[lgrawet]

Summary

Three defects in the Solana LP band construction, all in solana_dex_lp_expert:

• The width cap was a follow-up check, not a ceiling. The playbook derived a
  width from volatility and then tried to shrink it to fit the venue's bin/tick
  limit — a step an agent could skip, and did: an Orca open at 20.8% on
  tick_spacing=8 is 237 spacings → SIMULATION_FAILED, no funds moved, slot
  left empty.
• A flat percentage floor is impossible on fine-grained pools. A Meteora
  bin_step=10 pool caps the half-width at 3.40%, below any sane
  volatility-derived "4% floor", so a percent-first clamp is guaranteed to
  over-reach.
• The side guardrail contradicted the one-sided ranges the same skill defines.
  It required lower < P < upper unconditionally, while BUY (side=1)
  deliberately places both bounds below P and SELL (side=2) both above.

Changes

• lp_range_config skill: W_max tables for Meteora (68 bins, one under the 69
  cap) and Orca/Raydium (119 spacings, one under the 120 cap), derived from
  bin_step/tick_spacing; one-line off-table formula; the clamp expressed as
  ln(upper/lower) ≤ ln((1+W_max)/(1−W_max)) and required to preserve the
  band's side.
• lp_slot_operator loop: the same table-driven clamp; range bounds split into
  RANGE (asymmetric by base_pct, venue-flipping) and one-sided (ε-offset,
  venue-independent, lower=upper·(1−W) / upper=lower·(1+W)); the RANGE
  formulas explicitly restricted to 0<base_pct<100.
• AGENT.md: side-aware guardrail; width described as W = min(vol, W_max).

Verification

• Every table entry recomputed from its formula — all 19 values match.
• Each rounded Orca/Raydium entry re-counted against the cap: max 119.012
  spacings, under 120, so the four-significant-figure rounding cannot push a
  W_max band over.
• pytest tests/test_lp_skill_loop_agreement.py tests/test_lp_correctness.py
  → 57 passed.

Supersedes

#252, folded in here: both PRs rewrote the same clamp block, and their
one-sided formulas had already drifted apart.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge; no blocking issues remain.

Summary

The PR derives range limits from each pool's bin_step or tick_spacing and keeps the agent, loop, and skill aligned.

• LP bands now sit below, above, or around the price based on their side.
• Each pool's bin or tick size sets the maximum LP band width.

_{Reviews (4) · Last reviewed commit: "Scope the LP band's RANGE formulas and g..." · Reviewed by Greptile}

[lgrawet]

Note for whoever merges: this PR and #252 are two halves of the same fix.
This one makes the width clamp side-preserving, but the GUARDRAIL line just
above still reads "always verify lower < current_price < upper" — it still
demands a bracketing range. #252 is what turns that line into the per-side
check. Landing either alone leaves strategy.md self-contradictory.
Intended order: #251 first, then rebase #252 onto it — they touch the same
clamp paragraph.