Original model 01 analysis. These figures retain the original calculation basis. Read the new expert-built model and like-for-like comparison →
DIRECTIONAL - no registered model | 8 September 2026 | Codex / Marvin
Recommendation
Change the capital and operating proposition before optimising debt. Prioritise an operator-funded/ground-lease option and a genuinely standalone, lower-capital first phase. Compare a residential-income alternative only with explicit Portion 4 planning and infrastructure dependencies. No alternative is yet demonstrated to be better or bankable: operator terms, measured schemes and separate cost plans are missing. Mo and the partners retain the investment and release decisions.
The tested 40-key hotel and 1,000 m² office GLA is a rejection test of one hypothesis, not the recommended development mix or a valuation of the whole landholding. It currently produces R3.379m annual net operating income against R163.588m pre-finance development cost excluding land and VAT, a 2.07% yield.
Ranked work priorities
- Who funds and operates each use? Test land-retaining ground leases, operator-funded buildings and carefully priced joint ventures. This changes both the owner's capital requirement and the income/risk retained. An operator contribution is not free capital: compare lease duration, escalation, guarantees, servicing obligations, handback, control and the value of income surrendered. Do not compare ground rent divided by zero land cost with an own-build development yield.
- Prove the hotel's demand and margin. Occupancy of 60% instead of 45%, at the same R1,500 net average daily rate, adds R2.405m annual NOI and raises project yield to 3.54%. R1,800 ADR at 45% adds R1.443m and raises yield to 2.95%. Both together give R7.707m NOI and 4.71% yield, not the sum of two isolated deltas. Obtain monthly operator forecasts supported by comparable achieved trading, seasonality, staffing, distribution costs, management fees and replacement reserves. Higher ADR and occupancy may trade off, not improve together.
- Reduce capital per occupied room and per rentable square metre. A 25% hotel cost-per-key cut, R1.8m to R1.35m, saves R26.727m after linked fees, contingency and escalation but only raises yield to 2.47%. Even a 30% reduction in the whole pre-finance budget gives 2.95% at base income. Ask the architect/operator/QS to test efficient gross-to-net areas, specification, common facilities, services and parking together. A smaller hotel must have its own operating model; cutting keys pro rata while retaining shared infrastructure can worsen returns.
- Choose the right first-phase income mix. Do not assume conventional offices, a boutique hotel or residential is the winner. Office rent rising from R100 to R150/m²/month adds only R0.383m annual NOI at this small office scale, taking total yield to 2.30%. Test smaller professional/medical modules, operator-led accommodation and conditional rental residential against actual demand. Medical use, serviced accommodation and QSR are not automatically permitted or pre-let. Community uses need an explicit funded capital and operating allocation. Any sale component must show the trade against the client's long-term income and land-retention objective.
- Right-size infrastructure and reduce time to income. Halving the unquoted civil, electrical and public-realm allowances saves R10.394m including related costs but only gives 2.21% yield. Six months less preconstruction saves R4.836m and gives 2.13% under the original fixed-income screen. Separate common infrastructure from phase-specific works and obtain engineering and access evidence; do not simply delete allowances or statutory lead time. The full benefit of earlier income needs a dated discounted cash-flow model, not only a stabilised yield.
- Optimise funding after the operating case works. Reducing assumed interest from 11.5% to 9.5% raises the hypothetical 60% loan's DSCR only from 0.246x to 0.275x, still below the provisional 1.30x screen. Cheaper debt changes financing cost and equity cash flow, not the unlevered development yield. More equity may make the cash requirement fundable but does not repair the asset's weak operating return.
The threshold that matters
At the provisional 10% development-yield hurdle, R163.588m cost needs R16.359m annual NOI, versus R3.379m tested. Alternatively the tested NOI supports only R33.789m of development cost on that same pre-finance, land-excluded basis, a 79.35% cut. These are screening thresholds, not a valuation or a feasible cost target.
At the base R1,500 ADR and other inputs, the hotel would need 126% occupancy for the whole mix to meet 10%, which is impossible. At 60% occupancy, it would need about R3,149 net ADR while maintaining the same ancillary, cost and office assumptions. That is a break-even calculation, not a suggested market rate. A more expensive product capable of that ADR may also cost more to build and operate.
The return hurdle itself needs Mo's confirmation. At 8% the current cost needs R13.087m annual NOI; at 6% it needs R9.815m. Lowering the hurdle does not improve the asset. The existing combined upside case, which also assumes higher office rent/occupancy and 10% lower total cost, gives 5.44%; the downside has negative R0.280m NOI. Those are non-probability-weighted stress cases, not forecasts.
Expert review of the first model
The original workflow record shows finance-scenario, not property-development-finance-expert. This revision applies the property-development-finance-expert skill. The review is sequential and non-independent; it is not an independent professional sign-off.
The major comparability issue is timing. The original model escalates construction costs by 6% to the construction midpoint, but leaves rents, ADR and operating costs at their nominal starting assumptions. Its 2.07% is therefore a deliberately conservative mixed-date screening ratio, not a fully consistent nominal feasibility return. A separate constant-September-2026-money diagnostic removes that escalation: R140.196m pre-finance cost and a 2.41% yield. As a separate illustration, escalating both operating revenues and operating expenses uniformly at an unverified 4% for 4.5 years produces R4.031m NOI and 2.46% against the original future cost. Neither diagnostic rescues this tested mix. The next model should use a dated nominal cash-flow basis consistently, or show a distinct constant-money analysis, not silently blend them.
The cost basis needs reconciliation. The R1.8m/key allowance including FF&E implies R30,000 per hotel GFA m² at 60 m²/key, before infrastructure and soft costs. This is a hypothesis calibrated to historical benchmarks, not a measured QS estimate. Check room standard, FF&E and operator requirements against the actual accommodation schedule. Remove fee, diagnostic, working-capital and contingency duplication only where evidenced; reducing contingency is not value engineering.
Allocated component returns are not standalone phases. The workbook allocates common/soft costs largely in proportion to direct hotel and office cost. Those allocated yields cannot establish the price of an office-only or smaller-hotel scheme. Each alternative requires new measured scope, common infrastructure, opening/ramp costs and operator economics. Ranking between unmodelled alternatives is unresolved.
Funding remains a screen. Straight-line construction draw/simple-interest reserve is not a lender draw model. NOI as a cash-flow-available-for-debt-service proxy, VAT recovery/timing, tax, land opportunity cost, ongoing reserve timing and a complete equity waterfall remain unresolved. No IRR, MOIC or investment-grade residual land valuation is established. No subsidy is assumed.
Exact sources and verification
- Live workbook read on 8 September 2026: Inputs!B2:B41; Cost!B2:B17; Operations!B14:B15 and B20:B30; Funding!B2:B6; Sensitivity!B2:H4; TimingLand!A2:D4.
- Hotel demand inputs: Inputs!B22:B27. Hotel and office capital: Inputs!B6:B7; fees/contingency Inputs!B14:B15. Office demand: Inputs!B28:B30. Infrastructure: Inputs!B11:B13. Timing: Inputs!B18:B21. Finance: Inputs!B34:B38.
- Readback values and formulas are saved in live-values.json and live-formulas.json. analyse_levers.py reproduces base cost, NOI, yield, DSCR and the governing occupancy threshold before calculating sensitivities. Full deterministic outputs are in lever-results.json. The source workbook has not been edited.
- Current authority: Haarties finance/model-registry.md remains STUB_ONLY; no approved model. Current project evidence: _status.md and working/2026-09-08-feasibility/review-pack.md. Supporting framework: akoma-wiki/wiki/concepts/haarties-stage-2-option-screening.md. The older wiki registration restriction is superseded by the current registry's permission for labelled directional work.
- Source assumptions remain unresolved, despite verified arithmetic. The live workbook cites the AECOM July 2024 benchmark and one asking-rent comparator, not current QS quotes or achieved trading. No new market research was performed for this lever review.
- Own, Akoma-wide and other-project judgement files were searched for funding/procurement/feasibility principles; no applicable tagged judgement line was found or borrowed.
All scenarios are decision-material. Confirm the brief, hurdle and available equity with Mo; obtain planner, operator, QS, engineer, tax and lender evidence before reliance or external issue.