You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This repository demonstrates using the BinaryPredictor agent by Chronulus AI to estimate win probabilities for match-ups in the 2025 NCAA Men's Basketball Tournament.
BinaryPredictor Config
Chronulus Session Setup + First Prediction
For the first match up, the following prompt was used in Claude to setup the Chronulus Session and get the first prediction. Replace the items in {{brackets}} with your own values.
The 2025 NCAA Men's basketball tournament starts tomorrow. I would like to make predictions for the first round of the tournament.
Please create a new Chronulus Session for this and setup a data model that you can plug in data for each match up separately. I want to reuse the same BinaryPredictor for each match up. In the task, be sure to specific that we want to predict the probability that team 1 wins the matchup.
As input data, I will give you images and injury reports (if one is available) for each match up in folder in my workspace. I have also include a PDF of the bracket and a text file with the current schedule in my workspace.
Pass these images and documents to Chronulus. Additionally, please include one field for "Additional context" and one for a list of additional images. In some matches, I will want to provide more details or additional images that are not available for other matches. These fields will be used for those.
In the fields that you create that contain information about a specific team, prefix the fields with 'team_1' and 'team_2' according how the team is listed in the match up I provide.
When you have the predictions, please save the results as html in `picks'.
Let's start..
Get the Chronulus predictions for this match up in {{'round1-south'}}:
{{(8) Louisville vs. (9) Creighton, 12:15 p.m. | CBS}}
From this conversation, we asked Claude to save a JSON copy of the sessions information for future reference. This file is located at mens-bracket/sessions/session.json and contains:
session_id - a unique id that can be reused across Claude conversations orient the BinaryPredictor agent about the situation and task.
situation - the situation defined in the Chronulus session
task - the task defined in the Chronulus session
input_data_model - a example of the input data model that Claude successfully used when passing inputs to the agent on during the first conversation.
For all predictions after the first match-up
In subsequent conversations, the following prompt was used in Claude to reuse the session from the first match up. Again, replace the values in {{brackets}} with your own.
The session id referenced in this prompt is taken directly from the session info that was saved to mens-bracket/sessions/session.json. You would want to replace this with the session id from your own conversation.
The 2025 NCAA Men's basketball tournament starts tomorrow. I would like to make predictions for the first round of the tournament.
Please reuse the Chronulus session with session id = {{'1b330f4b-f0ac-5c06-975d-9f41c5abf58d'}} for this and setup a data model that you can plug in data for each match up separately. I want to reuse the same BinaryPredictor for each match up.
As input data, I will give you images and injury reports (if one is available) for each match up in folder in my workspace. I have also include a PDF of the bracket and a text file with the current schedule in my workspace.
Pass these images and documents to Chronulus. Additionally, please include one field for "Additional context" and one for a list of additional images. In some matches, I will want to provide more details or additional images that are not available for other matches. These fields will be used for those.
In the fields that you create that contain information about a specific team, prefix the fields with 'team_1' and 'team_2' according how the team is listed in the match up I provide. This needed because the session is setup to predict the probability that team 1 will win. So need to make sure this is correctly labeled.
When you have the predictions, please save the results as html in `picks'.
Let's start..
Please ask Chronulus to predict the probability that the first team listed in the following matchup will win. Use 5 experts.
Get the Chronulus predictions for this match up in {{'round1-midwest'}}:
{{(4) Purdue vs. (13) High Point, 12:40 p.m. | truTV}}
Tournament Schedule with Win Probability Predictions
This is the schedule of games for the 2025 NCAA Men's Basketball Tournament along with win probability predictions for each team provided by Chronulus AI.
First Four
Date
Region
Team 1 (Seed)
Team 2 (Seed)
Team 1 Win %
Team 2 Win %
Mar 18
South
Alabama State (16)
Saint Francis (16)
50.65%
49.35%
Mar 18
South
San Diego State (11)
North Carolina (11)
45.43%
54.57%
Mar 19
East
American University (16)
Mount St. Mary's (16)
51.41%
48.59%
Mar 19
Midwest
Texas (11)
Xavier (11)
49.87%
50.13%
First Round
Date
Region
Team 1 (Seed)
Team 2 (Seed)
Team 1 Win %
Team 2 Win %
Mar 20
South
Louisville (8)
Creighton (9)
46.32%
53.67%
Mar 20
Midwest
Purdue (4)
High Point (13)
85.76%
14.23%
Mar 20
East
Wisconsin (3)
Montana (14)
81.15%
18.85%
Mar 20
Midwest
Houston (1)
SIU Edwardsville (16)
97.20%
2.80%
Mar 20
South
Auburn (1)
Alabama State (16)
97.80%
2.20%
Mar 20
Midwest
Clemson (5)
McNeese (12)
69.89%
30.11%
Mar 20
East
BYU (6)
VCU (11)
49.95%
50.05% [upset watch - refresh is more experts and fresh news]
Mar 20
Midwest
Gonzaga (8)
Georgia (9)
49.01%
50.99%
Mar 20
Midwest
Tennessee (2)
Wofford (15)
91.79%
8.21%
Mar 20
West
Kansas (7)
Arkansas (10)
51.16%
48.84% [upset watch - refresh with injury news / tips]
Mar 20
South
Texas A&M (4)
Yale (13)
52.81%
47.19% [upset watch - refresh with injury news / tips]
Mar 20
West
Missouri (6)
Drake (11)
46.53%
53.47% [upset alert]
Mar 20
Midwest
UCLA (7)
Utah State (10)
44.61%
55.39% [upset alert]
Mar 20
West
St. John's (2)
Omaha (15)
88.90%
11.10%
Mar 20
South
Michigan (5)
UC San Diego (12)
57.86%
42.14%
Mar 20
West
Texas Tech (3)
UNC Wilmington (14)
82.15%
17.85%
Mar 21
East
Mississippi State (8)
Baylor (9)
50.78%
49.22%
Mar 21
East
Alabama (2)
Robert Morris (15)
89.71%
10.29%
Mar 21
South
Iowa State (3)
Lipscomb (14)
81.42%
18.58%
Mar 21
West
Memphis (5)
Colorado State (12)
55.11%
44.89% [upset watch - refresh with news on injuries]
Mar 21
East
Duke (1)
Mount Saint Mary's (16)
97.95%
2.05%
Mar 21
East
Saint Mary's (7)
Vanderbilt (10)
61.78%
38.22%
Mar 21
South
Ole Miss (6)
North Carolina (11)
47.13%
52.87% [upsert alert - if you can call UNC an underdog]
Mar 21
West
Maryland (4)
Grand Canyon (13)
71.93%
28.07%
Mar 21
West
Florida (1)
Norfolk State (16)
97.26%
2.74%
Mar 21
Midwest
Kentucky (3)
Troy (14)
83.20%
16.80%
Mar 21
South
Marquette (7)
New Mexico (10)
56.45%
43.55
Mar 21
East
Arizona (4)
Akron (13)
76.60%
23.40%
Mar 21
West
UConn (8)
Oklahoma (9)
50.84%
49.16% [Update with news on Godwin's status]
Mar 21
Midwest
Illinois (6)
Xavier (11)
47.67%
52.33%
Mar 21
South
Michigan State (2)
Bryant (15)
88.68%
11.32%
Mar 21
East
Oregon (5)
Liberty (12)
65.53%
34.47%
Second Round (before 12pm EST on March 20)
These probabilities were estimated before 12pm EST on March 20th using data for games through March 19th.
The games listed here explore the bracket hypothetical case that each of the picks from the first round will be / were correct.
Date
Region
Team 1 (Seed)
Team 2 (Seed)
Team 1 Win %
Team 2 Win %
South
Auburn (1)
Creighton (9)
64.42%
35.58%
Midwest
Clemson (5)
Purdue (4)
38.37%
61.63%
East
VCU (11)
Wisconsin (3)
28.25%
71.75%
Midwest
Houston (1)
Georgia (9)
80.78%
19.22%
Midwest
Utah st (10)
Tennessee (2)
23.22%
76.78%
West
Kansas (7)
St. John's (2)
29.08%
70.92%
South
Michigan (5)
Texas A&M (4)
42.19%
57.81%
West
Drake (11)
Texas Tech (3)
30.31%
69.69%
South
North Carolina (11)
Iowa St (3)
38.35%
61.65%
South
Oregon (5)
Arizona (4)
44.39%
55.61%
South
Memphis (5)
Maryland (4)
47.62%
52.38%
Second Round (after 12pm EST on March 20)
These probabilities were estimated after 12pm EST on March 20th using data for games through March 19th. However, none of the match-ups estimated in this section depend on results of games from March 20. So while they are not useful for bracket-making (past the deadeline), they are still free of potential data leakage.
The games listed here explore the bracket hypothetical case that each of the picks from the first round will be / were correct.
Date
Region
Team 1 (Seed)
Team 2 (Seed)
Team 1 Win %
Team 2 Win %
South
Marquette (7)
Michigan St (2)
36.39%
63.61%
Midwest
Xavier (11)
Kentucky (3)
27.52%
72.48%
West
Florida (1)
UConn (8)
51.94%
48.06%
East
Saint Mary's (7)
Alabama (2)
33.16%
66.84%
East
Duke (1)
Mississippi St (8)
73.85%
26.15%
Second Round - Day 1 (Updated with data on games through March 21)
Date
Region
Team 1 (Seed)
Team 2 (Seed)
Team 1 Win %
Team 2 Win %
Midwest
McNeese (12)
Purdue (4)
23.40%
76.60%
West
Arkansas (10)
St. John's (2)
29.10%
70.90%
South
Michigan (5)
Texas A&M (4)
42.94%
57.06%
West
Drake (11)
Texas Tech (3)
26.12%
73.88%
South
Auburn (1)
Creighton (9)
61.81%
38.19%
East
BYU (6)
Wisconsin (3)
39.09%
60.91%
Midwest
Houston (1)
Gonzaga (8)
69.78%
30.22%
Midwest
UCLA (7)
Tennessee (2)
28.78%
71.22%
Second Round - Day 2 (Updated with data on games through March 22)
Date
Region
Team 1 (Seed)
Team 2 (Seed)
Team 1 Win %
Team 2 Win %
West
Florida (1)
UConn (8)
50.95%
49.05%
East
Duke (1)
Baylor (9)
73.82%
26.17%
Midwest
Illinois (6)
Kentucky (3)
40.21%
59.79%
East
Saint Mary's (7)
Alabama (2)
30.46%
69.54%
West
Colorado St. (12)
Maryland (4)
34.75%
65.25%
South
Mississippi (6)
Iowa St. (3)
36.09%
63.91%
South
New Mexico (10)
Michigan St. (2)
29.74%
70.26%
East
Oregon (5)
Arizona (4)
41.68%
58.32%
Sweet 16- Day 1 (Updated with data on games through March 26)