · 9130 Lack of fuel (outside plant control), at 88 % — and it still waits for Ines.
“Output was held by supplier gas pressure, not by plant equipment.”
The report checker ran the 12 rules from the reporting instructions and Lumina's coding rules on all 62 Q3 events: 8 of 12 pass. It caught three records. A 36 h maintenance extension on Harrow Point 1 was for a crack found during the outage, not in the scope. A Kestrel Bay gas-turbine flame-out on 23 Jul carries 4260, a steam turbine code; the cause coder proposes 5130 Gas turbine combustor. And an 18 MW Brannock tracker stow is under the 20 MW solar threshold.
Limit: the report checker proposes a fix for each failure and never changes a record on its own.
Under rule ER-3, Ines recodes the Harrow Point 1 extension as U1 · 1400, with WO-1-40227 behind it. Rule 03 now passes — 9 of 12 — and the quarter has 8 left to clear. In the Q3 report, Harrow Point 1's forced outage rate moves from 0.4 % to 2.1 %, and the fleet's from 2.2 % to 2.4 %.
“Harrow Point 1’s Q3 forced outage rate rises; that is the honest number.”
The report writer computes service, planned, maintenance, forced and reserve hours, equivalent availability, forced outage rate and forced MWh by unit: fleet equivalent availability 97.2 %, capacity-weighted. It drafts the board summary for Margit Halvorsen, the reporting lead. Send for sign-off stays blocked until every record is coded and passes all 12 checks.
“Every figure links to its events. Drafted by the report writer from the coded records; Margit Halvorsen signs.”
Coded to the failed part, Harrow Point 3's three waterwall leaks — 3 Mar, 14 Jun and 20 Sep — add up to 201.8 h and 66,610 MWh, the largest cause in the fleet. The tube map shows all three on the front wall within 2 m of each other, near burner row 3. Ines sends the pattern to Callum, who decides whether to survey the front wall at the November maintenance outage.
“Coded to the failed part, the three outages read as one problem, not three bad nights.”
The same outage, seen by the four people who write it, code it, investigate it and sign it — what their quarter looked like, and what it looks like now.
Six agents build each event, read the log and the work order, propose the cause code and check it; the analyst confirms.
Turns breaker, output and status signals from the historian into events at 1-minute resolution: start, end, MW lost and a first event type. An event opens when the unit leaves service or loses more than 2 % of capacity for 15 minutes.
Reads the control-room shift log from 2 hours before each event to its end and picks out the equipment named, the symptoms, alarms and orders given, with their times.
Links the maintenance work orders on the same unit in the event window, matched on the equipment the log names, and reports the failed part and the repair. For a cause outside the plant, it attaches the supplier or system operator notice instead.
Proposes the event type by the timing rules and the cause code by the failed part, using the coding rules and the code library, with a rationale citing the log and the work order, a confidence and two alternatives. Below 80 % it shows the event without a proposal.
Runs 9 checks on every event and 12 rules on the quarter — type, code family, failed part, times, MW, overlaps, thresholds, contributing conditions, hours — names each failure and proposes a fix.
Computes equivalent availability, forced outage rate and lost energy per unit and for the fleet, and drafts the board summary: the result, the biggest causes, what changed.
Confirms each code; the reporting lead signs the file. The agents propose; a named person decides.
Ask about an event, a code, a check or the quarter — every answer cites the log, the work order or the rule.
Why is Harrow Point 3’s 20 Sep outage proposed as 1000?
The night-shift log names a waterwall tube leak on the front wall at 02:38 and 02:47, and WO-3-55812 repaired tubes 114–115 at el. 38.2 m. The boiler came off at once on the shift manager’s order, which makes it an immediate forced outage, U1. It is the third front-wall leak on the unit this year.
What is left before Q3 can be filed?
Seven events to code and three failed checks: an outage extension on Harrow Point 1 for work outside the scope, a steam code on a Kestrel Bay gas turbine trip, and an 18 MW solar event under the 20 MW threshold. The file is due Monday 16 Nov, because 15 Nov is a Sunday.
Which cause cost the most energy this year?
Waterwall tube leaks on Harrow Point 3: three forced outages, 201.8 hours and 66,610 MWh, all on the front wall within 2 m of each other. That is the largest cause in the fleet.
Is the Kestrel Bay derate on 4 Aug outside plant control?
Yes. Corran Gas Transmission held the offtake at 24 bar from 13:15, and the block was on its fuel pressure limit with no plant fault and no work order. The proposal is D1 · 9130, lack of fuel outside plant control, with the notice attached as evidence. You confirm it.
13 screens from the working solution, on its sample data. Pick one to see it large.
Deadlines, thresholds, units and who signs are settings. Changes take effect after the reporting lead approves them.
| Setting | Default | Choose from |
|---|---|---|
| Quarter due | GADS | 45 days after the quarter ends; next business day if a weekend |
| Lumina sign-off | Fri 6 Nov | The reporting lead signs 10 days before the due date |
| Board pack | Mon 23 Nov | Quarterly, with the GADS figures |
| Propose a code at | 80 % | Below this confidence the event is shown without a proposal |
| Confirm codes | Always | Every code is confirmed by an analyst; nothing is filed on a proposal |
| Repeat-cause alert | 3 events | Same code on the same unit within 12 months |
| Solar event threshold | 20 MW | Smaller solar events stay in Lumina’s own log |
| Contributing conditions | On | Asked for when the log names weather, fuel or the grid |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample fleet data · “target” = the design goal, measured in the pilot · “estimated” = our arithmetic or estimate: $0.5M = 0.3 % × 1,700 MW × $269.92/MW-day (PJM 2025/26 capacity auction) × 365; the days split is an estimate · “typical” = published figures: conventional forced outage rate 9.2 % in 2025 (NERC State of Reliability 2026, via Utility Dive); NERC Section 1600 GADS reporting, mandatory for conventional and solar units of 20 MW and up, due 45 days after each quarter (NAES); solar threshold 20 MW from January 2025 (pv magazine USA, 2024). Cause code numbers are from the demo utility's library. Lumina Utilities, its units, suppliers, people, events and records named on this page are fictional — characters and sample data in the working solution.
Software that turns a generating fleet's outages, derates and reserve shutdowns into the event records NERC's Generating Availability Data System asks for — event type, start and end, MW lost and cause code — ready for each quarter's filing. Generation Availability Reporting adds six agents that build the events from historian signals, read the shift log and the work order, propose the code with its reasons and check every record. The analyst confirms each code and the reporting lead signs.
The cause coder decides the event type from the timing in the log — immediate, delayed, postponed or start-up failure — and the cause code from the part that failed in the work order, not the alarm that tripped the unit. It writes a rationale citing both, gives a confidence and keeps two alternatives. Below 80 % confidence it shows the event without a proposal.
The report checker runs 9 checks on every event and 12 rules on the quarter, nightly. It catches extensions for work outside the original scope, codes from the wrong equipment family, times that do not match the breaker, overlapping derates, solar events under 20 MW, outside causes without their notice and hours that do not reconcile. Each failure comes with a proposed fix the analyst applies.
Yes. Solar plants of 20 MW and up report from January 2025, and only events losing 20 MW or more go in the file. The solar threshold is a setting; smaller events stay in the utility's own log and still count in the board figures.
The report writer computes service, planned, maintenance, forced and reserve hours, equivalent availability, forced outage rate and forced MWh lost per unit and for the fleet, capacity-weighted, from the coded records only. Every figure links to its events, and the rates update as codes are confirmed or fixed.
Yes. Utilities outside NERC track the same indexes — equivalent availability factor, forced outage rate, EFORd under IEEE Std 762 — for their regulator, their board and availability-linked power and water purchase agreements. The coding rules, deadlines and thresholds are settings.
Yes. The agents propose and draft; they never confirm a code, change a record or send anything. The analyst confirms every code, the reporting lead signs before the file is built, and any coding-rule change needs a second person and a test on last year's events.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your coding rules, cause code library, unit list, and a sample of historian data, shift logs and work orders — and it goes live once every quality gate has passed. We will walk you through it on one of your own quarters first.
We'll run Generation Availability Reporting on a quarter of your own historian data, shift logs and work orders.