# Experimental YLT emission laboratory

Research only, 22 October 2026. No final tokenomics decision. Entry point:
[`site/experiments/index.html`](../site/experiments/index.html). The landing page
adds one navigation link; `simulator.py`, assumptions, historical files, investor
cabinet and its engine are unchanged. Open the experiment HTML directly or serve
`site/` with any static server. No build, CDN chart dependency or network needed.

## Inputs reviewed and provenance

Reviewed current README, `assumptions.yaml`, `simulator.py`, browser treasury
engine/export, data sources/validation, August audit (outdated), October simulation
report, decision register and emission allocation discussion, and existing tests.
The live landing page at https://yieltop.com returned HTTP 200 and matched the
repository landing before the added link. The old audit's “not integrated” claims
are superseded by current code. Working documents disagree: older README refers
to 48M mining, while 12 October allocation and browser engine use 55M. Experiment
default is 55M, editable, with a separate external seed of 5M YLT/$500k. This does
not ratify either allocation. Other team/DAO/airdrop/early buckets are excluded.
`NAV/YLT` divides by experimental minted supply plus seed, **not total 100M FDV**.
Remaining supply means remaining LP mining budget only.

The experiment calls the **existing** `YieltopEngine.simulate` over the checked-in
`site/app/data.js`, not invented historical data. It retains BTC/ETH prices,
Curve/Convex/Stake DAO and Pendle pools, Yield Basis yields, CRV/sdCRV/SDT/CVX/
cvxCRV/YB prices, veYB/vlSDT yields, boost, acquisition costs and CRV routing in
that historical calibration run. The UI displays its actual date window, export
date, BTC/ETH source, primary/secondary totals and terminal treasury NAV.
Current export BTC/ETH source is `yield_basis_file`; this is explicit inherited
fallback metadata, not a fresh Yahoo/Binance fetch. Existing CSV simulation
results are references, not falsely relabeled future observations. The regression
suite checks the browser treasury engine against Python.

For each scenario TVL and strategy, re-run that historical engine. Primary monthly
rate = mean(income / portfolio TVL); secondary monthly yield = sum(secondary) /
sum(treasury NAV). Thus changing TVL recalibrates historical stable dilution and
execution effects. All projection months are **synthetic**, use those average
rates, and do not repeat or splice historical dates/prices. Future treasury is an
aggregate balance, not a future per-asset replay. No future BTC/ETH capital gains
are implied. Growth/decline changes external TVL, not a claimed strategy return.

Underlying limitations remain: Yield Basis V3 reconstruction before launch;
pre-listing YB stand-in and pre-history median locker yields; Curve survivorship;
Pendle issuer concentration; fixed locked SDT; assumed execution depths; no Yield
Basis capacity dilution. See [data sources](data-sources.md) and
[validation](data-validation.md). This is especially material at 500M TVL.

## Emission formulas

Monthly zero-based t; S = mining supply; N = selected horizon × 12; q =
(1 − annual decay)^(1/12). Scheduled budget b_t = S(1−q)q^t/(1−q^N), or S/N
for zero decay. All emissions are capped by unspent supply R_t.

- A: scheduled declining quantity b_t, independent of revenue and market price.
- B: primary revenue / predetermined reference USD per YLT. This is an explicitly
  unsafe lifetime comparator; **never** revenue divided by endogenous spot.
- C: E_t = min(R_t, carry_t + b_t, primary_t / reference_t, cap_t).
- D: same C bounds, with revenue demand multiplied by g_t.

For protected duration L (120/180/240 months), cap_t = R_t/(L−t) while t<L;
afterwards cap_t=R_t. carry_(t+1)=carry_t+b_t−E_t. Carry is an authorization
budget, not cash or additional supply. It cannot bypass the cap. There is no
forced catch-up on a price crash. Reference policy is fixed or an **ex ante**
annual index; neither policy reads spot, NAV or market depth.

**Lifetime proof:** R_(t+1) ≥ R_t (L−t−1)/(L−t). By induction R_t ≥
S(L−t)/L for t≤L. Therefore positive budget remains before the last protected
month, even with unbounded revenue. Exhaustion may occur at the end of month L,
never earlier. This protects timing of mining budget, not LP reward value. Zero
revenue may leave an indefinite unused budget. If the displayed horizon is
shorter than L, no claim is made about unseen exhaustion. Null means “not
exhausted within window”, not infinity; numerical depletion tolerance is 1e−7 YLT.

D exposes every component:

```
lockScore = clip(opening locked YLT / opening issued YLT / 0.50, 0, 1)
yieldScore = clip(calibrated secondary monthly yield × yieldScale / (0.10/12), 0, 1)
liquidityScore = clip(opening AMM USD depth / TVL / 0.02, 0, 1)
g_t = 0.25 + 0.25 × (lockScore + yieldScore + liquidityScore)
```

This is a transparent bounded heuristic, **not an optimized controller**.
Persistent assumed locking is a proxy for veYLT sustainability, not empirical
retention evidence. Thresholds and equal weights are experimental assumptions.
Opening depth is twice USD reserve; no inverse-price term. Lower spot alone
cannot increase the controller's target. Other simultaneous state changes can
change issuance; this is not a blanket monotonicity claim for coupled paths.
At large revenue C/D can coincide because the same schedule or cap binds.

## Economic accounting and metrics

Primary yield is foregone by LP and invested once in treasury, never also paid
as LP cash. Secondary is earned on opening treasury NAV, not same-month purchases.
Months 1–6 reinvest all secondary; later veYLT gets
`1 − 0.9 sqrt(1 − opening lock share)` if locks exist, otherwise zero. The rest
funds YLT buybacks. At zero locks all distributable secondary goes to buybacks.
No secondary term enters issuance demand. Ledger identities:

```
secondary = reinvestment + veYLT cash + YLT buyback cash
closing NAV = opening NAV × (1 + annual revaluation)^(1/12) + primary + reinvestment
closing mining budget = opening mining budget − emission
```

Future acquisition costs are not separately applied: historical calibration
reflects execution in observed treasury yields, but future dollar contributions
are added at par. No redemption promise attaches to treasury value.

One fee-free constant-product AMM has YLT reserve x and USD reserve y. Buyback B
removes xB/(y+B) YLT into protocol reserve. External scenario demand is separate
funding and removes tokens into external holders. LP sells only
`emission × (1−lock ratio) × sell ratio`, receiving yQ/(x+Q). Average execution
versus pre-sale spot defines price impact. Depth is `2y`, not order-book executable
depth. Stress = sold quantity valued at pre-sale spot / pre-sale USD reserve.
No bought token is burned, re-emitted, or paid to LP a second time. Unsold YLT and
locked positions have no assumed cash realizability. AMM seeds are external
funding, not revenue. Crash withdraws equal fractions of both reserves; withdrawn
assets belong to the external seed provider, not an LP reward or treasury cost.

Realizable APY is a **simple annualized monthly cash rate**, not a compounded
forecast: `(LP sale cash + veYLT cash) / TVL ×12`. Cash payouts cover the aggregate
mining cohort. Direct opportunity cost is primary cash foregone minus realized
cash; no compounding, tax, direct strategy execution cost or terminal sale of held
tokens. veYLT APY = monthly ve cash / closing locked-token market value ×12.
This ratio can surge as price collapses without any additional cash. NAV/YLT
and ve APY are diagnostics, not price floors or redemption rights. Monthly and
annual emission charts, other metrics, comparison and downloadable CSV are
computed from the same rows. CSV records synthetic status and scenario inputs.

## Scenarios, reproducibility and reviewed results

TVLs 10/50/100/300/500M; strategies conservative/balanced/custom; horizons
10/15/20/25 years; minimum C/D lifetime 10/15/20 years. Growth is +15% TVL/year,
decline −15%/year, crash is starting TVL divided by ten from month 61, with 50%
AMM reserve withdrawal by default (allowed fraction 0 ≤ w < 1; a fully empty
market is undefined and rejected). Sale/lock ratios, reference policy, decay,
supply, seed liquidity, treasury revaluation, external demand and yield multiplier
are editable. Lock ratio applies to new emissions and locks never expire.

```
python -m pytest -q
node tests/emission_experiment.test.cjs
node research/run_emission_experiments.cjs
```

The offline runner writes 80 comparisons (five TVLs × four paths × four mechanisms)
and full monthly stress output under `data/results/emission-experiments/`, with
SHA-256 of the unchanged exported historical payload. Defaults: balanced, 25-year
schedule, 15-year protected duration, reference $0.10, 80% sale of unlocked new
YLT, 50% lock, no external demand or treasury revaluation. Reviewed 500M → 50M:

| Mechanism | Exhaustion month | Emission YLT | Max liquidity stress |
|---|---:|---:|---:|
| A | 300 | 55M | 0.8381 |
| B | 3 | 55M | 2.1577 |
| C | 300 | 55M | 0.9230 |
| D | 300 | 55M | 0.9230 |

C/D coincide here because revenue demand exceeds their common budget bound.
This is not evidence that D improves economics. Treasury NAV is identical across
mechanisms because emission does not create treasury revenue.

**Do not interpret terminal AMM prices as forecasts:** the default model can
produce extreme prices (stress run B roughly $590k/YLT) because repeated buybacks
shrink AMM YLT reserves after early exhaustion and old holders never sell. This
is an explicitly exposed structural limitation, not value validation. Sensitivity
to treasury losses, low yields, high selling, low liquidity and outside-holder
selling must precede economic conclusions. No arbitrary price ceiling conceals
that behavior. Arbitrage, outside sales, AMM fees/MEV, lock expiries, liabilities,
asset defaults, emissions to other buckets and real market depth are absent.

Verification: 156 pre-existing Python tests passed before additions. New Node
suite (13 tests) verifies geometric schedule, floor over all protected lifetimes and TVL
paths, carry/supply conservation, spot independence, zero locks/revenue, economic
ledger, AMM execution, annual/CSV consistency, real historical calibration across
all TVLs/strategies, input validation, and UI render/change/download behavior with
a minimal in-memory DOM. This is **not a visual browser test**. Local file browser
preview was blocked by browser URL policy; no screenshot validation is claimed.