DE.CIPHYR

4.1 Credits (Compute Currency)

The base accounting layer of the platform is the compute credit. Pegged strictly at 1 credit = $0.01 USD, it provides a stable, non-volatile unit for all compute pricing on the network.

Users purchase credits via standard payment infrastructure (Stripe). There is no cryptocurrency, wallet setup, or blockchain interaction required. Credits are immediately available for use across all supported models.

Credit balances are backed by USD deposits held in regulated payment infrastructure. Unused credit balances are refundable subject to the terms of service.

4.2 Benchmark-Pegged Pricing Rates

De.Ciphyr maintains a continuous evaluation engine that tests all listed models daily against rigorous datasets. The resulting scores programmatically adjust the pricing rate (credits per 1M tokens) for that specific model.

• If GPT-5.5's reasoning score drops, its compute becomes cheaper on De.Ciphyr.
• If an open-source model releases an update that outperforms proprietary models, its pricing rate increases to reflect the higher quality.

4.3 Platform Stability Framework

De.Ciphyr employs multiple safeguards to ensure pricing stability and operational reliability:

• Pricing Rate Circuit Breaker: If any model's pricing rate changes by more than 25% in a single recalibration cycle, rate updates are temporarily paused for that model pending manual review. This prevents cascading effects from provider-side pricing shocks or benchmark anomalies.
• Provider Fallback: When a primary provider experiences downtime, De.Ciphyr automatically routes requests to equivalent alternative models, maintaining service continuity.
• Rate Change Notifications: Significant pricing rate changes are published with advance notice, and historical rate data is available via the public API for transparency.

4.4 Pricing Formula Specification

Pricing rates are derived from a transparent, hybrid cost-plus-benchmark formula:

The α parameter controls how strongly benchmark performance influences pricing beyond raw cost. This parameter is published with each recalibration cycle and may be adjusted with a minimum 7-day notice period.

4.5 Benchmark Integrity Framework

To defend against Goodhart's Law—where providers optimize specifically for known benchmark tests rather than genuine capability—De.Ciphyr employs a multi-layered evaluation architecture:

• Multi-Axis Scoring: Models are evaluated across four independent axes—Code Generation, Reasoning, Tool Use, and Instruction Following—preventing single-dimension gaming.
• Hidden Test Sets: 30% of daily evaluation uses proprietary, rotating test sets that are never published. These sets are refreshed monthly to prevent data leakage.
• Adversarial Prompts: 20% of evaluation uses adversarial prompts specifically designed to detect overfitting, including paraphrased variants of known benchmark questions to identify memorization.
• Contamination Detection: Models are tested for verbatim reproduction of known benchmark answers. Statistical anomalies in response patterns trigger manual review and potential delisting.
• Methodology Transparency: Aggregate benchmark methodology and scoring weights are published openly. Individual test cases from the hidden set are rotated monthly with a 30-day embargo before public release of retired test items.

6.1 API Gateway

Developers integrate with De.Ciphyr using a single, unified API endpoint that is fully compatible with existing OpenAI SDKs. Requests can be routed dynamically based on task priority (e.g., routing to the cheapest model with a benchmark score > 80 for basic tasks, or the highest-quality model for critical operations).

6.2 Execution Layer

The API routing gateway operates at the Edge to guarantee less than 50ms of gateway compute overhead. Requests are authenticated, metered, and forwarded to upstream providers with minimal latency impact.

Billing is fully atomic—credit deductions are processed using PostgreSQL atomic operations with idempotency keys to prevent double-charging, even in the event of network interruptions or retries.

6.3 Payment Infrastructure

All payment processing is handled via Stripe. Credit purchases, auto-topups, and refunds flow through standard, PCI-compliant payment rails. De.Ciphyr does not hold cryptocurrency, manage wallets, or interact with any blockchain network.

7.1 Risk Factors

• Pricing Rate Volatility: Model pricing rates may fluctuate due to provider pricing changes, benchmark methodology updates, or shifts in model performance.
• Benchmark Methodology Risk: While De.Ciphyr employs multi-axis hidden-set evaluation, no benchmark methodology is immune to all forms of gaming. The platform continuously evolves its evaluation approach.
• Provider Dependency: De.Ciphyr depends on upstream provider API availability. Provider outages, rate limiting, or silent model version changes may affect service quality.
• Data Privacy: API requests are routed through De.Ciphyr's infrastructure. Users should assess data sensitivity before using any model, and review upstream provider data policies.

7.2 Governance

De.Ciphyr operates under centralized governance, with all parameter changes (benchmark weights, tier thresholds, α sensitivity) subject to a published change policy with a minimum 7-day notice period before implementation.

The roadmap includes introducing community feedback mechanisms for parameter adjustments, ensuring the platform evolves based on user input and transparent decision-making.

7.3 Legal Status

De.Ciphyr is a compute services platform. Credits represent prepaid compute access, not securities, currencies, or investment instruments. Credits do not represent equity, profit-sharing, or governance rights in Ciphyr AI.