# Cash, reserves and the statutory bridge

As of **19 September 2026**. All entities, contracts, valuations, model distributions and market quotes in this module are fictional. Amounts are USD; tables labelled `m` use USD millions. The official sources support the stated rules, not the fictional transactions or their prices. This is an educational design and standalone calculation artifact, not a production installation, accounting opinion or assertion of regulatory compliance.

The implementation is [reserve_ledger.py](../src/reserve_ledger.py), driven only by [numerical-inputs.json](../fixtures/numerical-inputs.json) and the public results of [numerical_fixture.py](../src/numerical_fixture.py). The accounting-unit and contractual choices are shared with [economic-specification.md](economic-specification.md), [capital-rulebook.md](capital-rulebook.md) and [valuation-capital-annex.md](valuation-capital-annex.md). The mutually exclusive liquidity events are in [wrong-way-branches.md](wrong-way-branches.md); the cash buffers in those events are not extra resources for this book.

## 1. What is being measured

Three different measurements must not be collapsed into one reserve balance:

1. **Actual cash and statutory stocks.** Cash comes from named executed transactions. A reserve charge, mark, capital release or AVA deduction does not create cash or a bank borrowing. The cash engine records the funded assets, legal title, sale financing, legacy funding, settlement payments, reinvestment, facility draws and unresolved claim.
2. **Management funding-value and XVA pots.** Positive funding opportunity `G`, hedge value `H`, gain deferral `D` and XVA deduction `X` form a management bridge. They are not automatically additional IFRS assets or liabilities. The base funding-unit calculation includes economic CVA and standalone funding CCR/CVA KVA; it is not all-in bank XVA.
3. **Prudential valuation and capital.** An identified, recognised fair-value exposure can support an incremental AVA. This changes the CET1 numerator, not cash or P&L. The controlled AVA example is separate from the retained-risk financing book and is not added to that book without an accounting-unit mapping.

The distinction corrects two causal shortcuts in the requested framing. Hedging P&L and actual funding interest are not, merely by naming them, reserve movements. Nor does a noncash charge become an unsecured debt balance because it is described as “unfunded.” The module includes both a charge-only counterexample and a separate executed cash transaction under which the proposed quadratic funding effect really occurs.

### Execution and solvency limitation

The base opening contributed equity is USD100m; the consumed setup service costs USD1m, leaving USD99m statutory equity and free cash. There is no automatic capital contribution. `fullbook_capital` compares the supplied stocks with the disclosed risk-based and leverage requirements and reports the additional qualifying capital required. Statutory equity is assumed to qualify as Tier 1 in that comparator; this is a synthetic quality mapping, not a legal conclusion.

The cash paths are **execution counterfactuals conditional on separate capital approval**. A path can be cash-feasible while undercapitalised or insolvent. The explicit deficit facility has no automatic acceleration triggered by the synthetic accounting loss. That contract assumption permits the cash mechanics to be displayed; it does not approve new execution, cure negative equity or make the bank solvent. Each composed cash path carries the opening `capital_execution_gate`. Its subsequent equity is shown, rather than topped up by a plug.

## 2. Cash and title over the full twenty years

### Opening and executed funding

Debit-positive journal balances are maintained independently of the reported attribution.

| Stage | Assets | Liabilities | Statutory equity |
|---|---|---|---:|
| Before the new transaction | Retained U 1,000m; unrestricted cash 100m | Legacy funding 1,000m | 100m |
| Sell legal title to U and retain substantially all risks through the TRS | Add cash 1,000m; keep U recognised | Add sale financing 1,000m | 100m |
| Make the five-year EM advance | Replace sale cash with deployment asset 1,000m | No additional liability | 100m |
| Consume the separate day-one service | Cash falls by 1m | Unchanged | 99m |

The provider P holds legal title to U. Failed-sale accounting keeps U on the desk's statutory balance sheet and recognises the consideration received as a financial liability. The TRS does not justify recognising a second asset for the same retained exposure. IFRS 9 paragraphs 3.2.15, B3.2.14 and B3.2.16(o) support this retained-risk treatment; paragraph 3.2.22 prevents an unsupported offset of the retained asset and associated liability. The fictional regulatory classification is separately evidenced and does not rewrite this accounting unit.[IFRS9]

The instrument has USD1bn original quantity and notional. Its coupon is OIS + 100bp. The executed financing coupon is OIS + 70bp, consisting of the locked initial 20bp input plus a 50bp execution margin. The cash engine binds this rate to `funding_contract.locked_financing_spread`, rather than allowing a later internal funding-curve move to reset the coupon. OIS is 4%; the five-year deployment pays OIS + 400bp; legacy funding pays OIS.

### Quarterly settlement identity

For quarter `i`, original quantity `Q=N`, clean price per unit `P_i`, interval `q=0.25`, reference distribution `dist_i` and locked financing spread `k`:

```text
reference_coupon_i = N * (OIS + 0.01) * q
reference_redemption_i = N at year 20, otherwise 0
price_return_i = Q * (P_i - P_(i-1))
dist_i = reference_coupon_i + reference_redemption_i
TRS_net_due_i = dist_i + price_return_i - N * (OIS + k) * q
```

The provider receives the actual U distributions; the desk receives only the contractual net TRS amount. There is **no second desk cash receipt for U's coupon or redemption**. The reference coupon is retained as an attribution input, not duplicated as a cash flow.

Before the final distribution, the clean reference price follows the published pricing convention in `reference_price`. At maturity its ex-distribution price is zero. The negative price return and the redemption distribution therefore offset principal; the final scheduled net cash flow is not a second USD1bn repayment to the desk. The retained U and the sale-financing balance are then extinguished through the provider's retention of U redemption, a noncash principal retirement. Revaluation and settled principal-return adjustments are recorded consistently so that the same U loss is not also booked as an independent derivative loss.

### Year five and year twenty

At year five the EM borrower returns USD1bn. That amount is explicitly reinvested in an OIS cash instrument; it does not disappear, remain a five-year EM exposure forever or earn the old EM spread after maturity. Its principal and OIS interest remain separately visible through year twenty. At year twenty the OIS investment returns USD1bn, which repays the opening legacy funding. The sale-financing retirement described above is a different principal path.

Unrestricted cash earns OIS. Contractual cash income first funds cash expenses. An executed unsecured deficit facility, with the published USD500m limit and own-bond funding spread, funds only an actual remaining deficit. Surplus unrestricted cash repays that facility. On exhausted capacity or absent execution, the unmade payment becomes an explicit payable and the path stops as infeasible; the engine does not print a fictitious settled twenty-year continuation.

For every quarter the public result contains opening and closing stock balances, legal title, distribution and price-return components, the net settlement, interest and principal flows, facility draw/repayment, any shortfall, journal entry identifiers and an independently computed equity change. The base, mark-only and reference-stress paths each cover all eighty quarters when their stated cash contracts remain feasible.

## 3. The two management pots and the P&L identity

Use positive values for assets/gains and positive deductions for liabilities/reserves:

```text
E_management = C + M + G + H - B - D - X
```

`C` is cash; `M` is the other recorded assets; `B` includes sale financing, legacy debt, actual deficit borrowing and unsettled payables. `G` is the funding opportunity; `H` is an actually selected hedge's clean value. `D` is the gross management gain deferral. `X` is management CVA/KVA plus any identified statutory claim allowance **once**.

For the funding and hedge components over a period, with their cash coupon income `CF_G`, `CF_H`, actual funding cost `f` and operating/service cost `o`:

```text
P&L = CF_G + delta(G) + CF_H + delta(H) - f - o - delta(D) - delta(X)
```

Whole-book attribution adds the separately recorded deployment, legacy-funding, reference-revaluation and OIS-cash results. Cash or debt principal transfers and returnable VM are excluded from P&L. They still change balance-sheet stocks and liquidity.

Define `G_roll`, `D_roll`, `X_roll` at the ending date with elapsed time and contractual cash flows but unchanged market factors and no discretionary action:

```text
carry = CF_G + G_roll - G_open
clean_mtm = G_close - G_roll
D_theta = D_open - D_roll
D_observability_or_action = D_roll - D_close
X_theta = X_open - X_roll
X_market = X_roll - X_after_market
X_action = X_after_market - X_close
hedge_pnl = CF_H + H_close - H_open
```

The implemented aggregation adds carry, clean MTM, deferral release, reserve theta, reserve MTM, reserve action and hedge P&L, then subtracts cash funding drag and operating cost. An instantaneous scenario has no carry or theta. A year of ordinary runoff has theta, even when spreads do not change.

**Expected reserve release is P&L once.** It is not an extra windfall on top of a closing-stock change that already includes it. Unexpected spread-driven release is also P&L, but represents a risk position rather than time accrual. If a reserve covers an actual future cash expense, its consumption and that expense must both be recorded; the separate prefunding example below demonstrates this rather than treating reserve consumption as free revenue.

### Pot D: frozen pricing-time release

The management opportunity at inception is:

```text
G_0 = N * 0.003 * A(20)
A(h) = q * sum_(j=1 to h/q) (1 + OIS*q)^(-j)
D_t = G_0 * A(20-t) / A(20)
```

The inception spread is frozen for `D`. Funding or provider spreads changing do not re-mark that schedule. A separate observability event can change the deferral only with evidence applicable to the actual accounting unit and the recognition rule. Merely being older, moving to a favourable mark or selecting a different internal spread is not such evidence.

### Pot X: recomputed CVA and capital cost

The base management allocation is:

```text
X_t = CVA_core(t) + KVA_core(t) + statutory_unresolved_claim_allowance(t)
```

The core recomputes exposure, future loss and capital on the quarterly grid from remaining contracts and public parameters. No exposure path is a supplied answer. The actual funding reserve in the unhedged base is zero because no cash borrowing results from a reserve charge. The cash-stress path reports the cost of its actual drawn debt separately.

The core's KVA is a management opportunity-cost allocation for standalone funding CCR plus CVA capital; it is not a statutory provision or a dividend that the cash ledger silently pays. No shareholder distribution is specified. The whole-book max-constraint KVA is separately reported and must not be added to the standalone component KVA.

The base `incurred_CVA` used for regulatory exposure deduction is explicitly zero. An economic CVA number in this management pot is not evidence of an incurred accounting write-down on the regulatory unit. Values calculated under an obsolete automatic economic-CVA deduction are not the reference anchors for this bundle.

### Statutory bridge

For the unhedged base:

```text
E_statutory = E_management - G + D + management_only_X
```

The statutory claim allowance stays in both presentations; it is not restored and deducted again. A separately recognised hedge needs its own unit-specific bridge. Every reserve-schedule row supplies both equities and the bridge components.

This presentation makes gross `D` and `X` useful management schedules without pretending that IFRS requires an identical pair of statutory reserve accounts.

## 4. Day-one recognition is tested on the valid unit

The actual base is a retained-asset financing, not a newly recognised free-standing USD41m derivative. It has no recognisable funding-opportunity profit on day one. The separately consumed USD1m service is a real expense, so **total day-one statutory P&L is minus USD1m**, even though the report field `day1_statutory_profit` is zero for the funding-opportunity gain. `recognition.total_statutory_day1_pnl` and `reporting_panel.day1_statutory_profit` expose that distinction.

The supplied internal add-on is not observable evidence for IFRS recognition. A fictional executable quote used for the experiment's replacement-value mechanics does not by itself establish the evidence required by IFRS 9 paragraph B5.1.2A. That paragraph tests whether the initial fair value is evidenced by an active-market quoted price for an identical asset/liability or a valuation technique using only observable market data. Otherwise the initial difference is deferred, with subsequent release only for changes in factors, including time, that market participants would take into account.[IFRS9]

An explicit **alternative standalone-unit example**, not an additional asset in the failed-sale book, shows why unit definition matters:

| Valid separate derivative unit, USDm | Unobservable case | Evidenced observable case |
|---|---:|---:|
| Clean value | 41 | 41 |
| Statutory exit-price XVA | 12 | 12 |
| Fair value after statutory XVA | 29 | 29 |
| Transaction price | 0 | 0 |
| Initial carrying amount after day-one treatment | 0 | 29 |
| Deferred initial difference | 29 | 0 |
| Day-one P&L | 0 | 29 |

Deferring the gross 41 and immediately charging 12 would produce a spurious loss of 12 in the unobservable case. The difference to which the rule applies is the fair value of the valid unit after its statutory valuation adjustments, minus its transaction price. Internal KVA cannot be inserted into that statutory adjustment without a separate accounting basis.

## 5. Scenarios that distinguish the pots

### 5.1 Base one-year roll

`management_scenarios.one_year_roll` contains two deliberately different views:

- `whole_book`: the first four quarters after paid day-one fees, including EM income, legacy interest and interest on accumulated cash. Its opening and closing equity are actual management stock snapshots.
- `funding_component`: USD3m of first-year net funding cash, the funding mark roll and the two management pot releases; it excludes EM, legacy and cash-account carry. It is not an alternative full-bank profit number.

In the funding component, falling `G` contributes negatively to carry while declining `D` contributes positively to deferral release. Since the frozen base `G` and `D` coincide, those two stock movements cancel in total. The actual funding coupon and XVA release are not counted a second time. The whole-book view adds only the other contracts' recorded result.

### 5.2 The divergent-pot shock selected for the accounting report

At book year one, first roll the base to nineteen years remaining. Then, with **zero elapsed time**, change the matching funding replacement spread by +100bp and provider CDS from 80bp to 160bp. Reference credit and the locked financing coupon are unchanged. This is `management_scenarios.instant_funding_100bp_provider_80bp`.

| Quantity | After unchanged-market roll | Immediately after joint shock |
|---|---|---|
| Remaining years | 19 | 19 |
| `G`, USDm, rounded | 39.792 | 172.434 |
| `D`, USDm, rounded | 39.792 | 39.792 |
| `X` | `core.funding.year1_roll.CVA + .KVA` | `core.funding.year1_stress.CVA + .KVA` |
| Clean MTM, USDm, rounded | — | +132.641 |
| Deferral release and reserve theta | — | 0 |
| Reserve market P&L | — | `X_roll - X_stress` |
| Cash movement / new debt | — | 0 / 0 |
| Total management P&L | — | `delta(G) - delta(X)` |

The full-precision XVA levels and independent P&L are generated by the public core and ledger, not a frozen handwritten capital approximation. `D_open`, `D_close`, `X_open`, `X_close` and the attribution in `accounting.reserves` select **this instantaneous panel**, not a mixture with the year's carry. The alternate annual panel remains available. The AVA and terminal fields in that report have different, explicitly labelled observation scopes below.

The independent journal records a funding-asset revaluation and an XVA deduction remeasurement; it contains no cash or debt entry. Closing equity is read from those closing account balances and compared with the attribution. It is not defined as opening equity plus the reported P&L.

### 5.3 Charge-only control

`management_scenarios.charge_only` holds `G`, `D`, contractual payments, cash and debt fixed and increases `X` to the stressed level. Its journal records only the deduction and expense. Equity falls by the XVA charge; cash and debt are unchanged. This is a direct counterexample to “reserve expense mechanically creates borrowing.” It remains an expense despite being noncash.

### 5.4 Actually executed hedge defence

`management_scenarios.hedge_defence_one_year` isolates an alternative executed CDS defence, funded from a named allocation of the existing post-service cash. It is **not** added to the unhedged base and does not create another cash contribution.

The contract buys USD320m protection on P, at an 80bp premium, from independent H for five years beginning on execution at business day two. It pays a USD250,000 setup fee and USD25,000 annual administration cost. The hedge quote is observed **one year after CDS execution**, not exactly one year after base-book inception; `quote_book_time` is execution time plus one, and remaining CDS term is four years. Four full quarterly premiums total USD2.560m; setup plus first-year administration costs total USD0.275m.

The quoted clean mark changes from zero to:

```text
H_1 = 320m * (160bp - 80bp) * A(4)
hedge_pnl = H_1 - 2.560m
```

This annuity conversion is the matched fictional replacement quote, not a universal CDS-pricing model. OIS interest on allocated cash is separate carry. Actual premium and service cash costs are paid; the independent hedge journal computes the result from closing stocks. The executed cash-VM CSA gives cash equal to the positive mark and an equal return liability. Receipt of VM is not a second gain and does not extinguish the hedge asset.

This panel proves hedge cash, mark and P&L mechanics only. Eligibility, designation, hedge-provider CVA/CCR, fees, expiry and unprotected tail are evaluated in `core.actions.CDS.normal_designated`, `stress_designated` and `stress_undesignated`, as described in [capital-actions.md](capital-actions.md). The isolated hedge mark is not advertised as the all-in economic benefit, and the five-year hedge is not silently renewed for the twenty-year funding exposure.

## 6. Actual cash funding versus conditional quadratic UFVA

### 6.1 Mark-only move: no actual borrowing

`cash_scenarios.mark_only_funding_widening_100bp` widens the marking curve by 100bp and uses the stated stressed own-bond spread, while reference credit stays unchanged. Neither input changes the locked TRS financing coupon. Without an actual deficit, the facility is undrawn and the cash spread cost is zero. Charging CVA or KVA does not change that conclusion.

The actual cost definition is:

```text
PV_actual_funding_spread_cost = sum_i discount(t_i) * debt_(i-1) * own_bond_spread * q
PV_actual_borrowing_interest = sum_i discount(t_i) * debt_(i-1) * (OIS + own_bond_spread) * q
```

`cash_funding_drag` in statutory and management P&L is the full actual interest expense. The spread-only discounted cost is separately exposed as actual UFVA, so OIS interest is not silently omitted from the P&L tie. The own bond spread relative to OIS determines the contractual funding expense. Bond-minus-own-CDS basis is a comparative benchmark; it is not a cash borrowing rate by itself.

### 6.2 Reference-credit settlement stress: actual debt

`cash_scenarios.actual_reference_credit_widening_250bp` applies the +250bp reference-credit move at the first scheduled quarterly settlement and holds it thereafter. Funding marking spread is +100bp and the actual own-bond spread is 150bp. The quarter-end U price follows the published reference pricing rule with 19.75 years remaining. At subsequent quarters the price evolves with tenor, so the whole cash path, repayment and funding cost are recomputed rather than multiplying the first draw by twenty years.

The independently prepared W3 arithmetic exhibit supplies the following first-event anchors; it is not a claim that the complete runtime gate has already been run:

| First quarterly event | Amount |
|---|---:|
| U price per unit | 0.7434976916757938623 |
| Price-return cash component | -USD256,502,308.324206 |
| Reference coupon less financing coupon | +USD750,000 |
| Net TRS payment due | USD255,752,308.324206 |
| Cash available before the TRS settlement | USD109,990,000 |
| Actual facility draw | USD145,762,308.324206 |

The source is the original flat-bundle snapshot [branch-arithmetic-evidence.json](../evidence/baseline/branch-arithmetic-evidence.json), with scenario facts in the public W3 exhibits. The available cash includes the post-fee USD99m, its quarterly OIS interest and EM-minus-legacy coupon cash. The W3 funding opportunity can rise while the full pre-payment contractual value and cash position worsen. That is why a positive funding mark is not liquidity.

The additional W3 deployment market markdown and provider-credit shock belong to their own market and credit calculations; the contractual deployment cash engine does not turn an unrealised EM markdown into a cash principal loss. No hidden equity issuance rescues the resulting statutory loss. Independent quarter-by-quarter results disclose the minimum equity and the explicit distinction between cash feasibility, capital approval and solvency.

### 6.3 Separate executed exposure-prefunding experiment

`conditional_facility` is a distinct opt-in transaction with its own contributed USD100m cash/equity. It is not another facility or cash pot inside the base book. It borrows and segregates an external OIS deposit according to:

```text
x = incremental funding-spread move
B_i = N * A(20-t_i) * max(x, 0)
own_spread(x) = 0.005 + x
quarter_i_spread_cash_cost = B_i * own_spread(x) * q
```

Primary borrowing proceeds go into the segregated deposit; they cannot also fund the base deployment or serve as reusable base VM. The deposit earns OIS while the executed liability costs OIS plus the own-bond spread. Only the net spread is an incremental primary funding cost. As the target falls, deposit principal pays down matched primary debt. Its interest cash cost is paid from the separately contributed buffer; only an actual buffer deficit could use the separately disclosed secondary facility.

For the fixed quarterly grid and stated no-default path:

```text
I = N * sum_(i=0 to 79) q * discount(t_(i+1)) * A(20-t_i)
  = 119,725,575,814.7565 USD-year per unit spread, rounded
UFVA_primary(x) = I * x * (0.005 + x), for x >= 0
UFVA_primary(0.01) = USD17,958,836.372, rounded
initial_deposit = N * A(20) * 0.01 = USD137,220,515.265, rounded
```

Thus the quadratic term is real **conditional on this executed cash requirement and its own-funding rule**. It is not a general consequence of credit exposure, a positive replacement value or a reserve charge. USD14.5m is not the result under these quarterly inputs; obtaining it would require changing the convention or funding assumptions.

The funding-cost reserve is the remaining PV of those identified future cash spread costs:

```text
R_80 = 0
R_i = discount(q) * (quarter_i_spread_cash_cost + R_(i+1))
```

At each quarter the journal records the actual interest expense and reserve consumption/time roll separately. Reported P&L is OIS income on the independent buffer plus the reserve reduction minus actual net funding expense. The release therefore does not manufacture new cash or erase the funding bill. Principal and deposit movements have no P&L. The independent ledger exposes both primary and secondary funding costs and does not identify the reserve balance itself with a second debt draw.

## 7. AVA: uncertainty, CET1 and the conditional capital-cost link

The complete applicability and source analysis is in [valuation-capital-annex.md](valuation-capital-annex.md). The implementation selects the identified EU-style **core approach, model-risk category, Method 1** for two already recognised positive fair-value asset exposures. Their CET1 sensitivity is 100%. Other AVA categories remain unchanged in this incremental comparison, not assumed absent from the bank.

For each supplied uniform range of plausible current exit-cost model estimates `[a,b]`:

```text
expected_cost = (a+b)/2
q90_cost = a + 0.90*(b-a)
accounting_FV = gross_value - expected_cost - same_source_accounting_prudence
prudent_value = gross_value - q90_cost
residual_uncertainty = max(accounting_FV - prudent_value, 0)
APVA = (1-0.50) * residual_uncertainty * CET1_sensitivity
```

The 90% criterion is about a prudent **current exit valuation**, not future-P&L VaR. Uniformity is a disclosed synthetic assumption, not a regulatory prescription. Existing accounting prudence offsets only the same identified source of uncertainty. Expected cost is not by itself an uncertainty cushion. Method 1's aggregation alpha is 50% under the identified post-2020 rule; the expired 66% factor is not used.[RTS]

| USDm | X1 | X2 |
|---|---:|---:|
| Gross value | 100.000 | 50.000 |
| Plausible cost range | [1.750, 4.250] | [1.375, 2.625] |
| Expected cost | 3.000 | 2.000 |
| Same-source accounting prudence | 0.200 | 0.100 |
| Accounting fair value | 96.800 | 47.900 |
| 90th-percentile cost | 4.000 | 2.500 |
| Prudent value | 96.000 | 47.500 |
| Residual uncertainty | 0.800 | 0.400 |
| Method 1 APVA | 0.400 | 0.200 |

The incremental model-risk AVA is **USD0.600m**. Under the controlled fixed USD100m RWA denominator, CET1 falls from USD12.000m to USD11.400m, or 12.0% to 11.4%. CRR Article 34 supplies the CET1 deduction; it is neither a P&L charge nor cash borrowing.[CRR] The actual retained-U book is not assigned this separate deduction again: `fullbook_capital.AVA_in_actual_base_book` is zero.

If management elects to restore a 12% target in this controlled example, the incremental equity required is USD0.600m. No issuance is actually booked. With an explicit constant twenty-year forecast for this component and a 10% equity hurdle, the disclosed opportunity-cost increment is:

```text
replacement_equity_KVA = 0.600m * 0.10 * A(20) = USD0.823323092m, rounded
```

This is a conditional future capital-cost allocation, not automatic cash, not an accounting reserve and not a universal KVA response to any AVA increase. At a slack target no replacement capital need follow; a different forecast or binding constraint changes the result. Real exposure/RWA treatment is not assumed to retain the controlled fixed denominator.

Two counterexamples make the uncertainty distinction testable:

- Shift X1's entire cost range upward by USD3m without changing width or same-source prudence. Expected accounting cost rises from USD3m to USD6m, but its residual uncertainty remains USD0.8m and APVA remains USD0.4m. The accounting loss still matters to equity; the AVA does not mechanically grow with reserve level.
- Widen X1's range symmetrically from `[1.750,4.250]` to `[0.500,5.500]` while its expected cost remains USD3m. Its q90 cost becomes USD5m; residual uncertainty is USD1.8m and APVA USD0.9m. With X2 unchanged, the incremental AVA becomes USD1.1m even though the expected-cost level did not rise.

The sources identified as current are CRR consolidated **26 June 2026** and RTS consolidated **26 June 2020**, including the applicable amendment. A consultation or the July 2026 draft-RTS submission deadline is not treated as enacted replacement law. Governance, category coverage and actual recognition must be evidenced before using the controlled result in a real capital return.

## 8. Settlement, not age, closes the final reserve

At contractual year twenty the funding mark, priced-time deferral and scheduled core CVA/KVA profile run off. That does **not** settle an unpaid provider amount.

The base terminal event withholds USD500,000 from the otherwise positive final provider settlement. It records a gross receivable of USD500,000 and a full USD500,000 allowance. The net claim is zero, but the claim and its uncertainty still exist. Thus:

```text
scheduled G_20 = 0
scheduled D_20 = 0
scheduled core CVA_20 = KVA_20 = 0
unresolved gross claim = 500,000
terminal X = claim allowance = 500,000
all_obligations_settled = false
```

The implementation publishes two subsequent **alternative** resolutions, neither booked into the base path:

1. Actual payment of USD500,000: debit cash, credit the receivable, and reverse the now-unneeded allowance into P&L. The reserve becomes zero because the claim was settled.
2. Evidenced final write-off/legal discharge: use the allowance against the gross claim. With a full allowance there is no second loss or cash receipt. The reserve becomes zero because the exposure was extinguished.

A zero terminal reserve obtained solely by setting remaining maturity to zero is wrong. Settlement evidence must also cover any unmade payable and actual deficit debt. The core's final scheduled row is not a certificate that every legal and cash obligation has closed.

## 9. Public calculation and report mapping

### APIs

```python
compute_reserves(parameters: dict, core: dict) -> dict
simulate_cash(parameters: dict, reference_credit_move: Decimal,
              funding_move: Decimal, own_spread: Decimal) -> dict
```

`core` must be the public `compute_core(parameters)` result. The reserve module consumes `funding.normal`, `funding.year1_roll`, `funding.year1_stress`, `actions.CDS.hedge_mark_year1` and the public `wholebook_capital`, `annuity`, `discount` and `reference_price` functions. Missing required fields raise `ReferenceError`; there is no local fallback capital formula or implicit exposure series. Public money and rate values are exact decimal strings for JSON interchange.

The standalone command prints JSON and writes no report files. Run it from R, the organized bundle root containing `requirements.txt`:

```text
python3 src/reserve_ledger.py --parameters fixtures/numerical-inputs.json
python3 src/reserve_ledger.py --parameters fixtures/numerical-inputs.json --summary
```

From another working directory, use `python3 "$R/src/reserve_ledger.py" --parameters "$R/fixtures/numerical-inputs.json" --summary`, with R set to the absolute bundle path. The reserve CLI does not accept `--root`.

The full form includes all eighty cash quarters, eighty-one reserve observation rows including inception and terminal maturity, independent journals, the conditional facility and separate valuation/capital examples. The summary selects anchors without changing calculations. Command examples document the interface; execution evidence belongs to the integration gate, not to this document's existence.

### Stable exported locations

| Public field | Observation and source |
|---|---|
| `accounting.recognition` | Retained-asset financing at inception; `day_one_accounting`; zero funding-opportunity recognition before the separately expensed service |
| `recognition.total_statutory_day1_pnl` | Actual day-zero statutory stock change including the consumed service |
| `accounting.reserves.D_open`, `.D_close`, `.X_open`, `.X_close` | Instantaneous year-one funding/provider shock after the base roll |
| `accounting.reserves` attribution and independent equity change | Same instantaneous panel; no year's cash or theta added |
| `accounting.reserves.AVA`, `.CET1_after_AVA` | Separate recognised-FV model-risk example from `reserve_uncertainty`; not that instant panel's equity |
| `accounting.reserves.terminal_reserve`, `.all_obligations_settled` | Year-twenty unpaid claim event from `terminal` |
| `reserve_schedule` | Base funding opportunity, D, X and statutory bridge at every quarter, plus the final claim allowance |
| `management_journal` | Opening and closing independent trial balances, management overlay entries and an explicit link to the base cash source journal |
| `management_scenarios.one_year_roll` | Whole-book and funding-only annual attribution, kept separate |
| `management_scenarios.charge_only` | Same year-one stocks, only X changes; no cash or debt |
| `management_scenarios.hedge_defence_one_year` | Executed CDS countercase, one year after its execution, not base-book year one |
| `cash_scenarios.base` | Unchanged reference and funding marks; own-bond spread 50bp |
| `cash_scenarios.mark_only_funding_widening_100bp` | Funding move +100bp; reference move zero; own spread 150bp; locked cash coupon unchanged |
| `cash_scenarios.actual_reference_credit_widening_250bp` | Reference +250bp at the first quarterly price settlement; funding move +100bp; own spread 150bp |
| `credit_xva.actual_ufva` | Discounted actual drawn-debt spread cost in that reference-credit stress |
| `credit_xva.base_actual_ufva`, `.mark_only_ufva` | Separate cash-path costs, not replaced by the conditional experiment |
| `credit_xva.conditional_quadratic_ufva` | Separate executed exposure-prefunding transaction and its explicit own-spread rule |
| `fullbook_capital` | Each beginning-of-quarter actual base stock passed to the core risk/leverage max comparator; discounted capital service over the full life |
| `terminal` | Scheduled maturity versus open claim, plus unbooked subsequent settlement/write-off alternatives |

The scenario parameter sources are `scenarios.normal` and `scenarios.joint_stress` for the funding/provider profiles; `cash_ledger.reference_stress_move`, `cash_ledger.funding_stress_move`, `own_bond_spread` and `stressed_own_bond_spread` for cash; `conditional_reserve_facility` for the separate prefunding transaction; `actions.CDS` and its matched replacement quote for hedge defence; `reserve_uncertainty` for AVA; and `terminal` for the residual claim. `scenarios` is an ID-indexed conceptual table, stored as an array: consumers resolve entries by `id`, not by assuming array order.

The binding exhibit/report pointer map is owned by the public schema bundle. This module provides exact named outputs; it does not silently select a different observation for each field to make an identity pass.

## 10. Acceptance evidence and rejected shortcuts

The integration owner runs the standalone calculators and report grader after artifact owners finish. Workers do not claim validation from source inspection. The significant gates are:

- Recompute all cash distributions, principal paths and actual facility draws from public terms over the full horizon; reconcile opening and closing stocks without assuming a settlement because time elapsed.
- Replay debit/credit journal entries independently and compare the resulting equity change with the separate attribution for every quarter and named scenario.
- Check that base D follows its frozen annuity, that annual expected release appears once, and that an instantaneous market shock has zero theta but nonzero X remeasurement.
- Hold all contractual cash obligations fixed while charging X. Debt and cash must remain unchanged, even though P&L and equity fall.
- Recompute the reference-loss cash stress and its accumulated debt spread cost; then compute the separate prefunding experiment under its own legal cash requirement. The two UFVA values are not interchangeable.
- Check recognition on the retained-risk unit and on both controlled standalone-unit observability cases; do not impose gross deferral plus a duplicated statutory XVA loss.
- Recompute the current-valuation uncertainty, same-source prudence, Method 1 aggregation and CET1 numerator; do not deduct management KVA as if it were an AVA.
- Preserve the USD500,000 final allowance until actual payment or evidenced extinguishment. A candidate that reports zero because the core's scheduled exposure ends must fail.

A candidate P&L that is arithmetically consistent with its own invented closing equity is not evidence. The grader must compare it with the authoritative independent journal and opening/closing stocks. Likewise, a true `expected_release_counted_twice=false` flag cannot rescue a numerical attribution that counts the same release twice.

## Official references

- **[IFRS9]** [IFRS 9, issued 2026 text](https://www.ifrs.org/content/dam/ifrs/publications/html-standards/english/2026/issued/ifrs9.html), paragraphs 3.2.15, 3.2.22, B3.2.14, B3.2.16(o), 5.1.1A and B5.1.2A. Retrieved 19 September 2026. Retained-risk transfer accounting and initial difference recognition are separate tests.
- **[RTS]** [Delegated Regulation (EU) 2016/101, consolidated 26 June 2020](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX%3A02016R0101-20200626), Articles 8, 11, 12, 13 and Annex; with [Regulation 2020/866](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32020R0866). Retrieved 19 September 2026. The relevant 50% post-2020 Method 1 alpha and 90% current prudent-value criterion are applied only within their stated scope.
- **[CRR]** [Regulation (EU) 575/2013, consolidated 26 June 2026](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX%3A02013R0575-20260626), Articles 34, 92, 105 and 111. Retrieved 19 September 2026. The CET1 deduction and real exposure/denominator rules do not make the controlled fixed-RWA example a compliance calculation.