Gas estimation is suboptimal

[paulbarclay]

There's a number of issues with the current code, but they all boil down to very poor gas estimation.

1. The calculation only looks at transactions - it ignores blocks that aren't full.

2. The calculation only looks at gas price for the transaction, so smaller transactions (which tend towards higher gas prices) take up a disproportionate piece of the total gas.

[FIX 1&2]: These two can be solved by weighting by gas cost - make the percentage the % of 20,000,000 that the token fills.

3. The buffer only holds 2-3 blocks worth of transactions. To calculate a more stable price, you need a longer lookback.

[FIX 3]: Rather than calculating over a buffer, run the calculation for each block individually (including the weighting from fix 1&2), store the values for each block using, and weighted average over a different number of blocks for each one (highest 10 of the last 15 blocks for safelow, highest 7 for standard, highest 4 for fast, and highest 2 for fastest)

4. The percentage calculations value outliers incredibly highly.

[FIX 4]: Use 98% rather than 100% for "highest". There's no need to match a single over-high data point when your goal is "just get this confirmed"

5. The calculation doesn't take into account how long pending transactions have been waiting - "safelow" and "fast" transactions complete at the same speed.

[Berky99]

Bb