@ArgusPlatformAI

Went from the Navy to a SWAT team to building defense software. Co-founder of ARGUS. I write about wargaming, AI that shows its work, and building from zero.

 Florida
argusplatform.ai
Joined September 2026
  • Tweets755
  • Following619
  • Followers123
35Media
A decision on land rarely stays on land. ARGUS is being built for combat forecasting and wargaming across land, air, sea, cyber, and space. Evidence informs. People decide. argusplatform.ai
1
7
632
We wrote the rule before the first score: after 8 confirmed weeks, ARGUS claims forecasting skill only if it beats the base rate with a 95% interval fully above zero. One week in. Where's your money?
0%It beats it
0%It doesn't
0%Coin flip
0%Show me week 8
0 votes • Final results
1
25
Our model's challenger scored better in testing. We still didn't promote it. The interval on the difference crosses zero, so the gain could be noise. A gate that holds beats a number that looks good.
Dark ARGUS card titled "Better score. Not promoted." Two bars, champion and challenger, nearly equal length. Below, an interval on the difference spans zero. A red stamp reads NOT PROMOTED. Caption: the gain could be noise, so the champion stays.

ALT Dark ARGUS card titled "Better score. Not promoted." Two bars, champion and challenger, nearly equal length. Below, an interval on the difference spans zero. A red stamp reads NOT PROMOTED. Caption: the gain could be noise, so the champion stays.

2
29
Someone dropped a real fruit fly brain map into an AI-generated world, and the fly can't tell it's fake. Wargamers know the feeling. Build a good enough sim and nobody stops to ask which parts are real. Label them anyway.
🤖 Made with AI
Dark night city seen from above, split down the middle. Left half is a real lit city tagged REAL, right half is a teal wireframe city tagged GENERATED, with a small fruit fly hovering on the dividing line. Text: "News, simulation. The fly can't tell. Label the parts anyway."

ALT Dark night city seen from above, split down the middle. Left half is a real lit city tagged REAL, right half is a teal wireframe city tagged GENERATED, with a small fruit fly hovering on the dividing line. Text: "News, simulation. The fly can't tell. Label the parts anyway."

2
43
ARGUS's AI analyst, first 11 answers in production: 10 drafts tried to state a number it couldn't source. 7 fixed on repair. The rest got cut. 0 untagged numbers reached the screen. Showing it live at HAAC today.
Dark scoreboard card titled "No source, no number" with four boxes: 11 answers, 10 drafts with an unsourced number, 7 fixed on repair, 0 untagged numbers shown. Text below: "The rest of those sentences got cut. Every number on screen ties to a tool result. Live at HAAC today." Footer: ARGUS, production, from 19 Oct 2026.

ALT Dark scoreboard card titled "No source, no number" with four boxes: 11 answers, 10 drafts with an unsourced number, 7 fixed on repair, 0 untagged numbers shown. Text below: "The rest of those sentences got cut. Every number on screen ties to a tool result. Live at HAAC today." Footer: ARGUS, production, from 19 Oct 2026.

2
43
Scorecard, week 1 of 8. Ground changed hands in 6 of 176 cells. Our top 10 picks caught 4 and missed 2. Brier 0.023 vs 0.034 for just guessing the base rate. One week proves nothing. Seven to go, and every miss stays posted. github.com/pearl-labs-llc/ar…
Dark two-column scoreboard. Text: "Forecast series, scorecard. Week 1 of 8. Called it: 4. Missed: 2. 1 week counted, not evidence of skill. Top 10 picks. Every miss stays posted."

ALT Dark two-column scoreboard. Text: "Forecast series, scorecard. Week 1 of 8. Called it: 4. Missed: 2. 1 week counted, not evidence of skill. Top 10 picks. Every miss stays posted."

31
Tomorrow we show ARGUS at HAAC. Our one rule for every screen: if it can't say where a number came from, it doesn't go up. Also praying to the conference Wi-Fi gods. They have never once been on our side.
Dark empty stage with a laptop glowing on a lectern; the screen shows three blurred grey placeholder bars, each with a SOURCE tag, and a Wi-Fi icon with one bar. Text: "Tomorrow, HAAC demo. Where did that number come from? If it can't say, it doesn't go up."

ALT Dark empty stage with a laptop glowing on a lectern; the screen shows three blurred grey placeholder bars, each with a SOURCE tag, and a Wi-Fi icon with one bar. Text: "Tomorrow, HAAC demo. Where did that number come from? If it can't say, it doesn't go up."

44
Reports this week put the Geran-5 at a turbojet and up to 600 km/h. A defense tuned for slow propeller drones now gets the same alarm with a lot less clock. If you model air defense, speed just became the input you can't hand-wave.
Dark radar scope with a slow track and a fast track, beside two timing bars, a long grey bar for a propeller drone and a short teal bar for the Geran-5. Text: "News, air defense. Less clock. Geran-5 / turbojet, up to 600 km/h. Same alarm. Shorter warning. Speed is the input you can't hand-wave."

ALT Dark radar scope with a slow track and a fast track, beside two timing bars, a long grey bar for a propeller drone and a short teal bar for the Geran-5. Text: "News, air defense. Less clock. Geran-5 / turbojet, up to 600 km/h. Same alarm. Shorter warning. Speed is the input you can't hand-wave."

34
Sunday we sealed this week's call before anyone knew the answer: 176 cells along the Ukraine front, odds that ground changes hands in each over 7 days. The hash is public now. It gets scored on the Oct 25 map, and misses stay up. github.com/pearl-labs-llc/ar…
Dark terrain map with a glowing blue front line divided into a grid of cells. Text: "Forecast series, this week's call. Sealed before the answer. Sealed, SHA-256 published. 176 cells, 7 days. Scored on the Oct 25 map. Week 2 of 8 / pending."

ALT Dark terrain map with a glowing blue front line divided into a grid of cells. Text: "Forecast series, this week's call. Sealed before the answer. Sealed, SHA-256 published. 176 cells, 7 days. Scored on the Oct 25 map. Week 2 of 8 / pending."

33
On this day in 1805, Nelson took 27 ships of the line into 33 off Cape Trafalgar and won anyway. A model that only counts hulls calls that one wrong. The count was never the hard part. Everything the count leaves out was.
Dark oil-painting style aerial view of two columns of sailing warships cutting toward a long enemy line at sea. Text: "On this day, 21 October 1805. Trafalgar. 27 vs 33 ships of the line. The count was never the hard part."

ALT Dark oil-painting style aerial view of two columns of sailing warships cutting toward a long enemy line at sea. Text: "On this day, 21 October 1805. Trafalgar. 27 vs 33 ships of the line. The count was never the hard part."

31
How a case like the Baltic fakes would go into the cog-war corpus we're building: five intake rules, all in the image. The one that stings: reach gets labeled reach, and effect stays "unknown" until someone shows it. Big numbers hate that rule. What would you add?
ARGUS card titled How a case goes into the corpus. Five intake rules as a checklist: public source, dated; real anchor tagged apart from the fake claim; audience and channel, not just topic; reach labeled reach, effect unknown until shown; human review before it goes in. Labeled design intent, building.

ALT ARGUS card titled How a case goes into the corpus. Five intake rules as a checklist: public source, dated; real anchor tagged apart from the fake claim; audience and channel, not just topic; reach labeled reach, effect unknown until shown; human review before it goes in. Labeled design intent, building.

1
33
Prokhorovka, July 1943. Two respected sources agree exactly: 650 Soviet tanks, 505 German. Case closed? Both trace to the same research lineage. That's one count wearing two hats. An independent count says about 800 vs 500. How many of your sources are secretly one source?
ARGUS card titled Two sources agree. That's one count. Prokhorovka, July 1943. Left panel, same lineage, counted once: 650 Soviet and 505 German tanks, from two sources that trace to one original count. Right panel, independent lineage: about 800 Soviet and about 500 German armored vehicles engaged. Caption: the gap is the honest uncertainty. Sources: HERO/TDI lineage (DBB vol. 6; CDB90) and Clodfelter, 3rd edition.

ALT ARGUS card titled Two sources agree. That's one count. Prokhorovka, July 1943. Left panel, same lineage, counted once: 650 Soviet and 505 German tanks, from two sources that trace to one original count. Right panel, independent lineage: about 800 Soviet and about 500 German armored vehicles engaged. Caption: the gap is the honest uncertainty. Sources: HERO/TDI lineage (DBB vol. 6; CDB90) and Clodfelter, 3rd edition.

1
27
The hard parts of this year's fights keep being the parts games white-card: interceptor reloads, repair crews, tanker hulls, insurance. Planners: which sustainment assumption would you force onto slide one every time? Looking for answers from people who've been burned.
ARGUS card titled Games white-card the parts this year's fights run on. Four white index cards labeled interceptor reloads, repair crews, tanker hulls, and insurance, each stamped assumed fine. The card asks which sustainment assumption goes on slide one, every time. Small line: answers from people who've been burned preferred.

ALT ARGUS card titled Games white-card the parts this year's fights run on. Four white index cards labeled interceptor reloads, repair crews, tanker hulls, and insurance, each stamped assumed fine. The card asks which sustainment assumption goes on slide one, every time. Small line: answers from people who've been burned preferred.

1
28
A Ukrainian sea drone just shot down a Shahed with a machine gun. Sensing, jamming, magazine depth and logistics shaped that fight before a shot was fired. ARGUS resolves them in that order at every time step. Which one do your models skip?
ARGUS Combat Forecasting Engine diagram showing the ordered resolution cycle at each time step: sensing and jamming, logistics and magazine limits, air, autonomous and maritime strikes together, then the ACE land combat model.

ALT ARGUS Combat Forecasting Engine diagram showing the ordered resolution cycle at each time step: sensing and jamming, logistics and magazine limits, air, autonomous and maritime strikes together, then the ACE land combat model.

4
46
"Dark ship" is a headline. Our rule: a missing transponder is a lead, never a verdict. Suspected, then ambiguous, then confirmed, and every step needs an independent sensor that agrees in time and place. Most cases should stay ambiguous. That's honest.
ARGUS card titled A missing transponder is a lead, not a verdict. Three stages left to right: suspected, no signal where one was expected; ambiguous, highlighted, most cases live here, that's honest; confirmed, an independent sensor agrees in time and place. A tag under the middle stage reads stay here until shown otherwise.

ALT ARGUS card titled A missing transponder is a lead, not a verdict. Three stages left to right: suspected, no signal where one was expected; ambiguous, highlighted, most cases live here, that's honest; confirmed, an independent sensor agrees in time and place. A tag under the middle stage reads stay here until shown otherwise.

3
58
Two routes, one map, every assumption on the table. This is ARGUS comparing two courses of action side by side, from a historical demo scenario with uncalibrated numbers. Work in progress. If you plan for a living, what should this screen show you first?
ARGUS wargaming screen in a historical demo scenario: a 3D terrain map with two route options, orange and blue, flanked by panels listing each option's assumptions and evidence. A label reads 'Modelled strengths, uncalibrated adjudication'.

ALT ARGUS wargaming screen in a historical demo scenario: a 3D terrain map with two route options, orange and blue, flanked by panels listing each option's assumptions and evidence. A label reads 'Modelled strengths, uncalibrated adjudication'.

3
35
Next two weeks: an Army xTech|Search 10 white paper due Oct 19 and an ARGUS platform demo Oct 22. Coffee budget is under review. The rule doesn't change with the calendar: every slide says what's built, what isn't, and says who.
ARGUS card titled Busy calendar. Same rules. Three date tiles: Oct 19, Army xTech Search 10 white paper; Oct 22, ARGUS platform demo; Monthly, Tradewinds pitch video. Line: every slide says what's built, what's not, says who.

ALT ARGUS card titled Busy calendar. Same rules. Three date tiles: Oct 19, Army xTech Search 10 white paper; Oct 22, ARGUS platform demo; Monthly, Tradewinds pitch video. Line: every slide says what's built, what's not, says who.

2
43
Cognitive warfare isn't a bolt-on domain. Real drones have strayed into Baltic airspace. Then a fake site posing as a defense outlet claimed one damaged NATO jets. Small event, bigger story. If your game adjudicates the drone and skips the story, it scored half the turn.
ARGUS card titled Score the drone, skip the story: half a turn. An adjudication sheet with two lines: real drone strays into airspace, checked, small event, adjudicated; fake outlet claims it damaged jets, unchecked, bigger story, skipped. A rubber stamp reads half a turn.

ALT ARGUS card titled Score the drone, skip the story: half a turn. An adjudication sheet with two lines: real drone strays into airspace, checked, small event, adjudicated; fake outlet claims it damaged jets, unchecked, bigger story, skipped. A rubber stamp reads half a turn.

2
27
Battles in our database with both sides' strength recorded: 1900 to 1945, 476. 1946 to 1990, 172. 1991 to 2020, 26. Since 2021, zero. The wars everyone is asking about are exactly where the public record goes dark. A model should say where its history stops.
ARGUS bar chart titled The historical record has a cliff. Battles with both sides' strength on record by era: 1900 to 1945, 476; 1946 to 1990, 172; 1991 to 2020, 26; 2021 and later, 0.

ALT ARGUS bar chart titled The historical record has a cliff. Battles with both sides' strength on record by era: 1900 to 1945, 476; 1946 to 1990, 172; 1991 to 2020, 26; 2021 and later, 0.

1
1
19
Some pressure campaigns aren't trying to scare you. They're trying to bore you. Run the same patrol often enough and the next one isn't news. Hard to game, because nothing "happens" on any single turn. How would you score normalization as an effect?
ARGUS illustrative chart: as the same patrol repeats, news attention falls toward zzz while what feels normal rises. How would you score that?

ALT ARGUS illustrative chart: as the same patrol repeats, news attention falls toward zzz while what feels normal rises. How would you score that?

2
33
Interceptor news lately: big contracts to build more. A contract isn't a round, though. CBO says rebuilding the stock takes at least five years. So why does almost every wargame start with the magazine full at turn one? That's the assumption I'd attack first.
1
32
Load more