College X · Aviation Node · Meteorology · Year 3 · senior project, a year early
Channel 7 or Channel 42? Nobody kept the receipts.
Musa Eldamaty grew up in a two-weatherman house. His mother trusted the Channel 7 forecast. His father swore by Channel 42. Every storm that missed started the same argument at dinner, and nobody could settle it, because nobody had written down what either of them said. So he is building the thing that writes it down.
ME
◆ The Question He Carries
His parents came to America from Egypt in the 1970s and learned the local weather the way most people do: from the television. They picked different channels: his mother’s Channel 7 on the old top dial, his father’s Channel 42 on the UHF knob underneath, the newer station. By the time Musa was old enough to follow the arguments, the forecast was a family loyalty. His mother remembered every storm Channel 42 missed, and his father remembered every one Channel 7 missed.
“Everybody in my house had an opinion about who was right. Nobody had a single number.”
When he went looking for the number, the question got bigger than two TV stations. What he wanted was a running report card for the forecast his own county gets, in plain words: how often it lands within a few degrees, whether it leans warm, and whether “60% chance of rain” rains 60% of the time. He knows the report cards that already exist, and he reads them. They grade stations and locations; his parents’ question was about their county, in plain words. So he started building one.
He set one rule on day one: the receipts never carry a person’s name. The forecasters from his dinner table stay unnamed, and so does the city. He grades the federal forecast guidance instead. “Somebody on air is doing a hard thing in public. I’m grading the forecast, not the face.”
Profile
CollegeX · Engineering
BuildingAviation Node · Birmingham-Shuttlesworth
DepartmentMeteorology & Atmospheric Physics
AdvisorKenny Spinks
Second readerCassandra Lorenz
Year3 of ~4 · Junior
Homea two-weatherman house
Student IDOPA-2024-0247
How He Got Here
OPA enrolled him in the fall of 2024, which is the 2024 in his number. He landed in meteorology at the Aviation Node because Kenny Spinks teaches weather as a language you learn to read instead of look up. Musa wanted to know how often the translation is right.
In his junior year he asked to start his senior project early. Spinks said yes on one condition: grade it before you build it. His second reader comes from DOSA, because under the numbers the question is outcome versus trust.
Tab 1 of 4Overview
NETOPA · Source Log · What the Receipt Is Made Of
Eight sources. One click from each.
He has no papers yet. He has data. Every number on his report card comes from one of these, and each one links to where it lives. Federal forecast guidance, airport observations and volunteer rain gauges, from a university archive open to anyone, and anyone can check them.
✎ Annotated Sources
KEYSTONE · the shelf that makes it possible
archive
Yesterday’s forecast, still on the shelf
Iowa Environmental Mesonet, Iowa State University. Model Output Statistics archive and API. · mesonet.agron.iastate.edu/mos · API docs
The whole project rests on this. You can’t grade a forecast after the fact unless someone kept it, and the Mesonet keeps past model forecasts station by station. A forecast made on August 30 can still be checked a month later. That’s the difference between an argument and a receipt.
his current pull: 50 airport stations · Aug 31 – Sep 29, 2026 · each forecast from the 12Z run the day before
SOURCE 01 · what the air actually did
observed
The airport thermometer
Iowa Environmental Mesonet, ASOS/AWOS airport observation archive. · mesonet.agron.iastate.edu/request/download.phtml
The other half of every temperature receipt: the high and the low, measured at the airport weather station the county is matched to. Each forecast is graded over its own hours (Sources 03 and 04), and every build prints how each one is counted, so anyone can check the definitions against the numbers.
a window needs three-quarters of its hours observed, or it isn’t graded
SOURCE 02 · the county’s own rain gauges
observed
The backyard gauges
CoCoRaHS, Community Collaborative Rain, Hail & Snow Network. · cocorahs.org · reports read from the Iowa Environmental Mesonet archive · USDA Climate Hubs CoCoRaHS fact sheet · climatehubs.usda.gov
Rain comes from the county itself, not the airport. CoCoRaHS calls itself “a grassroots volunteer network of backyard weather observers of all ages and backgrounds,” and its readings are the ones his parents’ neighbors could be filing. The fact sheet asks volunteers to read the gauge “at the same time each day, ideally at 7 am,” and admits that “this might not always be feasible.” Volunteers can also “file multi-day reports to cover the time on vacation,” so a reading may cover more than one day. Usually, a reading dated the next morning stands for the 24 hours before it.
rain day: the county’s median reading 0.01 in or more · no reports = not graded
SOURCE 03 · the forecast being graded
primary
The National Blend of Models
NOAA Meteorological Development Laboratory, National Blend of Models. · vlab.noaa.gov/web/mdl/nbm · station card: vlab.noaa.gov/web/mdl/nbm-textcard-v5.0
His primary forecast, the one the report card leads with. NOAA says “The goal of the NBM is to create a highly accurate, skillful and consistent starting point for the gridded forecast.” Its high and low are 18-hour values. The station-bulletin card says “Min is between 00Z-18Z” and “Max is between 12z(current day)-06Z(next day),” so the airport high and low he grades it against cover those same hours. For rain he takes its two 12-hour chances, one for the day and one for the night.
called rain: either 12-hour chance 50% or more · called dry: the day and night chances add up to 20% or less · anything in between isn’t scored
SOURCE 04 · the second yardstick
comparison
GFS MOS guidance
NOAA Meteorological Development Laboratory, Model Output Statistics. · vlab.noaa.gov/web/mdl/mos · weather elements: vlab.noaa.gov/web/mdl/weather-elements
Graded side by side with the Blend, on the same days, so the Blend’s score has something to be compared against. MDL lists its temperatures as “Daytime (7 a.m. - 7 p.m. LST) maximum; Nighttime (7 p.m. - 8 a.m. LST) minimum,” so GFS MOS gets graded on those hours instead. A grade with nothing next to it is just a number.
SOURCE 05 · the one the API won’t keep for him
not yet graded
The weather service’s point forecast
National Weather Service API, api.weather.gov. · weather.gov/documentation/services-web-api
The weather service’s API serves a point forecast in 12-hour periods over the next seven days, and its forecast endpoints drop periods that have passed, so grading it later from the API means saving a copy every morning. He saved his first copy on September 29, 2026, for 51 stations. The daily recorder isn’t running yet, so this forecast isn’t on the report card.
SOURCE 06 · the map pin
matching
Which station speaks for which county
U.S. Census Bureau, Gazetteer Files (county reference points). · census.gov › Gazetteer Files
Each county gets the nearest airport station that has forecast records and actually returned observations, within 40 km (about 25 miles) of the county’s Census reference point. Stations whose archive has ended are skipped. If nothing is that close, the county stays ungraded and the card says so. An empty cell is honest. A borrowed number from past the 25-mile line is not.
SOURCE 07 · who already keeps score
prior work
He isn’t the first to grade the forecast
ForecastAdvisor, “How It Works.” · forecastadvisor.com/docs/about · NOAA NCEI, National Digital Forecast Database (NDFD). · ncei.noaa.gov › NDFD · NOAA MDL, NDFD Verification FAQ. · vlab.noaa.gov/web/mdl/ndfd-verification-faq
Private services do it: ForecastAdvisor, powered by ForecastWatch, says “we’ve compared billions of forecasts from major consumer and government weather services against actual observations for over 2,200 locations worldwide since 2004.” And the weather service grades its own forecasts. NOAA runs NDFD verification at a published list of stations, and NCEI archives the weather service’s forecast grids, which it describes as the products the weather service uses to make its “text, graphic, and digital products for emergency managers and the public.” What Musa adds is smaller and plainer: one county at a time, in plain words, with the method printed and two models side by side.
NCEI period of record for NDFD: back to 6 Jun 2004 (about ten years online)
Tab 2 of 4Source Log
NETOPA · Advisor Log · Junior Fall
Grade it before you build it.
Two readers who come at it from opposite ends. Spinks feels the sky before the instruments read it and wants to know what the forecast missed. Lorenz watched her own trusted model break in eighteen minutes and wants to know what the forecast promised.
▤ Recent Meetings
The rain question. “Don’t ask the airport whether it rained. Ask the county.” The rain grade uses the county’s CoCoRaHS volunteer gauges, and each forecast is graded on its own hours. The receipt for Aug 31 to Sep 29 grades 91 of 95 counties for temperature and 81 for rain.
Action items
→ Keep the rain grade to what the two 12-hour chances can guarantee
→ Name the counties with no volunteer reports. Don’t fill them in
→ Outside readers, then public
→ Keep the rain grade to what the two 12-hour chances can guarantee
→ Name the counties with no volunteer reports. Don’t fill them in
→ Outside readers, then public
Brought him the first full receipt. Spinks skipped the averages and went straight to September 21–23, when the median high across his stations fell twenty degrees in two days and the Blend’s highs ran warmer each day. “That’s the one I’d have felt coming. Your averages say the forecast is good. That week says when it isn’t. Put that week on the page, not just the average.”
Action items
→ Rebuild the receipt once a week until the daily recorder runs
→ List the ungraded counties by name. Don’t fill them in
→ A light outside read before it goes public
→ Rebuild the receipt once a week until the daily recorder runs
→ List the ungraded counties by name. Don’t fill them in
→ A light outside read before it goes public
She read the plan as outcome versus trust. “Every forecast is a promise with a number on it. People plan a wedding around a 20%. You’re not grading the weather. You’re keeping the receipt on a promise.” Then she tested the no-names rule, and it held. “Good. When a model breaks, somebody needs a face to blame. Don’t hand them one. Grade the promise.”
Standing question
→ When the forecast says a number, how often does the world agree with it? And does it say so before the miss or after?
→ When the forecast says a number, how often does the world agree with it? And does it say so before the miss or after?
First meeting of junior year. Musa asked to start his senior project a year early. Spinks agreed, with one condition: “Decide what you’re grading and how you’ll score it before you write a line of the page. If you pick the scoring after you see the numbers, you’ll pick the scoring that flatters them.” Every build prints the definitions in its method block, so anyone can check them.
Cross-reference
→ METEO at the Aviation Node: Spinks’ weather-as-literacy course
→ Second reader from DOSA: Lorenz, CYBER 451 Week 1, the gap between outcome and trust
→ METEO at the Aviation Node: Spinks’ weather-as-literacy course
→ Second reader from DOSA: Lorenz, CYBER 451 Week 1, the gap between outcome and trust
Tab 3 of 4Advisor Log
NETOPA · Research Map · Tap each stage to open it
From the dinner table to the receipt.
Where the project stands, honestly marked. Green is settled. Aqua is where he’s working right now. Red is the wall. The wall is the part he hasn’t built yet.
◇ The Shape
1
Who was right?Settled · nobody can say from memory
›
Where he starts
Both of his parents remember the other channel’s misses. Memory keeps the misses and forgets the hits. It can’t settle the argument, but a written record can.
Both of his parents remember the other channel’s misses. Memory keeps the misses and forgets the hits. It can’t settle the argument, but a written record can.
2
The shelfSettled · the model forecasts are archived
›
Data in hand
The federal model forecasts are archived, station by station, next to what each airport measured. The receipt can be written after the fact.
The federal model forecasts are archived, station by station, next to what each airport measured. The receipt can be written after the fact.
3
The receiptActive · 30 days, 91 of 95 counties
›
Built Oct 1 · to be rebuilt weekly
Aug 31 – Sep 29, 2026, 50 reporting stations. The Blend’s highs landed within 3°F 87% of the time (1,419 graded station-days), with an average miss of 1.9°F and almost no lean. Its lows landed within 3°F 81% of the time and ran a little warm. GFS MOS runs at 37 of the 50 stations. On the 1,032 station-days where both models’ highs could be graded, the Blend’s landed within 3°F 86% of the time and GFS MOS’s 83%; on lows (1,034 station-days) GFS MOS did better, 83% to 82%. Four counties stay ungraded: Gibson, Macon, Sequatchie and Smith have no station with forecast records within 40 km. The temperatures are good. The rain is where the argument lives.
Aug 31 – Sep 29, 2026, 50 reporting stations. The Blend’s highs landed within 3°F 87% of the time (1,419 graded station-days), with an average miss of 1.9°F and almost no lean. Its lows landed within 3°F 81% of the time and ran a little warm. GFS MOS runs at 37 of the 50 stations. On the 1,032 station-days where both models’ highs could be graded, the Blend’s landed within 3°F 86% of the time and GFS MOS’s 83%; on lows (1,034 station-days) GFS MOS did better, 83% to 82%. Four counties stay ungraded: Gibson, Macon, Sequatchie and Smith have no station with forecast records within 40 km. The temperatures are good. The rain is where the argument lives.
4
The weather service’s forecastThe wall · his route needs a recorder every morning
›
Not running yet
The weather service’s API serves the next seven days, not the past ones. NCEI keeps the weather service’s forecast as national grids; the simpler route for his page is to save the API’s point forecast every single morning, and one missed morning is a hole in the record. This is the part that turns a model report card into the thing his parents argued about.
The weather service’s API serves the next seven days, not the past ones. NCEI keeps the weather service’s forecast as national grids; the simpler route for his page is to save the API’s point forecast every single morning, and one missed morning is a hole in the record. This is the part that turns a model report card into the thing his parents argued about.
5
Forecast Receipts, in publicNorth star · senior year, 2027–28
›
Live · not on its own yet
A public page where anyone can click their county and see the receipt. It’s live at receipts.opathorlokanuniversity.net, to be rebuilt by hand each week for now. The senior project is the page running on its own, graded in public, with no names on the scoreboard.
A public page where anyone can click their county and see the receipt. It’s live at receipts.opathorlokanuniversity.net, to be rebuilt by hand each week for now. The senior project is the page running on its own, graded in public, with no names on the scoreboard.
165 of 248
county-days when the Blend gave a 50% or better chance of rain for the day or the night, the county’s volunteer gauges caught rain: 67%. When its day and night chances added up to 20% or less, so the 24-hour chance couldn’t be higher, the county stayed dry on 1,509 of 1,556 county-days: 97%. When it called rain, the gauges caught it two days in three. It’s one month: the rain calls came on 12 dates, and 189 of the 248 came from four of them (Sep 11, 12, 22 and 23). 81 of the 91 counties with a station had gauge reports; 10 had none (Campbell, Crockett, Humphreys, Johnson, Lake, Perry, Rhea, Scott, Trousdale and Unicoi). Aug 31 – Sep 29, 2026; 50 stations serving Tennessee counties, 3 across the state line.
▤ Does a 30% chance rain 30% of the time?
The question from the dinner table, asked properly. The Blend gives a day chance and a night chance, not one 24-hour number. Together they pin the 24-hour chance between two values: at least the bigger of the two, at most the two added up. If the forecast keeps its promise, the share of days that actually rained should land between them.
| Bigger 12-h chance | County-days | 24-h chance between | It rained | |
|---|---|---|---|---|
| 0–9% | 1,387 | 3–4% | 2% | 1 pt below |
| 10–19% | 256 | 14–20% | 13% | 1 pt below |
| 20–29% | 185 | 24–35% | 30% | inside |
| 30–39% | 138 | 34–51% | 41% | inside |
| 40–49% | 105 | 45–69% | 58% | inside |
| 50–69% | 141 | 58–85% | 55% | 3 pts below |
| 70–100% | 107 | 81–99% | 82% | inside |
Four of seven groups land inside; the other three run a little drier than even the lowest chance allowed. Aug 31 – Sep 29, 2026, every graded county-day, rain from each county’s CoCoRaHS gauges. One month only, and neighboring counties share weather, so the county-days aren’t independent.
▤ Junior-Year Class List
| Course | Home | Section | Status |
|---|---|---|---|
| METEO — Meteorology & Atmospheric Physics (Kenny Spinks)Weather as literacy, not lookup. It’s where he learned what a forecast is before he started grading one. | Aviation Node · College X | — | real / canonical |
| CYBER 451 · Week 1 — The Gap (Cassandra Lorenz)Outcome versus trust. The week that gave him the frame for everything under the numbers. | DOSA · College I · Bldg 1 | 4.1.5 | real / canonical |
| Forecast Verification & Statistics [code TBD]Hit rates, typical miss, lean, and whether a 30% chance rains 30% of the time. | Aviation Node · College X | — | proposed |
| Research Integrity & Sourcing [code TBD]The Three Gauge Test: one source is a guess, two a hypothesis, three is engineering. He treats the forecast, the observation, and the second model as his three gauges. | HASS (Human-AI Systems Stewardship) | 4.9.x | proposed |
⚠ This class structure is aspirational and not yet formally set up. Course codes marked [code TBD] are proposed and still need to be reconciled against the master OPA catalog. METEO and CYBER 451 are the courses his two readers already teach on their own portals.
Where This Touches the Universe
Kenny Spinks · Meteorology, Aviation Node →
Advisor. He feels a pressure change before the instruments read it. He keeps Musa aimed at the misses, not the averages.
Cassandra “Spin Rate” Lorenz · DOSA →
Second reader. A trusted model did exactly what it was built to do and she wore the blame. Grade the promise, not the face.
The Wing · the weather side of the flight →
Spinks’ lab at the Aviation Node. A pilot needs to know how far to trust the forecast they’re flying into.
Forecast Receipts · the tool itself →
Click a Tennessee county and see how often the federal forecast guidance was right.
His Student Passport · live tracker →
The dashboard Musa sees when he logs in: milestone stamps and the daily check-in on where the project stands. This card is the portrait; the passport is the working desk.
Tab 4 of 4Research Map