Assumptions

SCORING ASSUMPTIONS AND THRESHOLDS

Last reviewed: 2026-09-25 Owner: Sean <[email protected]>

Everything below is a JUDGEMENT the software makes on your behalf. None of it is dictated by the test publishers; where a publisher does define a band it is called out as such. If a number here disagrees with what the code does, the CODE is what shipped - please report it (Settings > Report a Bug) so the two can be brought back together.

1. WATSON-GLASER (WG) - II AND III ARE READ DIFFERENTLY#

This is the WG assumption that actually affects your numbers, and it depends on which vintage of report you hand us.

OVERALL percentile - both vintages:

Read as printed. The number in the report is the number we use.

SUBSCORES (Recognise Assumptions, Evaluate Arguments, Draw Conclusions):

Watson-Glaser III   Measured off the printed bar and recorded as an actual
                    percentile, 1 to 99.

Watson-Glaser II    The report does NOT print a subscore percentile - it prints a
                    Low / Average / High band. We record the MIDDLE OF THAT BAND as a
                    representative percentile:

                        Low      -> 55
                        Average  -> 75
                        High     -> 90

                    So a WG-II subscore of 75 does not mean "this candidate scored
                    75th percentile". It means "the report said Average". Two WG-II
                    candidates in the same band get identical subscores, and a WG-II
                    subscore should not be compared like-for-like against a WG-III one.

Watson-Glaser I (2008 and earlier, the raw-score format) is NOT supported. It is warned about and skipped rather than guessed at as the test buckets are different from modern day test groups.

A Watson-Glaser report in a layout the app has never seen (for example an older report that is mostly a picture of the page) is NOT skipped: after the other files are read, the app opens it and asks you to type its overall percentile and three subscores. A report that is a picture with no readable text at all cannot be told apart from Watson-Glaser I, so it is skipped with a warning instead.

(The 50 line drawn across the WG and EAS group charts is the 50th-percentile reference mark. It is a chart gridline, not a threshold applied to anyone's data.)

2. CPI SENSITIVITY IS STORED INVERTED#

The CPI reports a "Sensitivity (Sn)" percentile. We store and report

Sensitivity = 100 - Sn

so that, like every other scale on the page, HIGHER IS BETTER for the trait being described. A candidate printed at Sn = 30 appears in our reports as 70.

In the Group Graphing charts this same value is LABELLED "Goal Orientation" rather than "Sensitivity". The stored key stays "Sensitivity" - only the chart label changes - so a Customer Table column headed Sensitivity and a chart bar headed Goal Orientation are the same number.

3. CPI LIFE SATISFACTION - 7-POINT SCALE#

Read off the position of the printed marker and bucketed into seven equal bands, 1 to 7, lowest to highest. When the marker cannot be located the app asks you to read it off the PDF; it only ever accepts a whole number 1-7.

4. CPI LIFESTYLE QUADRANT#

The quadrant (Implementer / Innovator / Supporter / Visualizer) is taken from where the candidate's symbol sits on the printed grid. When it has to be entered by hand, the value recorded is the CENTRE of the chosen quadrant, not an exact coordinate - so two candidates entered by hand into the same quadrant get the same point.

5. CPI LEADERSHIP SCORECARD#

Counted from the printed checkboxes, and the count is EXPECTED TO SUM TO 18. If it does not, extraction stops rather than reporting a partial score - a scorecard that does not sum to 18 means boxes were missed, and a quietly low number would read as a real result.

6. MISSING TESTS#

A candidate is never given a zero for a test they did not take. Missing tests are left empty, group averages are computed only over the candidates who DID take the test, and report pages for a test nobody took are dropped from the final PDF.

7. WHOSE REPORT IS THIS? - THE "LAST, FIRST" FOLDER DECIDES#

A person's name is read from each PDF's file name, and file names are not always consistent: a nickname on one test ("Bob") and the full name on another ("Robert"), the surname first with no comma, and so on. When that happens, a folder named "Last, First" (for example "Smith, Robert") is treated as the authority - the rosetta stone for that person's name:

Files for the same surname are gathered under the name of the person whose "Last, First" folder exists.
A name written surname-first is turned round to match its folder.

Two consequences worth knowing:

If two DIFFERENT people share a surname and only ONE of them has a "Last, First" folder, the other person's tests are gathered under the one with the folder. Give each person their own folder (or file names that already agree) to keep them apart.
Group Graphing and Headline Summary apply this automatically. Generate Company Table offers "Manual Oversight" instead, so you can check each merge yourself.

THE QUESTIONS ABOUT PEOPLE, IN THE ORDER THEY APPEAR

A result belongs to the NAME in its file name. Up to three questions can come up while a folder is read. They always come in this order, and each appears only when it is needed.

1. TWO RESULTS FOR THE SAME TEST (while the reports are being read)

The same name has two results for one test: a retest, a second copy of a report, or two different people who share a name. The question shows each file's name, its test date and its report type, and lists only the scores that differ, grouped by test. The NEWER result is highlighted, so pressing Enter keeps it; when either date is unknown nothing is highlighted and you choose.

Merge keeps the newer file's scores. Discard keeps the result you already had. Keep Both files the second result under the same name followed by "_1".
If you are not sure the two files are the same person, choose KEEP BOTH. It is the only answer that keeps both results. Discard throws the other result away for good, and no later question can bring it back.
"Remember this answer" (ticked unless you untick it) saves your choice for those two exact files, so re-running the same folder does not ask again. The answers are kept in a folder named ".merge_decisions" inside the run folder; delete it to be asked again. If either file changes, the question comes back.
"Use the NEWER result for every remaining conflict" answers the rest of that folder's questions the same way, except where a date is unknown.

2. ONE NAME, TWO COMPANIES (once all the reports are read)

The same name has results filed under two DIFFERENT company folders (folders named "1 - <Company>"). The app asks whether it is one person or two, instead of combining them silently. The question lists each company's tests with their dates and files, and shows the name each company's results would be kept under.

"Same person" keeps one record. Nothing more is asked about it.
"Two different people" keeps the company with the most results under the plain name, and files the other company's results under the same name followed by "_1". If you chose Keep Both at question 1, the rest of that company's results join the "_1" result you kept, so each person ends up in one record.
It asks only when BOTH companies are known. Files that sit under no company folder, or that arrived inside a zip, cannot tell two people apart, and are combined by name as before. To keep two such people apart, give their files different names.
One person assessed at two companies (a move, a second role) is common too, so the question has no default answer.

3. TWO NAMES, ONE PERSON (while names are tidied)

A nickname or a typo gave one person two names. When both names have results for the same test, the question lists each such test on its own row, and you choose which result to keep for EACH test: for example, the EA from one and the WG from the other.

The newer result is chosen for you on every row where both dates are known. A row with an unknown date must be chosen before "Merge as chosen" can be clicked.
Tests only one of the two names has are always kept, and a test whose scores are identical on both is not asked about.
"Keep Both" files the second name separately, followed by "_1".
A name followed by "_1" is never offered as a typo of the plain name. It was kept apart on purpose at question 1, 2 or 3.

8. WHAT THE APP CANNOT READ - AND WHEN IT ASKS YOU INSTEAD#

The app reads the text and the drawn marks inside each PDF. Some reports cannot be read that way, and it tells you rather than guessing:

A SCANNED REPORT (a picture of the page, with no text in the file) is harder. For the cognitive (EA) battery the app reads the scores off the page image itself, and only when it is sure of every number; if it is not, it asks you to type them from the report. For other tests a scanned file is skipped with a warning, and the check at the end of the run lists every result that is missing, so you can continue or stop.
Watson-Glaser I (2008 and earlier) is not supported - see section 1.
A file must have the person's name in its file name ("Last, First - CPI.pdf" or "First Last CPI.pdf"). A file with no name in it cannot be assigned to anyone and is reported instead of guessed.

Values you may be asked to enter by hand, with the PDF opened beside the question:

CPI lifestyle quadrant, when the report does not plot it (stored as the centre of the quadrant you choose - see section 4);
CPI life satisfaction, when its marker cannot be found (section 3);
the EA battery scores, when the report is a scan the app cannot read with certainty;
Watson-Glaser scores, when the report's layout is one the app does not recognise.

Anything you type is used exactly as entered.

Alongside the scores, every result records two facts about its report, so a result can always be traced back to where it came from:

the TEST DATE printed on the report (for the EQ-i, the date on the cover); and
the REPORT TYPE - which version of that test's report it was read from.

If either cannot be read with certainty it is recorded as "unknown", never guessed.

9. THE REPORTS#

Please review provided PowerPoints for all company unique combinations and report assumptions.

PEOPLE ON THE GROUP GRAPHING PLOTS

Each Group Graphing run asks how people should appear on its plots. The choice starts on the default set in Settings ("Group plots, by default"). "Average" and "National Senior Leader" are comparison lines, not people, and keep their names in every option.

1. NAMES SHOWN Each person has their own colour and their own entry in the legend.

2. ANONYMOUS: ONE COLOUR, ONE LEGEND ENTRY

Everyone is drawn in the same colour, and the legend has a single entry, "Participants".

A participants table is added at the end of the report: everyone whose results are in it, which tests each took, and whether they were plotted or counted in the average. It has no colours, so it can be shared with the report.

3. ANONYMOUS: A COLOUR EACH, ONE RAINBOW LEGEND ENTRY

Everyone keeps their own colour, but the legend has a single "Participants" entry drawn as a rainbow, so no colour is tied to a name on the report.

The same participants table is added at the end of the report, without colours.
A separate file, "<report name> - INTERNAL colour key.docx", is saved beside the report. It says who is which colour. Anyone holding it can tell who is who on the plots, so it is for the people who prepared the report only: never share or present it.

4. ANONYMOUS: NAMES LISTED, ONE COLOUR, SHUFFLED The legend lists everyone, but everyone is drawn in the same colour, the names are in alphabetical order, and the people are drawn in a random order. Everyone can see who took part, but no mark can be matched to a name. No table is added.