Name the two operating modes, and correct G6's fleet figure by 5x - #197
Open
defenwycke wants to merge 1 commit into
Open
Name the two operating modes, and correct G6's fleet figure by 5x#197defenwycke wants to merge 1 commit into
defenwycke wants to merge 1 commit into
Conversation
The project has one goal and two workloads, and most of the fleet arguments this board has had came from pricing them as one. GOALS already separates them as G2 and G6; what was missing is that they are sized differently, on different deadlines, and do not carry the same fidelity risk. G6 said "roughly six L40S-equivalents are needed to hold position". That came from block 741,000 at 55 minutes. G2, in the same document, measures a near-tip block at 17,340 s — 4.8 hours — which against a 600 s interval is ~29 cards, not six. Corrected, and cited to the measurement already on the page. The correction matters beyond the factor of five, because ~29 is BREAK-EVEN: it matches the chain's growth rate and burns down none of the backlog. Backfill is a separate purchase on top — 138-229 card-years of it — so +100 cards is roughly 1.4-2.3 years and +500 is 0.3-0.5. Which settles the sequencing question that has been open all week. While backfilling you are by definition nowhere near the tip, so "is this block proved within 600 s of appearing" cannot be answered until catch-up is in sight, and that is years out at any plausible fleet. Size for throughput now; the latency-versus-bounded-lag decision is real but deferrable, and TOPOLOGY_AND_SETTINGS section 1 says so rather than presenting it as the next decision to take. Also records the fidelity asymmetry, which is the part that is easy to miss. Backfill proves a closed, enumerable set: every signature it will ever see already exists and can be differential-tested exhaustively against libsecp. Tip proving cannot be. So an acceleration that is unacceptable at the tip may be defensible for history — and since backfill holds 138-229 card-years against tip-following's ~29 cards, that is where the leverage is. hazync#139 makes the same distinction in its own text. Nothing is decided here; it is written down so the option is visible. Claude-Session: https://claude.ai/code/session_015nyynoX7kXHe7zCF3j6Jbf
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The project has one goal and two workloads, and most of the fleet arguments this board has had came from pricing them as one.
GOALS.mdalready separates them as G2 (genesis→tip) and G6 (tip-following); what was missing is that they are sized differently, run to different deadlines, and do not carry the same fidelity risk.The correction: G6 was stale by ~5x
G6 said "roughly six L40S-equivalents are needed to hold position". That was derived from block 741,000 at 55 minutes. G2, in the same document, measures a near-tip block (962,000, ~7,200 inputs) at 17,340 s = 4.8 hours:
⇒ ~29 cards is BREAK-EVEN. It matches the chain's growth rate exactly and burns down none of the backlog. That is the part the "six" figure hid.
The two modes
Backfill is a separate purchase on top of break-even:
What this settles
The latency-vs-bounded-lag question is not live yet. While backfilling you are by definition nowhere near the tip, so "is this block proved within 600 s of appearing?" cannot be answered until catch-up is in sight — years out at any plausible fleet. Size for throughput; a bounded lag costs nothing that matters.
TOPOLOGY_AND_SETTINGS.md§1 now says so, rather than presenting it as the next decision to take.The fidelity asymmetry, written down because it is easy to miss
Backfill proves a closed, enumerable set of blocks: every signature it will ever see already exists and can be differential-tested exhaustively against libsecp. Tip proving cannot be, because its inputs have not been written yet.
⇒ An acceleration that is unacceptable at the tip may be entirely defensible for history. Since backfill holds 138-229 card-years against tip-following's ~29 cards, that is where the leverage is. hazync#139 (bigint2 ECDSA, 13.78x per verify) makes the same distinction in its own text.
⚠ Nothing is decided here. It is written down so the option is visible when #139 is actually assessed.
Docs only.
https://claude.ai/code/session_015nyynoX7kXHe7zCF3j6Jbf