Numerical infrastructure for decisions that must hold up.
API23 is a precision-controlled computation system. It preserves computational state, runs demanding numerical work, verifies the result against a defined oracle, and produces a reviewable execution record.
The executive decision
API23/API23v2 is positioned as licensable financial and algorithmic software infrastructure for quantitative finance, risk, valuation, optimization, scientific computing, and regulated analytics.
Its value is control over numerical behavior: a buyer can define the precision, workload, comparison oracle, and acceptance receipt for a computation instead of receiving an unexplained number from an opaque pipeline.
Who buys
Quantitative finance teams, risk and valuation groups, financial software vendors, scientific-computing organizations, and enterprises that need auditable numerical workflows.
How it is packaged
Buyer-specific evaluation, field-of-use license, OEM integration, or recurring runtime agreement with an acceptance receipt and controlled technical disclosure.
Make important calculations explainable, repeatable, and controllable.
Financial and scientific systems often depend on long-running calculations that are difficult to replay, compare, or defend after the fact. API23 turns the numerical execution itself into a controlled artifact.
Problem
Two environments can produce different results because precision, state handling, rounding, or execution conditions were not explicitly controlled.
API23 utility
Bind precision, workload shape, error budget, and comparison oracle to the execution contract.
Buyer outcome
Engineering, risk, and audit teams receive a reproducible result with a receipt showing how it was produced.
Where it fits
Quantitative finance
Use it around valuation, risk, optimization, and scenario workflows where model outputs must be compared across runs and environments.
Scientific and engineering compute
Use it for repeated-state calculations, simulation, transformation, and matrix workflows that need controlled precision and replayable evidence.
Software vendors
Integrate it as a numerical component with a defined interface, buyer-specific acceptance harness, and controlled field-of-use rights.
Governed analytics
Give technical reviewers a trace from workload and precision choice to comparison result without exposing proprietary implementation mechanics.
Technical proof
API23 is a numerical execution layer built around a bounded transform, persistent state, controlled precision, exact composition, and an explicit comparison oracle. The proof is the chain from mathematical operator to executable state, conformance tests, real-data workload, and receipt.
Pinpoint utility: where it operates
API23 sits between a buyer's model and its numerical result. It receives a defined workload and precision contract, executes the calculation through the API23/API23v2 state path, compares the result with the buyer's named oracle, and returns the result plus an evidence receipt. It is the controlled numerical layer inside a valuation engine, risk service, optimization job, scientific pipeline, or software product.
| Buyer supplies | API23 operates | Buyer receives |
|---|---|---|
| Model inputs, operation graph, matrix/tensor shape, precision path, tolerance, and reference/oracle path. | State-preserving transform, wide-precision arithmetic, composition/collapse, deterministic comparison, and fault-boundary checks. | Computed result, error versus oracle, runtime facts, pass/fail acceptance result, and a reproducible receipt. |
Operating sequence
- Bind: name the workload, dimensions, precision, oracle, and acceptance threshold.
- Execute: run the model through the API23 numerical boundary.
- Compose: preserve and collapse repeated state transitions without replaying the entire chain.
- Compare: calculate error against the declared reference path.
- Record: preserve the output, parameters, comparison, and execution facts for engineering, finance, and audit review.
1 · Mathematical operator
2 · Product layers
| Layer | Public technical function | Recorded evidence |
|---|---|---|
| Transform | Deterministic scalar state transition used as the numerical primitive. | Exact oracle comparison and finite-output checks. |
| API23 | Precision selection, error-budget handling, deterministic output checking, and C-compatible integration boundary. | Symbolic CPU integration: 6 operations and 2 fault boundaries. |
| API23v2 | Persistent wide-state execution, exact state composition, and repeated-stage collapse. | FP16K/FP32K conformance suite: 11 / 11 passed. |
| Receipt layer | Records workload, precision, oracle, comparison result, and execution outcome. | 10,000 oracle samples; correlation 1.000000000000. |
3 · Conformance and oracle results
| Test | Recorded result |
|---|---|
| API23v2 conformance suite | 11 / 11 passed |
| FP16K and FP32K tiny-state retention | Pass |
| Cancellation and finite-positive H evaluation | Pass |
| Five-stage state collapse | Zero recorded collapse error |
| Deep collapse | One-billion-stage O(1) equivalence recorded |
| API23 ↔ API23v2 exact comparison | Zero recorded maximum error |
| Single-path error versus double comparison | 1.1102230246251565e-16 maximum recorded error |
| Handshake error versus double comparison | 8.881784197001252e-16 maximum recorded error |
4 · Real-data execution
| Workload field | Recorded value |
|---|---|
| Input | Two real frontal radiographic images |
| Matrix shape | 320×390 × 390×320 |
| Arithmetic path | FP32K / 32,768-bit arithmetic |
| Operation count | 79,872,000 FLOPs |
| Recorded runtime | 584.954798 seconds |
| Maximum error versus double comparison path | 0 |
5 · What a buyer can inspect
Before execution
Workload shape, precision path, oracle, error budget, and acceptance threshold are named.
After execution
The result, comparison error, state-composition outcome, runtime facts, and receipt identity are preserved for review.
Technical boundary
The current product is the software/API numerical runtime and its controlled integration boundary. These results demonstrate numerical execution and reproducibility on the recorded workloads; they are not a claim of ASIC silicon performance.
[primary project evidence] Technical white paper, retained conformance receipts, exact-oracle rerun, symbolic CPU integration record, and real-data GEMM execution record.
ZC Symbolic Compute and API23/API23v2
Abstract. The system treats numerical computation as a persistent, inspectable state transition rather than a disposable answer. Precision, state, composition behavior, execution boundaries, comparison oracles, and receipts are explicit parts of the workflow.
Architecture
The architecture begins with a symbolic transform and state model, adds an executable software path, then adds API23 precision and integration controls and API23v2 exact state composition.
Evidence method
Define the workload and oracle, execute the retained harness, capture the raw result, compare the paths, preserve the receipt, and register the evidence without mixing incompatible workloads.
Recorded results
Publicly summarized evidence includes 11/11 API23v2 conformance tests passed, zero recorded API23↔API23v2 error, and direct wide-precision execution over real image data.
Commercial relevance
The resulting capability is a controlled numerical substrate for financial modeling, risk, valuation, optimization, scientific computing, and other workflows where reproducibility matters.
What the paper establishes
- An executable wide-precision software path.
- Exact state-composition behavior under the tested conformance contract.
- Deterministic evidence receipts and comparison discipline.
- A deployable basis for buyer-specific evaluations and field-of-use licensing.
White paper version: public executive edition · October 2026 · Technical claims are summarized from retained project evidence.
Accounting view
The portfolio is proprietary software, algorithmic know-how, and related intellectual property. The accounting workstream is organized around the applicable transaction and reporting facts.
| Framework | Public-facing purpose |
|---|---|
| ASC 820 | Fair-value measurement; unobservable inputs are treated as Level 3 when applicable. |
| ASC 350-30 | Identifiable intangible-asset analysis, useful life, and impairment workstream. |
| ASC 805 | Identifiable intangible treatment in an acquisition or business combination. |
| ASC 606 | Revenue analysis for licenses, subscriptions, support, and evaluation services. |
This is a public accounting framework summary, not an entity-specific GAAP opinion. Final treatment depends on executed agreements, reporting facts, and the reporting entity’s accounting advisers.
Valuation method
The public rNPV method is a probability-weighted discounted cash-flow model. Each dated inflow and outflow is assigned the probability of reaching the event that creates it, then discounted to the valuation date.
Public methodology references
| Reference | What it contributes | Use |
|---|---|---|
| Public primary filing — algorithmic software/IP | Disclosed technology-IP rNPV example | Methodology and market context only |
| Public primary filing — software IP | Level-3 fair-value example | Corroborative accounting method; not API23 rNPV |
Primary public filing — technology-IP rNPV example · Primary public filing — software-IP fair value example · Primary public filing — rNPV method
Commercial packaging
Fixed evaluation
$50,000 fixed fee · 30 calendar days
50% at kickoff and 50% on delivery of the accepted evaluation report.
Annual production license
$250,000 / year · 12-month term
Paid annually in advance for one agreed field of use.
What the buyer receives
Configured runtime/API rights, buyer adapter, acceptance harness, standard updates, controlled support, reproducible receipts, and technical reporting.
Evaluation and license terms
| Term | Included position |
|---|---|
| Evaluation | One workload, one baseline comparison, one acceptance card, and one final readout. |
| Annual license | Non-exclusive, 12-month, field-of-use production rights; renewal requires an executed order form. |
| IP and source | No source-code transfer, patent transfer, trade-secret disclosure, or ownership transfer. |
| Data | Buyer data remains buyer-controlled; handling and deletion terms are stated in the order form. |
| Enterprise/OEM | Separate pricing and terms apply for additional fields of use, sublicensing, redistribution, or OEM embedding. |
[Derived commercial proposal] Fee and terms are a proposed API23 offer established for this public brief; the executed order form controls.
Request the evaluation and license order form