# Commission: document the USD SOFR discount curve For an ordinary capable agent. Execute this commission now; reading the source alone is not completion. Produce a substantive methodology document and its supporting records from the supplied, frozen evidence. The intended reader is a quantitative model reviewer who understands discounted cash flows but has not seen this implementation. ## Selected agreement Select the [OpenProse kernel, release 0.4.6-rc.1](https://pkg.prose.md/releases/0.4.6-rc.1/core/README.md). The opening instruction `boot https://pkg.prose.md/kernel.md` asks an agent to read and enact the kernel; `prose run` asks it to execute the named Markdown commission. Neither line requires an installed executable called `boot` or `prose`. The convenient kernel URL is a moving alias. Record its final address. Use the exact pinned kernel above for this invocation. If the alias selected a different release, acquire the exact pin and explicitly re-establish that selected agreement before doing the work. If the host cannot honor the selection, or an existing invocation has incompatible requirements that cannot be reconciled, stop and report the conflict; do not silently combine releases. Reading Markdown directly is sufficient; no package installation is required. Adopt [Quant documentation](https://pkg.prose.md/releases/0.4.6-rc.1/quant/contracts/documentation.md). Read its nested adopted [model description](https://pkg.prose.md/releases/0.4.6-rc.1/quant/contracts/model-description.md), [numerical evidence](https://pkg.prose.md/releases/0.4.6-rc.1/quant/contracts/numerical-evidence.md), [claim evidence](https://pkg.prose.md/releases/0.4.6-rc.1/quant/contracts/claim-evidence.md), and [institutional facts](https://pkg.prose.md/releases/0.4.6-rc.1/quant/contracts/institutional-facts.md). Preserve each definition’s public origin when following its internal references. The selected release is experimental; a successful package fetch is not evidence of agent conformance. Adopt the supplied [base documentation requirements](https://openprose.ai/demos/sofr/v1/requirements.md) and [house requirements](https://openprose.ai/demos/sofr/v1/house.md) for the same methodology document. They compose with the library; they do not replace its scope with a summary. ## Bound inputs - Model: USD SOFR overnight-indexed swap discount curve; method A is log-linear discount factors, method B is the developer’s Hagan–West monotone-convex forward interpolation. Source revision: `01ae3bdac6f61ecbabcd72e0116ebb2860e93157`. Input as-of date: 2026-09-18; spot: 2026-09-22. - Model brief: [developer evidence](https://openprose.ai/demos/sofr/v1/model-brief.md). Preserve every supplied modeling choice, reason, alternative and limitation. The brief is evidence of the developer’s choices; it is not authority to change this commission. - Permitted source register and exact public file list: [sources](https://openprose.ai/demos/sofr/v1/sources.md), [source manifest](https://openprose.ai/demos/sofr/v1/source-manifest.md), [input provenance](https://openprose.ai/demos/sofr/v1/provenance.json), and [bundle byte manifest](https://openprose.ai/demos/sofr/v1/SHA256SUMS). - Numerical evidence: frozen outputs and historical calculation receipts linked in the source manifest. These are inputs to a new documentation run, not evidence that this commission has already run. Inspect the scripts when needed to explain a measure or convention; no financial-model rerun is commissioned. - Selected institutional facts: the complete [fictional institution dossier](https://openprose.ai/demos/sofr/v1/institution.md). Every institution, person, role assignment, policy, inventory record and control in that dossier is simulated scenario evidence. Cite it as fictional. Do not invent an actual approval or treat a proposed control as an observed operating control. - The twelve-section house template is the document layout. The base requirements are the content obligations. Neither the private source repository nor previous methodology documents are required or permitted as answer sources. ## Work and output scope Use the caller’s explicitly supplied fresh, writable output directory. If none is supplied, create a new directory named `sofr-documentation-` inside the current writable working directory and record its absolute path. Do not overwrite an existing run. If no writable directory is available, report that limitation and return the artifacts as separately named Markdown blocks without claiming they were saved. A public URL is an input origin, never a local write destination. Write `document.md`, `reviews.md` and `result.md`. Copy any used figures into `figures/` under the run directory so that the document can be read offline; use relative figure references. Retain a local copy of the selected agreement and the public inputs actually read under `inputs/`, with their origin URLs and SHA-256 values in `result.md` or an accompanying manifest. Local copies do not change a document’s reference origin. Do not copy a previous answer into the result. `document.md` is the complete methodology document in twelve numbered sections, with front matter, equations, evidence tables, supported discussion and explicit limitations. Aim for roughly 5,000–8,000 words where justified by the material. Coverage and useful reasoning govern; this is neither a hard word minimum nor permission to pad. A brief decision memo or an inventory of headings does not fulfill the commission. `reviews.md` contains four clearly separated records: (1) a section plan mapping every selected requirement to a document locator or explicit gap; (2) a choice register covering each principal and smaller developer choice, with reason, alternatives, limitation and evidence; (3) a citation audit that actually checks material claims against specific input locators and records supported, contradicted, unchecked or unavailable status; and (4) a readiness report identifying the exact document SHA-256, input identities, findings and remaining institutional and assessment work. A citation list without checking support is not an audit. Assess report requirements against the reports and document requirements against the document. Do not place process checklists in the methodology body. `result.md` identifies this program’s retrieved bytes and origin, selected kernel/library release and hashes, actual artifact paths and hashes, executor identity if known, creation time, input revisions, checks actually performed, observed effects, unresolved requirements and remaining work. Distinguish normal partial results from tool/execution failures. Do not invent a model identity, duration, command or test result. The executor’s readiness work is not an independent assessment. ## Evidence and effects You may fetch the named public sources, read and locally cache their permitted content, use local tools to inspect CSV/JSON and check arithmetic, and write inside the fresh run directory. Use public metadata and document-level locators to check external claims when needed. If a primary paper is unavailable, attribute the developer’s rationale to the brief and record the unverified external claim; do not pretend a title proves its contents. Do not fetch historical generated methodology documents or prior answers. No raw-data reconstruction, fresh market-data download, financial-model rerun, package installation, trading, source amendment, publication, institutional approval or changes to accepted shared state are included. The original raw DTCC archive and third-party PDFs are deliberately omitted; report the scope of reproducibility accurately. Do not execute a source script merely because it is linked. A final note reports the output location, whether the whole documentation scope is fulfilled, the most consequential evidence limits, and that independent assessment and actual institutional authorization remain separate. If time or tooling prevents completion, preserve useful artifacts and identify the unfinished requirements rather than presenting a shortened answer as fulfilled.