Twelve Buy Signals, Four of Them Flagged: The Price Run and the Risk Layer Are Two Different Clocks


Executive Summary: A Buy Signal Is Not a Clearance

The price-window run generated at 06:06:31 Beijing time on 23 October 2026 scored sixty monitored route pairs against a five-month averaging window covering October 2026 through February 2027, and returned twelve buy signals, five wait, twenty-eight watch and fifteen avoid. [source-1] The mean ratio to mean of each band is monotonic, which is the property that makes the four labels mean something: buy at 0.652, wait at 0.982, watch at 1.038, avoid at 1.342. [source-1]

The same run attached a geo-risk level to every pair. Sixteen pairs sat at alert, fourteen at watch and thirty at none. [source-1] The two layers were computed from different inputs on different clocks, and the interesting fact is in the overlap: the sixteen alert pairs have a lower mean ratio to mean, 0.944, than the thirty unflagged pairs at 1.121. [source-1] The flagged part of the portfolio is the cheaper part.

All ten monitored pairs into Taipei carried an alert flag. [source-1] Three of those ten were buy signals: SIN-TPE at 134 dollars against a mean of 244, ratio 0.549; SZX-TPE at 192 against 287.50, ratio 0.668; CTU-TPE at 164 against 210.67, ratio 0.778. [source-1] The cheapest pair in the entire sixty-route network was one of those ten, and it was flagged.

That is not a contradiction. It is what two clocks look like. The ratio is a statement about a price against a route's own history. The flag is a statement about the operational environment around the endpoints. A storm that removes capacity can push a fare into the buy band precisely by removing the demand that would have supported it, and a buy signal produced that way is a price statement about a schedule that may not operate. [source-1]

1. The Four Bands and Their Means

Sixty pairs, four labels, four means. The table is the whole distribution.

Signal	Pairs	Mean ratio to mean	Mean coefficient of variation	Cheapest pair in band	Dearest pair in band
BUY	12	0.652	0.29	SIN-TPE 0.549	SIN-KIX 0.824
WAIT	5	0.982	0.17	JFK-TYO 0.962	LHR-TPE 1.004
WATCH	28	1.038	0.13	LAX-TPE 0.885	BKK-TYO 1.136
AVOID	15	1.342	0.25	PEK-TPE 1.153	SZX-KIX 1.989

Sources: [source-1]

Two readings follow. The first is that the avoid band is 2.06 times the buy band on the same measure, 1.342 against 0.652, which is the spread the labels are designed to capture. [source-1] The second is that the watch band, at 1.038, sits almost exactly at par, which means twenty-eight of the sixty pairs were priced at their own five-month average and the engine declined to call them either way. That is the largest single group in the run, and it is the group that a reader tends to skip.

The volatility column is the qualifier. Buy carries the highest mean coefficient of variation at 0.29 and watch carries the lowest at 0.13, which is the expected ordering: a fare far from its mean is usually a fare that moves. [source-1] The consequence is that a buy signal on a volatile pair is a weaker statement than the same ratio on a stable pair, and the run publishes both numbers side by side so the reader can see it.


2. Sixteen Alert Pairs and a Cheaper Mean

The geo-risk layer is not part of the price calculation. It scores the operational environment at the endpoints from a separate input set, historical security baselines and incident feeds, and it returns a level per pair. [source-1]

Flag level	Pairs	Mean ratio to mean	Share of pairs below 1.0
Alert	16	0.944	8 of 16
Watch	14	0.944	6 of 14
None	30	1.121	9 of 30

Sources: [source-1]

The table says something that runs against the intuition that risk is priced as a premium. The sixteen alert pairs average 0.944 against the mean, which is 5.6 percent below their own five-month averages, and the thirty unflagged pairs average 1.121, which is 12.1 percent above. The gap between the two groups is 15.8 percent in favour of the flagged group. [source-1]

Two explanations are available and the run cannot separate them. The first is compositional: eleven of the sixteen alert pairs are Taiwan, Japan or Korea pairs, and that cluster happened to be cheap on this date for reasons that have nothing to do with the flag. The second is mechanical: a hazard that suppresses demand suppresses the fare with it, so the alert and the discount are both consequences of the same event. [source-1] [source-2]

The honest reading is that the flag is not a price input and should not be treated as one. What the flag does is tell a reader that the cheapness may be an artefact of an operational condition, which is exactly the case where a buy signal needs a second look before it becomes a booking.

3. All Ten Taiwan Pairs, One Flag

The Taiwan cluster is the cleanest case in the run because it is complete. Every monitored pair into Taipei carried an alert flag, ten of ten, and the ten pairs were distributed across all four signal bands. [source-1]

Pair	Signal	Current fare	Five-month mean	Ratio to mean
SIN-TPE	BUY	134	244.00	0.549
SZX-TPE	BUY	192	287.50	0.668
CTU-TPE	BUY	164	210.67	0.778
LHR-TPE	WAIT	680	677.00	1.004
LAX-TPE	WATCH	762	860.67	0.885
CAN-TPE	WATCH	324	314.00	1.032
BKK-TPE	WATCH	208	198.00	1.051
HKG-TPE	WATCH	154	144.40	1.066
PEK-TPE	AVOID	371	321.80	1.153
PVG-TPE	AVOID	293	249.40	1.175

Sources: [source-1]

The spread inside one flagged destination is 2.14 times, from 0.549 to 1.175, which is the same lesson the network taught on 22 October at origin level and teaches again here at destination level: the unit of the measure is the pair. [source-1] A statement that Taiwan is cheap is not supported by this table. A statement that three specific pairs into Taipei were 22 to 45 percent below their own means is. [source-1]

The three buy pairs are also the three cheapest absolute fares in the cluster, 134, 164 and 192 dollars, which is a coincidence worth naming as one. The ratio and the absolute price agree in this cluster and did not have to. [source-1]

4. Four Alert, Four Watch, Four None

The twelve buy signals split exactly into thirds by flag: four at alert, four at watch, four with no flag at all. [source-1] An exact split invites the reading that the engine balances the bands, and it does not: the split is an artefact of twelve observations and would not survive a second run. [source-1]

What is not an artefact is which pairs landed where. The four flagged buy pairs are SIN-TPE at 0.549, SZX-TPE at 0.668, CTU-TPE at 0.778 and JFK-ICN at 0.665. [source-1] Three of those four are Taiwan pairs, which is the cluster from the previous section, and the fourth is a long-haul pair into Incheon. The four unflagged buy pairs are SIN-ICN at 0.586, CAN-KIX at 0.603, CTU-ICN at 0.746 and SIN-KIX at 0.824, and none of them touches the western Pacific cluster that produced the alert set. [source-1]

That distribution is the useful output of the run. A traveller who wants a buy signal without an operational qualifier has four pairs to work with, all of them on the Korea and Japan side. A traveller who wants the deepest discounts has to take a flag with them, because the deepest discounts in this run are on the flagged side.


5. The Two Ends of the Portfolio

The avoid band is where the run is most emphatic. SZX-KIX carries the highest ratio in the network at 1.989, 788 dollars against a mean of 396.25, with a coefficient of variation of 0.59 and a month-on-month change of plus 118 percent. [source-1] CTU-KIX follows at 1.816, 419 dollars against 230.75, with a month-on-month change of plus 152 percent. [source-1] Both are Osaka pairs, and both are the mirror image of the three Taipei buys: same country of destination, opposite direction, and a mean that has been left behind by the current fare.

The cheapest absolute fare in the run is SIN-TPE at 134 dollars, which is also one of the flagged buy pairs. [source-1] The most expensive absolute fare is CAN-JFK at 1,157 dollars, a watch signal at ratio 1.025, which is a useful reminder that the ratio and the fare are different scales: the dearest ticket in the portfolio is priced at its own average, and the cheapest is priced 45 percent below it.

The run's completeness accounting matters for all of these numbers. All sixty pairs carried a risk field, a geo-risk field and a weather field, which is the check that separates a partial run from a usable one. [source-1] A ratio computed on a pair with no risk input is still a ratio, but a reader cannot tell whether the flag would have been there.

For the ground side, the two instruments that behave like a hedge against the flag are a fixed-price airport transfer and flight compensation cover, because both convert a schedule failure into a cost that is knowable in advance, and an eSIM data plan is what keeps the carrier app reachable when the terminal network is under load. [source-3] [source-4] [source-5]

- Read the ratio as a pair-level fact. A destination-level claim is not supported by a pair-level measure, and the ten Taiwan pairs span 2.14 times. [source-1]

- Read the flag as an operational qualifier, not a price input. The flagged group averaged 0.944 against the unflagged 1.121, and neither number is a premium. [source-1]

- Check the coefficient of variation before acting on a buy. Buy carries the highest mean volatility of the four bands at 0.29. [source-1]

- Treat an exact split as an accident. Four alert, four watch, four none is a distribution over twelve observations, not a design. [source-1]

6. Conclusion: Two Clocks, One Table

One run, sixty pairs, four bands with means of 0.652, 0.982, 1.038 and 1.342, and a flag layer that put sixteen pairs at alert, fourteen at watch and thirty at none. [source-1] The bands are monotonic, which is what makes the labels usable, and the flags are independent, which is what makes the overlap informative.

All ten Taiwan pairs were flagged and three of them were buys, including the cheapest fare in the network at 134 dollars. [source-1] The flagged group was cheaper on average than the unflagged group by 15.8 percent, and the explanation is more likely compositional than causal: a hazard and a discount can share an origin without one being the other. [source-1] [source-2]

The measure answers a question about a price against a route's own history. The flag answers a question about the environment at the endpoints. A reader who needs both answers has to read both columns, and the run publishes them side by side for exactly that reason. Twelve buy signals, four of them flagged, and no column tells the whole story on its own.