feat: model structural overhang vs bleed-off; fix net-benefit triviality (#2)

Addresses the core critique that "above-target" is not the same as
"liquidate" — under steady demand you pause replenishment and let stock
bleed off; you only clear inventory that won't self-resolve.

Model:
- New `clearWindowMonths` patience window. Above-target stock that sells
  through within it is "Bleeding Off" (no clearance); only the remainder
  is a "Structural Overhang" and the actionable base for carry/clear.
- Carry-vs-clear and recapture now compute on the overhang, not gross
  excess, so the rule stops trivially firing on everything.
- Recalibrated demand so the slow segments (EV, luxury) carry genuinely
  soft run-rates → a real months-long overhang, telling the EV-glut /
  slow-luxury story for the right reason. Fast pools (Tucson) read
  "Bleeding Off". Every documented figure reproduces from the data.
- New statuses Overhang / Bleeding Off; dashboard gains a bleed-off
  window slider, a months-to-clear reality check, and per-row gating.

Docs (METHODOLOGY.md):
- §3/§5 rewritten for the two-step rule; §7 baseline refreshed.
- New §11 (OEM-owned vs dealer-floorplanned capital) and §12 (the
  liquidation channel ladder: incentive / fleet / auction + residual hit).

Baseline now: $2.80B deployed, $603M above target, of which only $320M is
structural overhang; $58M/yr carry, $19M clearance, $301M recaptured,
$38M net, $216M stockout gap fully covered.

Verified: tsc --noEmit clean; vite build; data reproduces docs exactly.

Co-authored-by: privatejoel <privatejoel@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

Author: privatejoel

METHODOLOGY.md

@@ -22,22 +22,35 @@ the narrative is coherent, not pulled from Hyundai's actuals.
 
 ## 3. Core formulas
 
+The central correction over a naive DSI model: **above-target stock is not automatically a
+liquidation candidate.** Stock that will sell through within a patience window at current
+demand should simply have its replenishment paused — it *bleeds off* at no clearance cost.
+Only the residual still sitting after the window is a **structural overhang** worth a
+clearance haircut. This is what stops the model from recommending you discount demand you
+already have.
+
 ```
 dailySales      = avgMonthlySales × (1 + demandShift) / 30
 DSI             = stock / dailySales                       # days of inventory on hand
 targetStock     = targetDSI[category] × dailySales
-excessUnits     = max(0, stock − targetStock)
-excessCapital   = excessUnits × unitCost
-carryingOnExcess= excessCapital × carryingRate             # annual $ bled on the excess
-liquidationLoss = excessCapital × liquidationDiscount      # one-time clearance haircut
-recaptured      = excessCapital × (1 − liquidationDiscount)
-netFirstYear    = carryingOnExcess − liquidationLoss
+excessUnits     = max(0, stock − targetStock)              # gross above target
+windowUnits     = dailySales × 30 × clearWindowMonths      # what sells within the patience window
+bleedUnits      = min(excessUnits, windowUnits)            # clears organically — no action
+overhangUnits   = max(0, excessUnits − bleedUnits)         # structural overhang — the actionable base
+overhangCapital = overhangUnits × unitCost
+monthsToClear   = excessUnits / (dailySales × 30)          # organic sell-through of the excess
+
+carryOnOverhang = overhangCapital × carryingRate           # annual $ bled on the overhang
+liquidationLoss = overhangCapital × liquidationDiscount    # one-time clearance haircut
+recaptured      = overhangCapital × (1 − liquidationDiscount)
+netBenefit      = carryOnOverhang − liquidationLoss
 portfolioDSI    = Σ stock / Σ dailySales                   # volume-weighted, not a simple average
 ```
 
 **Status rule (segment-specific, not one universal cutoff):**
 
-- `At Risk` — `DSI > targetDSI` (over-stocked relative to the segment's velocity)
+- `Overhang` — `overhangUnits > 0` (above target *and* won't clear within the window → clearance candidate)
+- `Bleeding Off` — above target but `overhangUnits = 0` (self-resolves once replenishment pauses → no clearance)
 - `Stockout Risk` — `DSI < 0.5 × targetDSI` (too lean; turning away serviceable demand)
 - `Healthy` — otherwise
 
@@ -53,45 +66,63 @@ segments do not. A uniform threshold would contradict the thesis.
 | SUV | 60 | Mainstream mid-velocity |
 | Volume | 45 | High turnover; lean target protects ROIC |
 
-## 5. Decision rule
+## 5. Decision rule (two steps)
+
+1. **Pause, don't dump.** For every above-target pool, first stop replenishment and let it
+   bleed off. Anything that clears within `clearWindowMonths` never warrants a clearance
+   haircut — liquidating demand you already have is value-destructive.
+2. **Clear the overhang on net benefit.** For the structural overhang that remains, act
+   **only when `netBenefit > 0`** — the annual carrying cost avoided exceeds the one-time
+   clearance loss. Because the overhang sits *beyond* the window by construction, annual
+   carry is the right horizon to weigh against the one-time haircut.
 
-For each above-target pool, act **only when `netFirstYear > 0`** — i.e. the annual
-carrying cost avoided exceeds the one-time clearance loss. Recaptured capital is then
-redeployed against `Stockout Risk` pools first (relieving lost-sale demand), with the
-remainder funding strategic EV / growth priorities. This resolves the
-redirect-vs-liquidate distinction: transfer where a constrained pool can absorb units,
-clear only where none can.
+Recaptured capital is then redeployed against `Stockout Risk` pools first (relieving
+lost-sale demand), with the remainder funding strategic EV / growth priorities. This is why
+the model clears the IONIQ 5 (soft demand → a months-long overhang) but *not* the Tucson
+(above target, yet sells through in days). It also resolves the redirect-vs-liquidate
+distinction: transfer where a constrained pool can absorb units, clear only where none can.
 
 ## 6. Scenario parameters (defaults & ranges)
 
 | Parameter | Default | Range | Source logic |
 |---|---|---|---|
 | `carryingRate` | 18%/yr | 8–28% | Floorplan interest (~8%) + depreciation (~7%) + storage/insurance (~3%) |
-| `liquidationDiscount` | 6% | 2–15% | Wholesale/auction haircut to clear excess units |
+| `liquidationDiscount` | 6% | 2–15% | Wholesale/auction haircut to clear excess units (see §12 for the channel ladder) |
+| `clearWindowMonths` | 2 mo | 1–6 mo | Patience window: above-target stock that sells through within it bleeds off rather than being cleared |
 | `demandShift` | 0% | −20% to +20% | Demand stress test |
 
 ## 7. Baseline results (default assumptions)
 
 | Metric | Value |
 |---|---|
-| Capital deployed | $2.87B |
-| Capital above target | $412M (14% of deployed) |
-| Annualized carrying cost on excess | $74M/yr |
-| Capital recaptured (net of discount) | $387M |
-| One-time clearance loss | $25M |
-| First-year net benefit | $49M |
+| Capital deployed | $2.80B |
+| Capital above target (gross) | $603M (22% of deployed) |
+| Structural overhang (actionable) | $320M (11% of deployed) |
+| Annualized carrying cost on overhang | $58M/yr |
+| Capital recaptured (net of discount) | $301M |
+| One-time clearance loss | $19M |
+| Net benefit | $38M |
 | Stockout-risk funding gap | $216M (fully covered by recaptured capital) |
-| Portfolio DSI (volume-weighted) | 50 days |
+| Portfolio DSI (volume-weighted) | 52 days |
+| Overhang pools / bleeding-off pools | 3 / 1 |
+
+Note the gap between **$603M above target** and the **$320M actually cleared**: ~$283M of
+above-target stock bleeds off on its own once replenishment pauses, and is never liquidated.
 
 ## 8. Sensitivity (illustrative)
 
-- **Carrying rate is the swing factor.** At 8% the annual bleed falls to ~$33M and several
-  pools no longer clear the net-benefit hurdle; at 28% it rises past ~$115M and clearance is
-  obviously worth it. This is why the rate assumption must be owned by Finance, not assumed.
-- **Clearance discount** mostly affects *how much* is recaptured, not *whether* to act,
-  until it approaches the carrying rate.
-- **Demand shift** moves pools across thresholds — a −10% shock pushes additional volume/SUV
-  pools above target; +10% relieves them. The status logic recomputes live.
+- **The bleed-off window decides how much is even actionable.** This is the parameter that
+  encodes "don't liquidate demand you still have." At a 1-month window more above-target stock
+  counts as overhang; widen it toward 6 months and most pools are judged self-resolving, so the
+  overhang — and the recommended clearance — shrinks toward zero. At the 2-month default, $320M
+  of the $603M above-target is overhang.
+- **Carrying rate** scales the annual bleed on the overhang (≈$26M at 8%, ≈$90M at 28% on the
+  baseline overhang) and, with the discount, sets whether clearing clears the net-benefit hurdle.
+  It must be owned by Finance, not assumed.
+- **Clearance discount** mostly affects *how much* is recaptured, not *whether* to act, until it
+  approaches the carrying rate.
+- **Demand shift** moves pools across thresholds and resizes the overhang — a −10% shock lengthens
+  sell-through (more overhang); +10% lets more stock bleed off. The status logic recomputes live.
 
 ## 9. Known limitations (and the honest next steps)
 
@@ -100,8 +131,9 @@ clear only where none can.
 - **Contribution margin not modeled**, so redeployment upside is stated as the funded demand gap,
   not an incremental-profit figure. *Next step: per-segment margin to value recaptured capital.*
 - **Steady-state demand** — no seasonality, model-year transitions, or incentive elasticity.
-- **Single-period view** — `netFirstYear` mixes a recurring annual saving with a one-time loss;
-  a full NPV would discount multi-year carrying savings. *Next step: multi-period NPV.*
+- **Single-period view** — `netBenefit` weighs one year of carry on the overhang against a one-time
+  loss. The bleed-off window removes the cruder "annual-vs-one-time always clears" bias, but a true
+  model would integrate carry over each pool's expected hold path. *Next step: multi-period NPV.*
 - Figures are illustrative and not Hyundai actuals.
 
 ## 10. Retail & DMS layer
@@ -147,3 +179,39 @@ from the OEM layer, now visible as floorplan cost.
 yet modeled (would quantify the *margin* cost of aging, not just floorplan); dealer-trade
 logistics cost is assumed negligible; and a live DMS integration (vs. illustrative aging
 vectors) is the obvious productionization step.
+
+## 11. Whose capital is it — OEM vs. dealer floorplan
+
+A deliberate simplification worth stating plainly: the OEM layer values the **entire** national
+pipeline at OEM wholesale cost and frames clearing it as recovering OEM capital. In reality,
+once a unit is wholesaled to a franchised dealer it sits on the **dealer's floorplan**, not the
+OEM's balance sheet — which is exactly what the Retail & DMS layer (§10) measures. The two
+layers describe the same metal from two balance sheets:
+
+| Stage | Who owns it | Where the carry lands |
+|---|---|---|
+| In production / in-transit / port / unallocated | **OEM** | OEM working capital |
+| Allocated & wholesaled to a dealer (on the lot) | **Dealer** | Dealer floorplan interest (Retail layer) |
+
+The OEM's *direct* capital-at-risk is therefore the un-wholesaled slice plus its exposure to
+incentives and residual support; the on-lot overhang is borne by dealers and influenced by the
+OEM through **allocation and incentive policy**, not by repossessing units. Figures are stated on
+a single OEM-wholesale-cost basis for clarity, with the dealer view reconciled in §10. *Next step:
+split the pipeline into OEM-owned vs. dealer-floorplanned and carry each at its own rate.*
+
+## 12. How the overhang actually clears (the "6%" unpacked)
+
+`liquidationDiscount` is a single blended haircut; in practice clearance runs down a ladder of
+channels, each with a different cost and side effect — there is no single anonymous buyer:
+
+| Channel | Buyer | Typical haircut | Side effect |
+|---|---|---|---|
+| Retail incentive | Ordinary retail customer | ~2–8% (subvented APR/lease, cash) | Pulls demand forward; brand-equity drag on luxury |
+| Fleet / commercial | Rental & commercial fleets | ~8–15% | Floods the used market later → **depresses residuals** |
+| Dealer trade / wholesale | Other dealers | ~10–20% | Logistics cost; limited absorbing capacity |
+| Auction / remarketing | Dealers & exporters | ~15–30% | Last resort for 90+ day units (see §10 curtailment) |
+
+The default 6% corresponds to a **modest retail-incentive bump** — the cheapest rung. The model
+does not yet price the residual-value contamination from the fleet/auction rungs, which would feed
+back into a higher effective EV carrying rate. *Next step: a channel-mix input that blends the
+haircut and routes the residual hit back into `carryingRate`.*

README.md

@@ -18,23 +18,27 @@ one place while demand goes unserved in another.
 ## The approach
 
 Measure each model's **Days Sales of Inventory (DSI)** against a **segment-specific target**,
-then apply a **net-benefit decision rule**: clear above-target inventory *only when the annual
-carrying cost avoided exceeds the one-time clearance loss*, and redeploy the recaptured capital
-to relieve stockout-risk segments first.
+separate stock that will **bleed off on its own** (just pause replenishment) from a genuine
+**structural overhang**, then clear only that overhang *when the annual carrying cost avoided
+exceeds the one-time clearance loss* — redeploying the recaptured capital to relieve
+stockout-risk segments first. The point of the bleed-off step: never discount demand you
+already have.
 
 ## Key results (illustrative baseline)
 
 | Metric | Value |
 |---|---|
-| Capital deployed | **$2.87B** |
-| Capital above segment target | **$412M** (14%) |
-| Annualized carrying cost bled | **$74M/yr** |
-| Capital recaptured | **$387M** |
-| First-year net benefit | **$49M** |
+| Capital deployed | **$2.80B** |
+| Capital above segment target (gross) | **$603M** (22%) |
+| Structural overhang (won't bleed off) | **$320M** (11%) |
+| Annualized carrying cost on overhang | **$58M/yr** |
+| Capital recaptured | **$301M** |
+| Net benefit | **$38M** |
 | Demand-constrained gap funded | **$216M**, fully covered |
 
-The recaptured capital more than covers the volume shortfall — the EV-glut and slow-luxury pools
-fund the lean volume line, with ~$171M left for strategic priorities. Full derivation in
+Most above-target stock isn't cleared — it *bleeds off* once replenishment pauses; only the **$320M
+structural overhang** (EV-glut + slow-luxury) is liquidated. The recaptured capital more than covers
+the volume shortfall, with **~$85M** left for strategic priorities. Full derivation in
 [METHODOLOGY.md](./METHODOLOGY.md).
 
 ## How it's structured
@@ -43,9 +47,10 @@ The app is organized as four role-targeted views:
 
 1. **Strategy Brief** — answer-first recommendation with quantified impact, then a
    Situation / Complication / Resolution case. *(Consulting framing.)*
-2. **Decision Dashboard** — segment-targeted DSI table, corrected and accurately-labeled KPIs,
-   a ranked net-benefit recommendation engine, and **live scenario sliders** (carrying rate,
-   clearance discount, demand shift) so every figure can be stress-tested. *(Planning / analytics.)*
+2. **Decision Dashboard** — segment-targeted DSI table that separates a **structural overhang**
+   from above-target stock that simply **bleeds off**, accurately-labeled KPIs, a ranked
+   net-benefit recommendation engine, and **live scenario sliders** (carrying rate, clearance
+   discount, bleed-off window, demand shift) so every figure can be stress-tested. *(Planning / analytics.)*
 3. **Execution Plan** — phased rollout, RACI decision rights, governance KPIs, and a
    risk/mitigation register. *(Program management.)*
 4. **Retail & DMS** — drills the OEM thesis down to the dealer P&L: DMS-sourced aging buckets,
@@ -62,6 +67,8 @@ This was deliberately built to survive a probing interview:
   and a volume-weighted "Portfolio DSI," not a count ratio mislabeled as a financial return.
 - **The headline tradeoff is actually computed** — carrying cost avoided vs. clearance loss, per
   pool, with an explicit act/hold rule — not asserted in prose.
+- **Above-target ≠ liquidate** — the model strips out inventory that sells through on its own and
+  only clears the structural overhang, so it never recommends discounting demand you already have.
 - **Targets are segment-specific**, consistent with the core thesis.
 - **Assumptions are owned and sourced**, with sensitivity and limitations documented openly.