Which verdicts survive a price change
Which cost decisions survive a new price list, and what did pricing add to other trade-offs?
Which verdicts survive a price change?
This is Priya's question. The answer follows from asking whether a conclusion depends on a rate or on a ratio or structure.
| Conclusion | Depends on | Survives the new price list? |
|---|---|---|
| Compact small files to about 100 MB or more | S* = request fee ÷ byte fee, plus the 20–100 ms latency floor | Yes. If scan price halves, S* doubles from about 80 KB to about 160 KB. The target (hundreds of MB) is far above both. The latency floor does not change at all. |
| Prefer incremental refresh to full recompute | ratio of changed data to table size | Yes. Rates cancel: 10 TB at $2.50/TB is $25 per day, incremental is $0.25. Still 100× cheaper. |
| Cold tier worth it only if reads are rare | p(read) × retrieval cost < stock saving | Changes numerically. Storage rates fall 20%, so the saving shrinks and the break-even moves. The form of the test stays. |
| Commit at the demand quantile equal to the discount | shape of the loss (an unused unit costs the committed price, a missing one only the discount) | Yes, unless the discount itself changes: the quantile is set by the discount, not by the list rate. |
| Three-year commitment at 45% | rate falling about 20% per year | Check again. The headline discount erodes. A 45% discount over three years with 20% yearly price falls is worth about 32% (see below). |
| Compress as hard as possible | which side of the meter the CPU sits on | Inverts if bytes are counted uncompressed. Compression then has storage value only. |
| Result cache is worth everything or nothing | scan pricing versus occupancy pricing | Inverts if you move between those two meter types. |
| Do not optimize a saving below about $500 per month | labour cost versus infrastructure cost | Gets stronger. Infrastructure prices fall, labour rises, so each optimization is worth less every year. |
The rule of thumb. A conclusion that comes from a structure (a ratio of changing quantities, a fixed latency, an asymmetric loss, a multiplication) lasts. A conclusion that comes from a specific number in a price list does not. Write down, for every cost decision, which of the two it is.
Money buys position on every other trade-off. A design pays in waiting, wasted work or weakened guarantees; money is a fourth currency, and wherever a list of currencies exists, cost extends it by one.