Stray
OUTLIER (Advanced) — *a point far from the rest is a question with two possible answers: an error to fix, or a real thing to chase.* You investigate its provenance before you decide, and you never silently delete it — because quietly smoothing away the value that doesn't fit is how a true finding, or an honest fault, gets tidied out of the record.
Loading audio…
Press play to listen along. The line being read lights up as you go.
Show full transcript
Loading transcript…
The morning the club smoothed a spike out of their chart, Stray was the only one who flew down to find out what the spike had been before anyone erased it.
He was a crow, sleek and black with a knowing eye, a little lens on a cord at his chest. The club had drawn the pond's water level month by month — a tidy line — except one month leapt far above the rest. "A measuring mistake," someone said, reaching for the eraser. "Smooth it in." Stray hopped between the pencil and the page. "Stop. That point is an outlier — a value sitting far outside the pattern of the others. And an outlier is a question, not a verdict. It has two possible answers, and you can't tell which from the shape alone." He lowered the lens over the spike. "Either it's an error — a mis-copied figure, a slipped decimal, a broken gauge — in which case you fix it and say so. Or it's a real, rare, true thing — a storm, a burst dam — in which case it may be the single most important point on the page. Erase it before you know, and you might be deleting the news."
He named the discipline the academy's way. "Before you touch an odd point, check its provenance — where did it come from, who recorded it, by what instrument, in what units? Provenance tells you error or signal. Only then do you act: correct a genuine mistake, or chase a genuine surprise." He set the lens down. "There's also a quieter reason never to just delete it. A single far point drags a mean around, which tempts people to remove it to make the average behave. But that's backwards — if you're worried one point can swing your summary, that's a reason to use a robust measure like the median and to investigate the point, not to make the point vanish so the tidy number survives."
Stray had learned both halves in the rook-parliament of the tall elms. His mother was the one the flock sent toward anything strange — a scarecrow that moved, a field that glittered. "Everyone else is staying put," young Stray once called as she flew toward a distant flash. "Because everyone else decided it was nothing without checking," she called back. "The odd thing is where the news is." She returned having found a spill of grain from a broken cart, a feast the flock had nearly shrugged off. But she taught him the other half too: another day a "strange" glint was only broken glass, nothing at all. "Sometimes the odd thing is only odd," she said. "So you look — and then you know. You don't keep it or throw it before you've seen what it is." Years later, over a chart with a spike instead of a field with a flash, Stray would think: the point that doesn't fit is a question — go and answer it before you decide its fate.
In his workshop a mole named Digby brought a chart of test scores. "Everyone got around fifty, except one ninety-eight. I dropped it — it was throwing off my average." Stray raised the lens. "Dropped it — before checking where it came from?" "It didn't fit." "Who got the ninety-eight?" Digby checked. "A new student. Joined last week." "So your outlier is a real creature who really answered," Stray said gently. "Now the question splits. Maybe she'd studied this exact thing before — a true, fascinating point. Maybe her paper was marked wrong — an error to fix. Maybe a digit got transposed in your copying — also an error. You erased her before you knew which, purely because she made the line untidy." He nudged the lens toward Digby. "Fix it if it's an error. Chase it if it's real. Report it either way. But never delete a point just for standing apart — standing apart is a clue, not a crime."
Digby rubbed out the crossing-out and went to ask her, and came back to say she'd simply loved the subject for years — a real, honest, high point, no error at all. "I nearly erased the best thing on the page," he admitted, "for the crime of making my chart neat." "That's the danger of tidiness," Stray said. "It's so satisfying we forget it can hide the truth — or the fault. A quietly deleted outlier is a decision no reader ever gets to check." He tapped the restored score. "Now the surprise that made this chart worth reading is still on it."
He let the lens fall against his chest, and something in him settled — not smooth, exactly, but bright and awake, the pleasure of a mystery met instead of erased. Over the elms the rook-parliament would wheel and call, always sending one of their number toward whatever didn't fit, and Stray felt keen and glad, certain that the point standing apart is the one that most repays a closer look, and that the record must always show it was asked, not vanished.
The DataForge ensemble
Stray is part of DataForge's distributed-narrative cast. Each character embodies a different curricular primitive; together they teach the full subject.
-
Catch
Data collection — who-what-why-when posture (every dataset has a collector + purpose + omissions)
-
Tidy
Data cleaning — preparation-with-integrity posture (every cleaning choice changes meaning; document the choices)
-
Graph
Data visualization — shape-of-the-story posture (which chart tells the truth, not the loudest one)
-
Tell
Interpretation — correlation-not-causation posture (data shows patterns; humans interpret; confidence not certainty)
-
Guard
Data ethics — bias-privacy-harm-consent posture (who benefits, who's harmed, who decided; structurally present in every kit from kit 6)
-
Crux
Summary — honest-middle posture (which middle tells the truth: mean, median, or mode; one giant value drags the average away from where most of the crowd actually sits)
-
Cull
Representative sampling — fair-sample posture (a sample must mirror the whole; who gets left out silently bends the story)
-
Ladle
Rates & fair comparison — like-for-like posture (compare rates and fair portions, not raw totals; a bigger pot isn't a bigger share)
-
Waver
Uncertainty & margin — confidence-not-certainty posture (every estimate carries a margin of error; show the wobble honestly instead of hiding it)