Internal · pre-IPO markets · data to 20 Nov 2026
One funding formula, four versions, and what each one did to ANTHROPIC
Lighter computes funding the same way on every perp market. On the pre-IPO markets, two settings changes and one engine change gave four different behaviours in five months. This page shows each version, lets you change every input, and replays the real ANTHROPIC and OPENAI history through any settings.
v1, launch. Full premium against the market's own 30-minute average: funding swung hard and, on net, slightly charged shorts.
v2, 20 Jul. Multiplier 1 switched the premium off entirely. Funding was the interest rate every single hour.
v3, 13 Sep. An engine change shrank the dead zone below the interest rate, so quiet hours round to zero.
Entropy, same period as v3. What Hyperliquid's io:ANTH paid shorts while its open interest grew to about $200M.
How funding is computed
The same four steps run on every perp market, every hour. Numbers are in engine units: 1 unit = 0.0001%. Rates are per 8 hours until the last step. A positive rate means longs pay shorts.
Sample the premium
ask/bid are the impact prices: the average fill for a fixed order size. A sample is zero whenever the index sits between them. On pre-IPO markets the index is the market's own 30-minute average price, so it usually does.
Average and scale
m is the per-market premium multiplier, 1 to 100. 100 passes the premium through in full; 1 keeps 1% of it.
Pin small premiums to interest
If P is within c of the interest rate I, funding equals I exactly. Otherwise it is the premium minus c. c was 500 (0.05%). Since 13 Sep it is ⌊500 × m / 100⌋.
Divide by 8, truncate
Stored in whole units of 0.0001% per hour (0.88% a year), rounded toward zero. Interest 32 → 4 → 0.0004%/hr → 3.5% a year.
Playground
Pick a version or change any input. The table shows each step with your numbers; the chart shows what the formula pays across all premiums. The shaded band is where funding equals the interest rate.
Build the premium from a book
One minute's sample from the index and the two impact prices. "Use" treats all 60 minutes of the hour as equal.
Funding (per year) against the hour's premium
ANTHROPIC, version by version
Lighter's actual hourly funding, with each formula version shaded. Entropy (Hyperliquid io:ANTH) is the competing market that took the open interest. Switch the orange line on to replay history with your playground settings.
Lighter mark price
Replay: what would each setting have paid?
Average funding to shorts, per year, if the same hourly premiums had gone through a different formula. "Your settings" uses the playground. "Fix" keeps each hour's real multiplier but never lets the dead zone fall below the interest rate.
Read these as mechanics, not forecasts. The replay holds the order book fixed. With real funding switched on, traders would have leaned against it: OPENAI's huge replay numbers at m = 100 after 13 Sep come from long stretches in October when its book sat well above its own average with nothing pushing back.
Fixes
The 13 Sep change works as intended at multipliers of about 7 and up (it makes m = 50 a clean "half funding"). It breaks when ⌊500 × m / 100⌋ drops below the interest rate: m ≤ 6 at interest 32, m ≤ 19 at interest 100. Today that covers OPENAI, OURA, VARIATIONAL and XTEN; VARIATIONAL is configured at interest 100 (about 11% a year) and paid exactly zero in all 1,020 hours checked up to 19 Nov.
1. Keep the dead zone at least the interest
c = max(⌊c₀·m/100⌋, |I|). Restores the interest floor at any multiplier and changes nothing at m ≥ 7. The API's predicted-rate copy needs the same change.
2. Round, or keep finer units
The hourly rate is stored in whole 0.0001% steps and truncated toward zero, so anything under 0.88% a year vanishes. That only bites when rates are small, which is exactly the low-multiplier case.
3. Anchor the index to an outside price
v1's problem was the index, not the multiplier: a market measured against its own average only sees momentum. Blending in an outside mark (Entropy uses at least 5% external weight) lets premium funding pay for real mispricing. Not replayable from this data.
4. Announce changes, ramp them in
v2 and v4 went live without notice, v4 in the middle of the IPO debut along with leverage going from 5× to 10×. Post the new parameters 72 hours ahead and phase them in.