docs: V2 bond sizing and window analysis

[Kayce10]

Summary

Adds docs/src/V2_BOND_SIZING.md, the economic simulation document requested in issue #75. This is a research/design document — no contract code is changed.

The document models the three quantitative questions the issue scoped:

1. Default-frequency model

Derives the minimum reveal-turnout required for the against-side to reach a strict majority as a function of the honesty share among revealers. Shows the issue-body example (49% revealed against, 50% non-reveal → assertion stands) in formula form and generalises it to a table. Identifies that a lone disputer at W = 2B can never cross the threshold and concludes that N_voters ≥ 3 realistically achievable third-party registrants is the practical floor for a correctable deployment.

2. Whale-capture model

Derives the minimum capital required for a single address or coalition to reach strict majority as a function of base_bond, max_position, and expected honest counter-stake. Provides a concentration-threshold table and shows that max_position = 2 × base_bond is the tightest single-parameter whale control. Discusses the self-hedging cost-illusion (a coalition's net exposure is its registration deposit, not the asserter bond it can recover).

3. Window-length tradeoffs

Frames registration_duration_secs, reveal_duration_secs, anti_snipe_extension_secs, and anti_snipe_hard_max_secs as a joint optimization over participation latency (τ_react), coordination time (τ_coord), and attacker capital-movement speed (τ_atk). Provides a sanity-check inequality set and a summary table.

4. Recommended profiles

Three-tier table (private beta / public testnet / higher-value mainnet candidate) matching the style of BOND_SIZING.md, with numeric ranges for all five window parameters plus max_position and max_total_weight, followed by narrative guidance and scenario checks.

Files changed

• docs/src/V2_BOND_SIZING.md — new file (581 lines)
• docs/src/SUMMARY.md — adds the new doc to the book outline under Deployment & Operations: Protocol v2

Testing

Research/documentation change only — no contract code, no WASM, no tests affected.

Closes #75

[Kayce10]

All three requested changes have been addressed in commits 0e76300 and 636d4df:

1. Em/en dashes removed — all 21 em dashes and 16 en dashes replaced with plain
   punctuation (colons, semicolons, parentheses, or rephrased sentences) throughout
   V2_BOND_SIZING.md.
2. capture-cost phrasing fixed — the "three times / four times / eight
   times" prose now reads "just over twice / just over three times / just over seven
   times" to match Part 4's capture_cost_multiplier ~= 2 and avoid understating how
   cheap capture is at R_w > 2B.
3. threshold reconciled — the scenario check now uses the contract's actual
   comparison (side_weight > W - side_weight, equivalently 2 x side_weight > W) and
   states floor(W/2) + 1 for integer W consistently. The N_voters = 1 worked example
   is updated to match.

CI is fully green on 636d4df (test, sdk, demo, build all passing). Ready for
re-review.

[collinsezedike]

All three previous findings are genuinely fixed, not just cosmetically, the em/en dashes are gone, Part 2 and Part 4 now agree on the zero-counter-stake case, and the W/2 threshold section correctly explains the floor(W/2)+1 parity handling with a matching worked example. Two new things surfaced this round.

[Kayce10]

Both findings from this round are addressed in bae9ce2.

1. Self-hedging section: removed the incorrect claim that the asserter bond is
   recovered when the honest side wins. settle() for StrictMajorityAgainst pays only
   positions where agrees_with_outcome == Some(false); the asserter's Fixed(true)
   position has Some(true) and is forfeited along with R_w. The section now states the
   real net exposure on loss as B + R_w and uses that as the capture cost throughout.
2. N_voters rule of thumb: added a clarifying paragraph making explicit that
   N_voters >= 3 is a floor for a healthy electorate, not a sufficient condition for
   correction. The symmetric-split model shows the against-side weight lands at
   exactly W / 2 regardless of headcount; what determines correction is an honest-side
   imbalance in revealed weight, not just how many participants registered.

Ready for re-review.

[collinsezedike]

Both findings from the last round are genuinely fixed, the self-hedging exposure and the N_voters clarification both check out against the actual settle() mechanic. One new thing this round.

[collinsezedike]

This says the subtraction form avoids doubling overflow, but that's not why the code uses it. lib.rs:1376-1378's own comment says it's because integer division rounds down and would wrongly pass a boundary case on an odd W, not overflow, doubling side_weight can't overflow anyway since it's bounded well below i128::MAX. Please fix the stated rationale to match.

[Kayce10]

Fixed in 9b96191. The rationale now matches lib.rs:1376-1378: the subtraction form
is used because integer division rounds down and would wrongly pass a boundary case
on an odd W, not to avoid overflow. The overflow claim is removed.