Every projection names what would sharpen it
Every probability ships with the specific thing that would tighten it — a source to qualify, a record to resolve, a question to answer, so the next hour of work is chosen rather than guessed.
Decision science for the defense industrial base
Nørn is a decision product for the defense supply chain. Not a map of your exposure — a ranked list of the calls in front of you, each with a probability, the date the world will settle it, and what it would take to sharpen it. When that date passes, the call is scored against what actually happened.
A risk score that is never checked against an outcome is an opinion with a number on it.
Access is invitation-only today. Tell us what you’re trying to protect and we’ll come back inside two business days.
A risk register hands you a list of suppliers. Nørn hands you two different numbers for each one, because collapsing them is how this whole category misleads: the dollars whose date will move, and the dollars expected to come back off the award. Nationally those are 67% and 3.8% of the same book.
The decision
Qualify a second source before 14 November, or accept a single-source dependency through the window. Either way the finding becomes a situation that has to declare, before it is created, what will settle it and roughly when.
“This supplier’s next award holds its date” settles itself, from the federal record. “This supplier is a concentration risk” is a true statement and a useless one — no evidence can ever arrive that closes it. Nørn refuses to create the second kind. A queue of findings that can never be settled is how every risk register turns into wallpaper.
These are the difference between a decision product and a dashboard with a confidence column. Each one is a constraint the system will not let an analyst route around.
Every probability ships with the specific thing that would tighten it — a source to qualify, a record to resolve, a question to answer, so the next hour of work is chosen rather than guessed.
A supplier’s estimate is a beta-binomial posterior — its own record blended toward the population rate, weighted by how much record exists. A hard cut-off was tried and abandoned: it put a cliff at ten observations and collapsed almost every supplier onto the population rate. The blend is continuous; the basis is always stated.
Confidence is encoded in fill, not color — solid for a supplier’s own measured rate, hollow for a borrowed population rate. A rate earned from 118 awards must never render like one borrowed from a band.
Not “elevated risk” but a probability attached to a date the world will decide. When that date passes the call is scored against what happened, and the score is kept.
If this supplier stops, who else is approved to make this part? Counted per part, from the federal catalog — two firms approved against the same stock number are substitutes as a matter of record.
FLIS / PUB LOG · observed
The probability this award misses its date, banded by size and measured against resolved history — with the date it will be settled on.
FPDS · measured
Expected annual loss for the county a supplier sits in, and where a dozen suppliers that look diversified share one physical exposure.
FEMA National Risk Index · 3,232 counties
Every supplier screened against twelve government lists, with the entity resolution to keep a name collision from becoming an accusation.
Consolidated Screening List · 1260H · UFLPA
Suppliers running software with known-exploited vulnerabilities — scoped to what is actually measurable from the outside, and no further.
CISA KEV
Where the open dollars pool behind one buyer, one product family, or one place — the exposure no per-supplier score can express.
FPDS · derived
A supplier score tells you where you are. It does not tell you which move to make, which fix does not help, or what stops being possible in March. Below is a live board — inject a shock and watch the moves re-rank, or close a gap and watch the interval tighten without the number moving.
Candidate moves · scored
| Move | Wins | If wrong |
|---|---|---|
| Surge Columbia; accept Virginia rate slipleads | 44% | AUKUS certification becomes unmakeable. The alliance commitment is the thing that breaks. |
| Buy the constraint out — capitalize castings, forgings, a second weld source | 41% | Money lands three tiers down and takes 30+ months to show up as a hull. Slow where the door is fast. |
| Hold Virginia at rate; accept Columbia risk | 27% | A deterrent patrol gap. There is no recovery move on the far side of this one. |
| Re-phase the AUKUS transfer window | 22% | Relieves the rate requirement, spends allied credibility, and does nothing for Columbia. |
Inject a shock — the board re-ranks
Nothing injected. The board above is the baseline read.
One-way doors
Gaps · close one and watch the band tighten
3 open questions hold the interval 56 points wide. Closing a gap narrows it and never moves the number — new information makes you more certain of what is true, not more optimistic about it.
Reading that a plan is fragile is an intellectual event. Watching the odds swing when you break it yourself is a different one, and it is the one that changes minds. A plan that survives only under the assumptions you started with is not a plan — it is a forecast you are fond of.
The floor, stated up front
Nørn does show tier-2 — the part of it that was actually filed. Primes report subcontracts over $30,000 under FFATA, and Nørn reads those filings directly. Nothing here infers who supplies whom.
Across twelve measured monthly windows of national DoD, between 0.28% and 0.72% of prime awards carried a subcontract filing — median 0.41%. Nørn reports the range rather than a single figure, because a rate that moves by more than a multiple cannot be described by its middle, and a single number is what gets quoted.
Coverage looks like it is falling in recent months. That is filing lag, not decline. Primes report after the fact, so the newest windows are still filling. Reading that curve as “subcontract reporting is collapsing” would be a confident, wrong finding of exactly the kind this product exists not to produce.
So the sub-tier is a drill-down, not a network view. “Here is what this prime filed” is defensible at this density. A graph is not: an absent link means nobody filed, which on a network is indistinguishable from a prime that has no subcontractors at all — and a tool that drew it anyway would invent chokepoints out of missing paperwork.
What it will not do is infer the rest. No complete sub-tier map exists at any price. There is no tier-3. And federal award records carry no stock numbers, so we will never tell you what share of a supplier’s dollars sits behind a single-source part — that figure would be an estimate wearing a measurement’s clothes.
Slip probability is estimated from resolved awards, banded by size, because a single pooled rate describes nobody. Every band shows the population it was measured against — including the one that is thin.
| Award size | Slips | Median slip, when it does | Resolved awards measured |
|---|---|---|---|
| Under $250k | 1.6% | 133 days | 7,888,754 |
| $250k – $2.5M | 31.7% | 273 days | 89,952 |
| Over $2.5M | 61.9% | 365 days | 23,375 |
A large award is roughly forty times likelier to move than a small one, and when it does the median slip is a full year.
Most of that forty is attention, not delivery, and the number should not stand alone. A slip is a recorded schedule change — somebody has to file the modification. After the promised date passes, 97.2% of small awards see no action of any kind, against 46.6% of large ones. A $50,000 award delivered six months late and one delivered exactly on time leave the same record, which is none. Among awards somebody actually touched, the spread is 2.5×, not forty.
So the figure is the right answer to will the promised date move, which is what the system predicts and scores. It is not a delivery-failure rate, and it is not the amount at risk.
Those are two different claims, and either alone misleads.Across the whole open federal book, $489.3B is expected to move — 67% of it. Of that, $27.8B is expected to come back off its award: 3.8%. The first reads as a catastrophe; the second reads as though nothing is late. Together they say the true thing, which is that most slipped work still lands and the money worth watching is a twentieth of what “at risk” implies.
The loss signal is not where you would look for it. A large award that slips is no likelier to have money taken back (5.62%) than one that does not (5.97%) — slipping is how big programs run. A small award that slips is 16.6× likelier to be reversed. And de-obligated dollars are frequently re-obligated elsewhere, so this is a floor on disruption, never a measurement of waste.
A number also means nothing without knowing what normal is. A third of open federal dollars land on their promised date. Across the fourteen buying components holding over $1B, the middle half sits between 32% and 37%. A program at 33% is ordinary, not a crisis — and Nørn says which, rather than leaving a reader to do the subtraction backwards.
All of it counted, not estimated, from the public federal record — FPDS, FLIS/PUB LOG, the Consolidated Screening List, CISA KEV and the FEMA National Risk Index. Free, citable, and auditable by anyone holding the same sources.
Nørn is three general capabilities. Supply chain is where they are deepest today, and the hardest place to fake them — the federal record is public, so every claim above can be checked against a source you also hold. That is the point of leading with it.
Raw frontier-model forecasts are confidently wrong — on the hardest domains, worse than guessing the base rate. Nørn calibrates them against a corpus of resolved outcomes, then keeps scoring itself as new ones resolve.
The same organization across three spellings, two aliases and a shell resolves to one entity. Ambiguity is the failure mode that quietly corrupts every downstream number.
Build a twin of the situation, then change it. Remove an actor, sever a link, move a policy — and watch the calibrated picture move under the modified network.
The corpus of resolved outcomes that scores a supplier’s slip probability is the same corpus that scores Nørn’s geopolitical and scientific forecasts. It is one engine with one track record, not a supply-chain product with adjacent ambitions.
Pointed at, today and under the same calibration — click any role to see how it fits the work.
Blair Merlino is Nørn's founder. Former Army Special Operations, BU Questrom MBA + MS in Information Systems, management consulting at Deloitte, and infrastructure-focused VC and operator roles since.
The hardest part of high-stakes decision-making isn't access to information — it's the gap between information and a calibrated, actionable verdict. Nørn closes that gap because the founder spent a career living on the wrong side of it.
Access is invitation-only today. Tell us who you are and what you're trying to track; we'll get back to you within two business days.
Security & data handling documentation — tenant isolation, collective layers, controls and compliance trajectory — is available on request.