auction

auction is an experimental framework for managing a real-time auction for sending Notifications to users. Each team is granted a number of tokens which can be used to place bids on a specific user with a. Each bid is a transaction with a corresponding priority (0-10, 10 being highest priority). After receiving a bid, auction will calculate an associated cost of the bid, and if accepted, deduct that balance from the user's token quota. As priority increases, the associated cost of a bid also increases. auction also tracks the frequency for which given priorities are requested by teams. This frequency count is used to keep a running reputation score for a given team, in turn affecting the cost function which determines the price of a bid.

running locally

The included docker-compose.yaml will start a local instance of DynamoDB.

docker compose up

Running an auction:

go run cmd/auctiond/main.go

auction process

1. All teams begin with a fixed allocation of 1000 tokens and a reputation score of 100.
2. A team can bid on a given auction by specifying the targeted userID and a priority. The cost of a bid is calculated using the following function:

minMultiplier := 1.0       // No price increase at max reputation
maxMultiplier := 2.5       // 2.5x price increase at minimum reputation
priceMultiplier := minMultiplier + (maxMultiplier-minMultiplier)*(1-reputation)/100)

1. A team is eligible to bid if the cost of the bid is less than their current balance.
2. Bids are ranked according to a static weighting formula defined below.
3. The bid with the highest ranking wins and has the bid cost deducted from their balance.
4. The auction system also tracks a frequency count of priorities submitted by the bidding system.
5. If a team abuses a priority (more than 5 requests for priority 10 within a refill interval), their reputation score is penalized.

ranking bids

This describes a basic weighed ranking system to evaluate a bid based on its priority and reputation.

1. Normalize Priority: We can map priority from its original range (1-10) to a 0-100 scale.

$$ \text{{Normalized Priority}} = \frac{{(\text{{Priority}} - 1)}}{{9}} \times 100 $$

2. Normalize Reputation: If the reputation already falls between 0-100, we can use it directly. If not, we normalize it based on the max possible reputation score, let’s say maxReputation.

$$ \text{{Normalized Reputation}} = \frac{{\text{{Reputation}}}}{{\text{{maxReputation}}}} \times 100 $$

3. Weight the Components: You can assign a weight to each factor. For example, priority might be more important than reputation, so we could assign a weight of 70% to priority and 30% to reputation.

$$ \text{{Combined Score}} = (0.7 \times \text{{Normalized Priority}}) + (0.3 \times \text{{Normalized Reputation}}) $$

Example:

• priority = 9 (high priority)
• reputation = 50
• maxReputation = 100

The normalized values would be:

• Normalized Priority: $\frac{(9-1)}{9} \times 100 = 88.89$
• Normalized Reputation: $\frac{50}{100} \times 100 = 50$

Then, the combined score would be:

$\text{{Combined Score}} = (0.7 \times 88.89) + (0.3 \times 50) = 77.22$