Develop and test paired keys

Choose a useful difference

A feature helps a key only when it separates the records still being considered. Compare the observed values before choosing a question

1. 1Different values separate P is smooth and unlobed. Q is smooth and lobed. Their outlines differ, so outline separates them. Asking about their shared smooth edge does not
2. 2Either can work If two records differ in both edge and outline, either feature can separate them on its own. Both are not required for that pair
3. 3Neither may be enough Two different samples can have the same recorded edge and outline. These observations cannot separate them. Seek another clear feature instead of assuming that they are the same species

P has a smooth edge and an unlobed outline. Q has a toothed edge and a lobed outline. Which recorded feature, used on its own, can separate P from Q?

Exactly one alternative for each record

At a paired step, every record reaching it needs one route. Check each record against both alternatives. Two matches cause overlap; no matches cause an omission

1. 1A clear pair For these simplified edge records, smooth edge and toothed edge cover every record without overlap. Each record has exactly one of those recorded values
2. 2An overlapping pair Suppose a pair instead says smooth edge or lobed outline. A smooth, lobed record meets both alternatives, so it has two possible routes
3. 3A missing route A toothed, unlobed record meets neither of that pair's alternatives. The same pair can therefore have both overlap and omission. Finding one fault is not a reason to stop checking
4. 4Check the supplied collection A smaller collection may lack one of those counterexamples. Base the verdict on every record actually supplied, not on an imagined extra record

The collection contains X (smooth, unlobed), Y (smooth, lobed) and Z (toothed, lobed). A pair offers smooth edge or lobed outline. What happens for these three records?

Test X, Y and Z against both conditions. Is any record left with no route?

Consider only the records that reach a step

A feature can be useful at the first step but useless when repeated later. If step 1 has already sent all smooth-edged records to step 2, asking about edge again cannot separate that group

1. 1Follow the first split In the next model, A and C are smooth; B and D are toothed. Edge first sends A and C to step 2, and B and D to step 3
2. 2Find a difference within each group A is unlobed and C is lobed, so outline separates them at step 2. B is lobed and D is unlobed, so outline also separates the other group at step 3
3. 3Read the endpoints One code at an endpoint means one reference remains. Two codes mean those references are still together. None means no supplied record reaches that endpoint; an empty branch alone is not proof that a key is wrong

This draft asks about edge at all three steps. Why do A and C remain together at step 2?

Predict which records will separate if only step 2 changes to Outline, then try it. Change step 3 too and test all four records. Now change step 1 to Outline. Which later features need to change to distinguish all four again?

Why can repeating step 1's feature at a later step fail to separate the remaining records?

Test, revise, then test again

A successful route for one record does not test the whole key. Trace every record from step 1 and check that each has one supported route to its own endpoint

1. 1Keep a test record Record the choices and final endpoint for each sample. Look for a record with no route, two possible routes, or an endpoint shared by different references
2. 2Repair the relevant step Replace an unclear or unhelpful distinction with a clear observed difference. Retest the whole collection, including records that worked before the change
3. 3Try independent use With a teacher, record features from a suitable real collection, write your own paired key and ask a classmate to use it without extra hints. Compare their routes with the observations, then revise ambiguous wording
4. 4Keep the coverage clear Name the collection your key covers. These two model features are not enough to identify all real plants, and a route does not establish evolutionary relatedness

A revised key gives a separate, supported endpoint for three of its four references. The fourth has not been tested. What can you conclude?

A useful difference at every step

Choose clear recorded differences. Each record reaching a pair should fit exactly one alternative. Later questions must separate the records still together, not simply repeat an earlier question

1. 1Check the result Trace every record to an endpoint. Repair an overlap, omission or unresolved group, then retest the whole collection
2. 2Know when evidence is missing If the available features cannot separate two records, seek another reliable observation. Do not invent a distinction or infer that they are the same species
3. 3Beyond this model Choosing and repairing supplied alternatives does not fully assess building an unrestricted key, observing real specimens, explaining decisions in writing or testing with another person