# Fictional institution dossier — Harbor Example Research **All institutional content in this file is simulated.** Harbor Example Research, its people, inventory, systems, findings, policies and records are invented scenario inputs supplied by the demo author. They are not claims about an actual institution. The historical market data and numerical outputs are separately identified in the source register. This dossier is evidence of the scenario’s supplied facts, not evidence of live operation or real approval. Dossier revision: `HER-scenario-1`, authored 2026-10-12. Scenario model reference: `HER-QM-017`, revision `sofr-snapshot-2026-09-18`. Documentary status: proposed sandbox methodology, awaiting documentation assessment. No production deployment or trading is authorized. ## Inventory and intended use | Field | Supplied fictional scenario fact | |---|---| | Model name | USD SOFR OIS discount curve | | Inventory ID | HER-QM-017 (simulated) | | Tier | Tier 2 in the fictional policy: a material quantitative method with limited sandbox use and no production exposure | | Model owner role | Rates Research Lead, represented by fictional person Maya Chen | | Developer role | Curve Research Analyst, represented by fictional person Daniel Ortiz | | Review coordinator role | Model Review Lead, represented by fictional person Priya Shah; no review has been performed by this fictional person | | Monitoring owner role | Market Data Controls Lead, represented by fictional person Elias Brooks | | Accountable authorization role | Research Model Committee; decision pending in this scenario | | Information date | Numerical snapshot as of 2026-09-18; dossier authored 2026-10-12 | | Intended readers | Quantitative reviewers, rates research users and model-risk coordinators | | Approved uses | None. Proposed sandbox uses are method explanation, historical interpolation comparison and illustrative cash-flow sensitivities | | Downstream consumers | Proposed sandbox notebook `curve-research`, proposed demonstration report `rates-methodology`; no actual integrations or deployed downstream models | | Product scope | Hypothetical USD cash flows discounted at SOFR and single-curve compounded-SOFR OIS; no actual customer positions | | Portfolio | No real portfolio supplied. USD 100 million is a unit exposure used to explain a specified single-cash-flow difference, not an inventory of assets | The fictional tier is a supplied scenario classification, not a conclusion that would be appropriate for a real bank. A real deployment would have to determine its own materiality from actual exposures, reliance and controls. ## Data lineage and proposed daily process The scenario’s research intake starts with the retained NY Fed fixing and DTCC public dissemination report. The original developer’s `derive_quotes.py` filters reported executions and estimates an hour-16 New York close from adjusted median prints; `bootstrap.py` builds both methods. Public files carry their own historical hashes and stage receipts. This is the **demonstration lineage**. There is no internal licensed market-data feed, real market-data entitlement, observed reconciliation process or operational recovery record. The proposed fictional production design would replace derived public prints with a licensed, reconciled input feed, retain timestamped snapshots, reject missing or incompatible conventions, and retain instrument-level eligibility and exclusion reasons. That design is not implemented by this bundle. No source entitled to establish a real vendor close is supplied. The last published fixing known at the historical close must be distinguished from subsequently realized fixings available only for ex-post checks. Proposed snapshot policy: choose one consistently defined New York close, preserve trade and publication times separately, and flag a fixing not published when expected. Proposed stale-input rule: hold a candidate curve if any required input is missing, stale under its feed-specific service agreement, or has inconsistent conventions. No feed-specific service agreement or actual stale-quote tolerance is supplied; that remains a gap. The historical hour-16 estimate must not be presented as the institution’s licensed closing curve. ## Proposed controls and monitoring thresholds Every control in this table is a fictional proposed requirement. None has operating-effectiveness evidence. Owners are roles defined above, not real approvals. | ID | Risk or limitation | Proposed fictional control / trigger | Owner and response | |---|---|---|---| | C-01 | Calibration or solver error | Absolute par-rate error at most 0.01 bp for every input, with complete instrument table; any breach holds the candidate | Developer diagnoses and records failure; review coordinator determines whether a new assessment is needed | | C-02 | Weak or noisy derived tenor estimates | At least 5 qualifying prints and adjusted IQR at most 2 bp, preserving the original historical eligibility rule | Market Data Controls Lead marks unsupported tenor unavailable; no invented replacement quote | | C-03 | Public-print estimates mistaken for executable prices | Label every derived quote as a historical estimate; obtain a licensed benchmark before any proposed production use | Model owner holds production proposal pending benchmark evidence | | C-04 | Poor pricing outside the input set | Median absolute residual at most 1 bp on the defined held-out cohort, with separate spot and forward-start summaries | Developer investigates any breach; common-adjustment dependence remains a limitation even when below threshold | | C-05 | Sparse long end | Require a documented review of 20Y–30Y shape, input concentration and A/B disagreement for every future candidate | Model owner restricts long-end reliance pending evidence; this control supplies no quantitative production exposure limit | | C-06 | Extrapolation | Proposed sandbox displays beyond 30Y must carry an extrapolation label; production use beyond the last node prohibited pending a separately authorized policy | Model owner removes unauthorized use; no actual enforcement system exists | | C-07 | Locality and unstable sensitivities | Review the 5Y +1 bp comparison and the full 4.5Y cash-flow ladder; a proposed future campaign would bump all inputs across multiple dates | Developer records spillover; comprehensive risk-system bump policy remains unavailable | | C-08 | Negative-rate or regime change | Stop extension of this positive-rate snapshot’s conclusions to a negative-rate setting until collar behavior and admissible shapes are assessed | Model owner commissions a separate scenario study; no threshold inferred from this one date | | C-09 | Shared schedule/calendar defect | Independently reconcile business-day calendars, accrual boundaries and payment dates before production consideration | Review coordinator obtains separate evidence; agreement between two methods sharing schedules is insufficient | | C-10 | Model or dependency change | New code, convention, library version or input transformation requires a new immutable change record and impact assessment | Developer retains old result identities; committee decision remains separate | | C-11 | Evidence or document drift | Match the final document, source manifest and assessment by SHA-256 before any presentation of assessed results | Review coordinator preserves prior assessments and does not relabel them as applying to new bytes | The proposed recurring cadence is each candidate build for C-01 to C-04 and C-11, plus a quarterly review of limitations and dependencies and an event-triggered review for convention, data, model or regime changes. This dossier does not schedule jobs or prove those reviews occur. Escalation means recording the exception and withholding the proposed use, not silently changing a tolerance to fit the result. ## Implementation and change record Proposed system name: `HER research sandbox` (fictional). Historical numerical environment supplied by the developer: Python 3.12.14, QuantLib 1.43 and SciPy 1.18.1; further package versions and historical stage timestamps are in the actual retained environment JSON files. Those JSON files describe the historical numerical run, not a deployed institutional environment. Scenario change `HER-CR-001`: prepare the first methodology document for the fixed September 18 data. Scope is documentation from frozen outputs. Model code, financial outputs, input filters and original evidence are not changed. There is no real implementation ticket, production release, vendor agreement or disaster-recovery test. Document version history begins with the actual newly created `demo-v1` artifact and its recorded hash; do not invent prior document revisions. ## Open findings and disposition | Finding | Supplied scenario status | Evidence needed to close | |---|---|---| | F-01: licensed benchmark missing | Open; blocks production consideration | Actual benchmark curve, snapshot identity, eligibility, comparison tolerance and discrepancy analysis | | F-02: multi-date and stressed-regime assessment absent | Open | Authorized campaign with retained results and separate assessment | | F-03: calendar/convention reconciliation not independent | Open | Review performed with a separately sourced calendar and schedule reconstruction | | F-04: proposed controls have no operating evidence | Open | Real ownership, implementation, monitoring records, breaches and resolutions | | F-05: document assessment pending | Open at dossier authoring | Exact newly produced document and a separately performed assessment tied to its hash | These are fictional scenario findings authored for the commission, not actual past validation findings. No real prior findings or resolution history are supplied. A later document self-check may report its own state, but must not retrospectively claim that these fictional people reviewed it. ## Approval and required real-world gaps Scenario committee decision: **pending**. Sign-off block: **not signed**. Approval conditions are not known because no committee decision has occurred in the scenario. Proposed conditions for consideration include resolving F-01 through F-04 and separately assessing the final documentation; these proposals are not attached approval conditions. The methodology must preserve visible gaps for actual institutional inventory/tier, internal feed lineage and entitlements, real usage/exposure inventory, actual roles and authorization, real prior validation findings, approval conditions, monitoring implementation/effectiveness, production instrument/convention set, licensed benchmark and tolerance, fixed snapshot/stale-quote policy, extrapolation authorization, full risk-system bump methodology, actual implementation/change ticket and real signatures. The simulated dossier supports a coherent example of how those topics are documented; it does not replace their real evidence.