A passed software test is not called independent validation.
ACDS model governance
How ACDS governs model risk.
Inspect model ownership, approved uses, release controls, limitations, and the work required before ACDS makes stronger validation claims.
- 2 governed analyzers
- Frozen core formulas
- Pending independent review
Ratios are compared only after their inputs use the same basis.
Material model changes create a new reviewable release.
Reports preserve warnings, exceptions, and prohibited uses.
Project responsibility
Accountability should be visible.
ACDS is independently developed in Ontario. Responsibility for the current models, documentation, monitoring, and releases sits with ACDS model development. Future professional reviewers and approvers have a separate, objective role.
ACDS Decision
Questions about model methods, documentation, or a release can be submitted through the ACDS contact page.
- Current responsibility
- Model development, documentation, monitoring, and releases
- Project location
- Ontario, Canada
Model inventory
Every consequential model needs an owner and a status.
The register follows a proportionate model-risk structure. The current founding beta separates development ownership from independent review, but outside reviewers and approvers have not yet been appointed.
| Model ID | Model and version | Risk classification | Model owner / developer | Monitoring status | Independent review |
|---|---|---|---|---|---|
| ACDS-F-01 | Equipment Investment Analyzerv0.25.0 | High-impact decision support | ACDS Decision · Model Development | Internal automated verification | Not yet completed |
| ACDS-F-02 | Farmland Acquisition Analyzerv0.91.0 | High-impact decision support | ACDS Decision · Model Development | Internal automated verification | Historical-case protocol open |
ACDS-F-01
Equipment Investment Analyzer
v0.25.0- Origin
- Internally developed deterministic model
- Model dependencies
- User-entered assumptions; no external data feed
- Approved current use
- Equipment alternative screening, capacity testing, financing-pressure review, and decision-file preparation
- Material limitations
- Pre-tax; no CCA, legal, mechanical-condition, engineering, or lender-policy determination
- Most recent internal review
- Automated production release gate · September 4, 2026
- Next required review
- Frozen through 10 development cases, 5 blind cases, and objective review before the first paid release
ACDS-F-02
Farmland Acquisition Analyzer
v0.91.0- Origin
- Internally developed deterministic model
- Model dependencies
- User-entered property, farm, financing, evidence, and policy assumptions
- Approved current use
- Buy-rent-wait screening, projected generational-wealth comparison, guarded indefinite-ownership sensitivity, whole-farm capacity review, offer structuring, and decision-file preparation
- Material limitations
- Not an appraisal, environmental assessment, title review, tax or succession opinion, perpetual-growth forecast, or credit approval
- Most recent internal review
- Automated production release gate · September 4, 2026
- Next required review
- Before the first paid release or any material formula change
Method, implementation, documentation, monitoring, and releases.
Must be sufficiently objective from model development.
Future authority to approve defined higher-reliance uses.
And periodically after externally reviewed use begins.
Governance evidence in the file
The model-use record travels with the professional handoff.
A farmland Full Review package now includes a machine-readable governance record alongside the Decision File. This makes the version, permitted purpose, institution policy context, limitations, and prohibited uses easier to retain during a pilot or credit review.
Connects the result to the published model inventory and release.
Records the threshold profile applied without claiming it is authenticated.
Keeps decision support separate from approval, appraisal, and assurance.
Preserves the prioritized review queue in an editable CSV register.
Use boundary
Decision support is not decision authority.
- Compare entered investment alternatives on a consistent basis.
- Identify repayment, liquidity, collateral, renewal, and evidence constraints.
- Prepare questions, conditions, and supporting records for professional review.
- Document user-entered assumptions, their sources, and the model version used.
- Loan approval, credit adjudication, or a lender commitment.
- An appraisal, legal opinion, tax opinion, audit, review, or assurance engagement.
- Independent verification of user-entered records, property facts, or forecasts.
- A guarantee that the highest-ranked alternative will produce the forecast result.
Historical-case validation protocol
A repeatable path from internally tested to independently reviewed.
Case results must be anonymized and used only with permission. The protocol compares calculations and definitions; it does not treat a historical lending decision as proof that either conclusion was economically correct.
Freeze the model version
Keep the formulas and definitions unchanged while a group of cases is tested. Any correction creates a new version and restarts affected tests.
Normalize definitions
Reconcile cash available, annual debt service, lease treatment, current portions, and balance-sheet basis before comparing outputs.
Development comparison
Compare at least ten anonymized completed files covering financed, all-cash, liquidity-constrained, collateral-constrained, and renewal-sensitive cases.
Blind holdout
Run at least five additional cases that were not used to adjust the model. Record every difference before seeing the external conclusion.
Independent review
Have an objective agricultural lender, accountant, or qualified reviewer challenge conceptual soundness, definitions, calculations, limitations, and results.
Publish findings
Report procedures, case counts, differences, corrections, unresolved limitations, reviewer role, and review date without implying assurance that was not performed.
Proposed comparison criteria
Match arithmetic only after matching definitions.
These are preliminary acceptance criteria for reviewer approval. A difference outside tolerance is investigated and documented; it is not silently averaged away.
| Metric | Proposed criterion | Required condition |
|---|---|---|
| Annual debt service | Same to within $1 | Identical loan terms and payment timing |
| DSCR and current ratio | Same to within 0.01× | Definitions and statement dates reconciled |
| NPV and supported price | Same to within 1% | Identical cash flows, timing, horizon, and discount rate |
| Recommendation and constraint | Same classification or explained difference | Identical alternatives, policies, and evidence assumptions |
Reference framework
Built toward institutional review expectations.
ACDS is not an OSFI-regulated institution. OSFI model-risk guidance is used as a voluntary governance benchmark for ownership, independent review, approved uses, limitations, versioning, and monitoring.
OSFI Guideline E-23 — Model Risk ManagementFCC — Debt-service coverage ratioFCC — Financial statements and ratiosIAASB — Agreed-upon procedures