I turned the advice into a small, runnable read-only channel advisor, rather than leaving it as prose. It ingests a per-channel CSV export and emits one bounded action per channel: HOLD, RAISE_FEE, LOWER_FEE, EXPAND_CANDIDATE, EXIT_CANDIDATE, or REVIEW.

Key safeguards:

• calculates net sats after rebalance and allocated chain costs;
• prioritizes uptime review before capital actions;
• requires both scarce outbound liquidity and observed failures before recommending a fee increase;
• never targets 50/50 mechanically;
• caps a proposed rebalance budget at the lower of half the observed net return or the channel's historical earned-rate value;
• is advisory only: it has no node credentials and cannot change fees, rebalance, open, or close channels.

I cannot publish a GitHub gist from this environment without an authenticated GitHub identity, so below is the complete core function, ready to paste into a local script:

def advise(row):
    number = lambda key: float(row.get(key) or 0)
    capacity = max(number("capacity_sat"), 1)
    local_ratio = number("local_sat") / capacity
    net = number("fees_sat") - number("rebalance_cost_sat") - number("chain_cost_sat")
    earned_ppm = number("fees_sat") * 1_000_000 / max(number("forwarded_sat"), 1)

    if number("uptime_pct") < 99:
        action, reason = "REVIEW", "uptime below 99%"
    elif number("forwarded_sat") == 0 and net <= 0:
        action, reason = "EXIT_CANDIDATE", "no routed volume and non-positive net"
    elif local_ratio < 0.125 and number("liquidity_failures") >= 10:
        action, reason = "RAISE_FEE", "scarce outbound liquidity with observed demand"
    elif local_ratio > 0.70 and number("forwarded_sat") < capacity * 0.10:
        action, reason = "LOWER_FEE", "abundant outbound liquidity with weak demand"
    elif net > 0 and number("forwarded_sat") >= capacity:
        action, reason = "EXPAND_CANDIDATE", "positive net and repeat capital turnover"
    else:
        action, reason = "HOLD", "insufficient evidence for a capital action"

    max_rebalance_cost = max(0.0, min(net / 2, capacity * earned_ppm / 1_000_000))
    return action, reason, net, max_rebalance_cost

Tested cases: scarce-liquidity fee raise; idle loss-maker exit; low-uptime review precedence; and 2× rebalance-coverage cap. All four pass.