simpa: adversarial mining scenarios driven by DAGKnight counters (#1112)

[Kali123411]

Closes #1112.

Honest Simpa runs with a uniform broadcast delay produce almost no DAG conflicts, so DAGKnight's UMC cascade never runs and the DK counters stay ~0. This PR adds deterministic adversarial mining scenarios that manufacture the conflict shapes the convergence proof reasons about, and asserts — via Consensus::dagknight_counters() — that each scenario exercised the intended consensus flow.

What's added

• simpa/src/simulator/adversary.rs (new): the Scenario enum, seed-derived AdversaryParams/AdversaryPlan, and per-miner AdversaryRuntime.
• miner.rs: MinerOptions gains an optional adversary plan; Miner::mine() branches into private-fork / latency-burst handlers. Withhold-family adversaries ignore incoming honest blocks (keeping a pure private fork), insert locally, and release via staggered targeted env.send (a monotonic sequence guarantees parents arrive before children). Adversary blocks are mined empty — the DAG shape, not tx load, drives the cascade.
• network.rs: init_with_adversary() over a shared init_inner; the public init_with_lane_producer signature is unchanged.
• main.rs: --scenario/--seed + params; after the run it snapshots the DK counters, asserts the expected counters moved, logs SCENARIO <name>: PASS/FAIL, and exits non-zero on failure.

All schedules derive from --seed, so a (scenario, seed) pair is fully reproducible.

Scenarios and results

Run on the dagknight branch (seed 1, 2 miners unless noted, --sim-time 400, --tpb 32). Each scenario's "done-when" is the counter(s) it must move:

Scenario	Proof the flow ran (DK counters)
withheld-side-dag	total_calls=4296, cascade_flips=2808, voting_blocks=166387
latency-burst	total_calls=4278, max_flips=36 (transient parallel subgroups)
equal-rank-ties (3 miners)	total_calls=20493, max_cascade_flips=120, total_flips=192106
strong-shortcut	from_checkpoint=4669 vs from_scratch=298 (94% reuse)
weak-shortcut	from_scratch=219 (>0), total_calls=2658
variable-difficulty	total_calls=10347, flips=7682
gray-context-change (--features baseline-debugging)	total_calls=4465, 0 baseline/cascade disagreements — no panic == agreement

Determinism: the same seed reproduces total_calls exactly; a different seed differs clearly. total_calls (DAG-shape driven) is fully deterministic; minor variance in flip/voting sub-counts comes from the consensus's internal parallel processors, not the adversary schedule.

Notes for reviewers

• -t/--tpb is transactions-per-block, not sim-time (-s/--sim-time). Scenario runs should pass a modest --tpb: the stock OnetimeTxSelector bypasses mass-aware selection, so a large tpb makes the honest miner's own blocks exceed the compute-mass limit and panic — a pre-existing simpa property, unrelated to the adversary logic.
• weak-shortcut and strong-shortcut both move their required counter (from_scratch>0 / from_checkpoint>0), but both runs are checkpoint-dominant in practice — a sustained conflict zone reuses checkpoints either way. Cleanly forcing from_scratch dominance would need re-rooting the fork at a fresh recent ancestor, which the current pure-private-fork mechanism doesn't do; noting it as a possible follow-up.

🤖 Generated with Claude Code

[Kali123411]

Pushed 948bf380: the Lints failure was formatting only (import ordering + line wrapping in simpa/src/main.rs and simulator/miner.rs). Fixed with cargo fmt --all; cargo fmt --all -- --check is now clean and clippy shows no new warnings on the simpa changes. No functional changes.

The re-run is sitting in action_required (first-time-contributor workflow approval) — a maintainer "Approve and run" would let CI re-check. Thanks!

[STP-KAS]

Outsider review of the scenario assertions (not driving the adversary logic).

scenario_assertion for WithheldSideDag currently requires only total_calls > 0 && total_voting_blocks > 0. The PR table and the commit message already claim this scenario is the one that moves cascade flips (seed 1: total_calls=4296, cascade_flips=2808, voting_blocks=166387). Honest simpa with uniform delay barely conflicts, so total_calls alone does not prove UMC cascade ran.

Suggested one-line tighten in simpa/src/main.rs:

Scenario::WithheldSideDag => (
    s.total_calls > 0 && s.total_cascade_flips > 0 && s.total_voting_blocks > 0,

VariableDifficulty already requires total_cascade_flips > 0. Same bar for the withheld side-DAG is what makes DK-208 actually pin Alg. 6.

Follow-ups, not this PR:

• gray-context-change still only asserts total_calls > 0 (the baseline/cascade panic is already gone on dagknight / Dk UMC depth restriction #1127).
• weak-shortcut asserts from_scratch > 0 while both shortcut runs stay checkpoint-heavy, as the notes already say.

Patch: https://github.com/STP-KAS/dagknight-test-grok/blob/main/tests/pr1124_withheld_cascade_flips.patch

[STP-KAS]

Exact suggested change, since the previous comment mangled fences:

// simpa/src/main.rs  fn scenario_assertion
Scenario::WithheldSideDag => (
    s.total_calls > 0 && s.total_cascade_flips > 0 && s.total_voting_blocks > 0,
    format!(
        "total_calls={}, total_cascade_flips={}, total_voting_blocks={}",
        s.total_calls, s.total_cascade_flips, s.total_voting_blocks
    ),
),

Today it is total_calls > 0 && total_voting_blocks > 0 only. The PR table already reports cascade_flips=2808 on withheld-side-dag. Pin that.

Patch: https://github.com/STP-KAS/dagknight-test-grok/blob/main/tests/pr1124_withheld_cascade_flips.patch