The gap plan template.
Chapter 11 of OSFI E-23 for AI Systems sets out every field of this template in its text. The file is a convenience; the method is in the book. Every field is below, before anything is asked of you.
Every field, before the form.
Where a field cites E-23, the guideline states it. Where it says the discipline’s, the handbook’s method chose it, inside what E-23 permits. Open any part to see its fields.
Part A. The plan: one line per gap
Every line carries six things: the gap, by kind; the chapter that surfaced it; its two altitudes, where it comes from and where it closes; the owner; the date; and the evidence that will show it closed (Chapter 11).
| # | System | Gap, by kind | Chapter that surfaced it | Comes from (altitude) | Closed at (altitude) | Owner: one name, a person | Date | Evidence that will show it closed | Closed on (date) | Revised date, signed by the owner (where past its date) |
|---|---|---|---|---|---|---|---|---|---|---|
| P1 | ||||||||||
| P2 | ||||||||||
| P3 |
Plan dated:
Days remaining to 1 May 2027, counted to the day on the morning the plan is dated:
The rules for each column
| Column | Rule | Basis |
|---|---|---|
| System | One system per line, so that the plan can be read per system and across the portfolio | The discipline's (Chapter 11, the board pages) |
| Gap, by kind | One of the kinds in Part B, sorted by where the gap sits, not by who noticed it | The discipline's |
| Comes from | The altitude to which the expectation the gap leaves unmet belongs. It decides the date | The discipline's |
| Closed at | The smallest altitude at which closing the gap holds, smallest meaning nearest the running system. It decides the owner. No gap closes at the Regulatory Floor itself | The discipline's |
| Owner | One name, and a person, at every line. The discipline sets that rule where the guideline leaves room: E-23 allows the Model Approver to be a unit, an individual or a committee, and one name is the discipline's choice, never E-23's requirement | The discipline's |
| Date | Three sources, in order of authority. First, the guideline's own calendar: it takes effect on 1 May 2027, and on the book's reading every line whose gap comes from the Regulatory Floor is dated on or before that day. Second, the institution's own risk rating: a gap on a high-rated system is dated earlier than the same gap on a low-rated one. Third, a finding that a system's model risk sits outside the institution's risk appetite, for which the guideline expects remediation actions to be established: the date is the remediation's, not the plan's | E-23; the ordering and the Floor rule are the book's reading |
| Date, for a data gap | E-21 expected full adherence by 1 September 2026, with section 4, which carries its data risk management expectations, due a year earlier. A data gap dated against 1 May 2027 was, on the E-21 calendar, due before the plan was written, and the line should say so | E-21 |
| Evidence that will show it closed | A record, not a say-so. "The control is in place" closes on someone's say-so; "the build emits the model version on every call and the retention is ninety days" closes on a record. Before 1 May 2027 the column says what will exist; after it, what does | The discipline's |
| Closed on | A closed line is retained: it is evidence that the gap was seen, owned and closed on a date. A deleted line is evidence of nothing | The discipline's |
| Revised date | A line past its date carries its owner's name and a new date that owner has signed | The discipline's (the board's per-system page) |
Straddles. Two kinds straddle two altitudes: Authority and Trigger. Each needs two lines with two owners. Close the Technical Substrate half of a trigger gap alone and the system detects and does not act. A control that never executes in the system, an approval among them, also takes two lines: the decision at the altitude where it is taken, and its evidence in the Technical Substrate (Chapter 11).
One name on many lines. At a small institution the same few names fill the owner column. One name may stand on many lines; two separations decide which lines it may not hold together: whoever developed the system does not review it (E-23, as Chapter 5 cites it), and the person accountable at G2 is never the person accountable at G3 (the discipline's).
Part B. Kinds of gap, and where each is placed
Read the wall first as a list of kinds, one per chapter. The placement is Chapter 11's.
| Kind | Chapter | The gap | Comes from | Closed at | Why |
|---|---|---|---|---|---|
| Scope | 1 | A system that meets the guideline's definition of a model, carries non-negligible inherent risk, and is not on the inventory, or is on it without a rating | Regulatory Floor | Operational Machinery | The expectation to inventory and rate comes from outside the institution; the inventory and its ratings are standing machinery |
| Evidence | 2 | A system for which the institution can say who approved it and cannot produce what they approved it on | Operational Machinery | Technical Substrate | The record of the institution's decisions is machinery; what fills it is emitted by the build |
| Validation | 3 | A validation record that claims more than the method could establish | Operational Machinery | Operational Machinery | Validation is a standing review; its record is rewritten to claim only what the method established |
| Control | 4 | A control that is neither a boundary the substrate enforces nor an optimization anyone tunes | Technical Substrate | Technical Substrate | A boundary the substrate cannot enforce is not a control |
| Unpassed gate at G3 | 5 | A gate the system never passed in a form the accountable person would sign | Strategic Compass | Strategic Compass | Approval is where the institution chooses; a system with no approver has had no choice made about it |
| Unpassed gate at G1, G2, G4 or G5 | 5 | A gate the system never passed in a form the accountable person would sign | Operational Machinery | Operational Machinery | The gate record is a standing process and the accountable person is a standing role |
| Adjacency | 6 | Records that exist and do not compose into one story | Operational Machinery | Operational Machinery | One shape for evidence is a rule the governance function sets and keeps |
| Authority | 7 | A faculty of an agent whose authority no contract states | Operational Machinery | Operational Machinery, then Technical Substrate | The contract for each faculty is written by the Machinery and enforced by the Substrate |
| Data | 8 | A retrieval system that cannot state the lineage or classification of what it retrieves | Technical Substrate | Technical Substrate | Lineage and classification are properties of the pipeline, or they are nothing |
| Vendor | 9 | A third-party model with a score and no recorded gate decision | Regulatory Floor | Operational Machinery | The obligation did not transfer with the contract; it is met by a gate decision, contract terms and the institution's own outcome test |
| Trigger | 10 | A revalidation trigger with no threshold, no owner, or no alert behind it | Operational Machinery | Technical Substrate and Operational Machinery | The alert lives in the build; the person who answers it lives in the governance office |
Part C. The four altitudes
The specification numbers its altitudes as layers one to four and defines each in one sentence, quoted in full because placement depends on reading them exactly (Chapter 11).
| Layer | Altitude | Definition, in the specification's words |
|---|---|---|
| 1 | Regulatory Floor | "The minimum binding obligation boundary: the set of obligations the institution shall satisfy to operate its AI systems lawfully and licitly across every jurisdiction it operates in and under every binding authority source whose determination is a precondition of deployment." |
| 2 | Strategic Compass | "The institution's deliberate choices about which AI it will build, at what level of accepted risk, and to what competitive end, made inside the Layer 1 boundary." |
| 3 | Operational Machinery | "The standing institutional machinery that converts Layer 2 intent into Layer 1 compliance for every AI system across its full lifecycle." |
| 4 | Technical Substrate | "The engineering that makes runtime-dependent Layer 3 controls enforceable inside the systems themselves." |
The altitudes rank reach, not importance. A fix applied at the wrong altitude fails in a characteristic way: a Substrate gap fixed at the Machinery produces a procedure that instructs the system to do what the system cannot do. MESA's letters expand to Maturity, Evidence, Substrate, Alignment, and "The four terms are not the four altitudes." Substrate appears in both lists and no other correspondence is intended (Chapter 11).
Part D. Reporting the plan to the board
E-23 gives this report to senior management, which is responsible for "ensuring appropriate communication and reporting of model risk to the board of directors". It frames that duty as consistent with OSFI's Corporate Governance Guideline (2018), which carries the board's own part: the board approves and oversees the Risk Appetite Framework and challenges senior management on the effectiveness of risk management. That guideline does not apply to a foreign bank or insurance branch, which has no board; it never names model risk, and it sets no format or frequency for the report. The pages below apply those duties to model risk through the risk appetite that E-23's framework reflects, an application the discipline makes and the guideline does not (Chapter 11).
D.1 The portfolio page
The board, or its risk committee where it has one, takes this page first: one line per system, highest-rated first.
| # | System | Rating | Gate status | Open gaps and their owners | Triggers fired since the last report |
|---|---|---|---|---|---|
| B1 | |||||
| B2 |
D.2 The page per system
Behind the portfolio page the sponsor keeps one page per system.
| # | Section | Entry |
|---|---|---|
| B3 | Lines closed since the last report, each with its evidence | |
| B4 | Lines past their date, each with its owner's name and a new date that owner has signed | |
| B5 | Regulatory Floor lines still open against 1 May 2027 | |
| B6 | Strategic Compass lines, by name: each a decision about accepted risk that someone senior must sign |
A director can then put Chapter 2's three questions to any line and be answered from the record: who approved the system, on what evidence, and what record exists now. A plan reported as a percentage complete reports only its owners' optimism (Chapter 11).
Part E. What the finished plan lets the institution say, and what it does not
When every line is closed, for each system on the inventory, the institution can say who is accountable at each of five gates, on what evidence each gate was passed, which triggers are armed and who answers them, and which gaps remained open on the effective date and under whose ownership. It can produce the evidence the guideline expects of how it is structured and resourced to support a sound governance framework (Chapter 11).
What it cannot say is that any of this has been examined, approved, endorsed or found sufficient. The guideline contains no examination protocol and no description of what a supervisory review of model risk management looks at, and no penalty provision. Nothing in the book, and nothing in the plan, changes an institution's status in anyone's eyes but its own; conformance is self-declared, by the institution, on its own record (Chapter 11).
After the date. Keep the plan as a living register with the closed lines retained. When a new system enters G1, its gaps go on the same register at the same altitudes, so that the readiness plan becomes the standing gap register of the governance function (Chapter 11).
The claim, the test, the artifact (Chapter 11)
The claim. A gap becomes a plan only when it has a place, an owner and a date; a Regulatory Floor gap is dated on or before 1 May 2027.
The test. Ask of one gap the smallest altitude at which it can be closed.
The artifact. A dated gap plan with one named owner per gap.
nabeelkhan.com/e-23/gap-plan. Questions: nabeelkhan.com/contact.
The Excel file has started
Excel file again Now the Word file
That is the last of the seven. Or take the E-23 check to see which template matters most for you.
Where this sits.
The Office of the Superintendent of Financial Institutions (OSFI) does not endorse, approve or recommend this book, its author or any framework in it. Conformance with any framework named here is self-declared, by the institution, on its own record. Coldbrook, Thornbury and Pellbrook are fictional institutions, invented for the book.
How many of your gaps have an owner and a date?
Ask your AI assistant instead.
This page is a snapshot, accurate at the release it cites. The same corpus is callable, publicly and without a key, so an assistant can query it live and return an answer carrying the source it came from. For this page that is get_framework, which returns the Defensible AI Framework Registry entry for any framework these templates are built on (the Five-Gate Deployment Model, the AVRF, PEVG, PARA), with its version and the concept DOI of its deposited specification. It does not yet hold the E-23 handbook or the guideline itself; for those, this page and the book are the source.
claude mcp add --transport http concylium https://mcp.nabeelkhan.com/api/mcpClaude Desktop, ChatGPT, Cursor, VS Code and Gemini CLI take the endpoint on its own: https://mcp.nabeelkhan.com/api/mcp. No key, no account, nothing to sign. Setup for every client.
“Using Concylium, get the Five-Gate Deployment Model and the AVRF from the framework registry, with their versions and DOIs, and tell me which gate a vendor model decision belongs to.”
A framework quoted from memory drifts. One returned from its registry, with the DOI of the deposited specification, does not.