Key takeaways
- A block is excess only relative to the remaining meld slots, not a fixed slogan about block count.
- Standard 13-tile hands need four melds; Taiwanese 16-tile hands need five. Both also need one pair.
- Edge and inside blocks are not automatically weakest; compare live tiles, improvements, and rules value.
- If two blocks accept the same tile, that tile cannot be added twice.
- Long runs and compound shapes may share tiles productively, so overlap is not an automatic instruction to break.
What is a block?
An incomplete two-tile shape with a path to a chow is a block:
- Two-sided: 45m accepts 3m or 6m.
- Inside: 46m needs 5m.
- Edge: 12m needs 3m.
A pair can develop into a pung and can serve as the head. Blocks and pairs therefore compete for limited structural slots; counting two-tile shapes without assigning roles is incomplete.
When are there too many blocks?
Count capacity first.
Standard 13-tile hand
The target has four melds. If two are complete, only two meld slots remain. Four additional blocks cannot all survive; at least two are beyond capacity.
Standard 16-tile hand
The target has five melds. With two complete, three meld slots remain. The same four blocks exceed capacity by only one.
The pair requires a separate role. If no stable pair exists, some block or floating tile must still create it. This is why a fixed “five-block” or “six-block” rule is unreliable.
Map the hand’s roles
Temporarily divide the hand into:
- Completed melds
- Pair candidates
- Meld candidates: two-sided, inside, edge, and pairs
- Floating tiles
One tile cannot fully occupy two roles at the same time. The run 3456m can be 345 + 6, 3 + 456, or 34 + 56, but it is not already two completed melds.
How do you find the weakest block?
Do not stop at “two-sided is better than inside, which is better than edge.” Compare in this order.
Does breaking it worsen shanten?
Normally reject a discard that worsens shanten unless scoring or defense supplies a deliberate reason.
How many live tiles remain?
The edge block 12m theoretically accepts four 3m. If three are visible, only one remains. An inside block with all four copies live may be stronger.
Which improvements exist?
An inside 46m completes with 5m and may become two-sided after 3m or 7m. Some edge blocks have fewer equivalent improvements. Compare the next shape, not only direct completion.
Does it carry rules value?
A Value Honor pair, dora-adjacent shape, flush route, or tai/fan requirement can justify a narrower candidate.
Does it retain defense?
Against visible pressure, choosing which block to break also determines whether useful safe tiles remain.
What is overlapping acceptance?
The block 23m accepts 1m and 4m. The block 56m accepts 4m and 7m. Both include 4m.
Adding each block’s theoretical eight copies gives sixteen, but the four copies of 4m were counted twice. The union is 1m, 4m, and 7m—twelve theoretical copies before subtracting visible tiles.
Overlap can:
- Inflate the headline acceptance count.
- Leave one block awkward after the shared draw fills the other role.
Compare the effective-tile set for the whole hand, not the sum of isolated block totals.
Why do adjacent pairs collide?
The shape 55p66p contains two pairs competing for one head position. It can work beautifully if one pair becomes a pung and the other the head, but sequence-related draws can force an awkward split.
Adjacent pairs are not automatic discards. Double-pair waits, All Pungs, or the extra meld in 16-tile Mahjong can increase their value. Ask:
- Does the hand still need a head?
- How many melds remain?
- How many live copies remain for each pair?
- Does breaking one side improve sequence development?
- Does the ruleset reward pung- or pair-based forms?
Which overlap is actually useful?
Long and compound runs create flexibility by sharing tiles. The shape 3456m preserves both end extensions and can choose different melds after later draws. A tile supporting several decompositions can be a source of efficiency.
Distinguish:
- Harmful overlap: candidates exceed capacity, acceptance is double-counted, and breaking the shape preserves shanten or cleans the hand.
- Useful compound shape: shared tiles create more acceptance, improvements, or decompositions, and breaking them narrows the hand.
Overlap is a diagnostic signal, not an automatic discard rule.
A six-step cleanup
- Set the target: four melds for 13 tiles or five for 16.
- Subtract completed melds: count remaining slots.
- Reserve a pair candidate: do not treat every pair only as a meld.
- Build the effective-tile union: merge duplicates and subtract visible copies.
- Simulate each break: compare shanten, live tiles, resulting shape, and rules value.
- Add game state: value, defense, turn, and opponent speed.
If candidates are close in shanten and live acceptance, break the narrower block with fewer improvements and more overlap.
Common mistakes
- Counting blocks without subtracting completed melds.
- Adding each block’s draws independently and duplicating shared tiles.
- Breaking every inside shape without checking live tiles or improvement.
- Keeping every pair even though only one head is required.
- Treating every shared tile as harmful and destroying flexible long runs.
Scope and limitations
This framework concerns standard hands. Seven Pairs treats pairs as completed units and needs a different capacity model. Yaku, fan, tai, flowers, defense, and visible game state can still override structural cleanup.
Practicing cleanup with Tingpai Lab
Before viewing a candidate ranking, label the completed melds, pair, blocks, and floaters, then write the union of effective tiles. When checking the explanation, identify whether the losing candidate exceeds capacity, double-counts acceptance, or leaves a weaker follow-up instead of memorizing only the discard.
Continue reading: