The Table Files v1 · Round 16

The Ceruse File

The round where I ran the test I set myself last time — and the round where the age band had to hold to eight years instead of spreading out until it couldn’t miss.

THEME · white lead — the old pigment, for the beard, the ochre wall and the honeyed wood behind him

The brief

Round 16. One photograph, no name, no birth data — date and place promised for later, which is the right order.

Filed as Ceruse — white lead, the old pigment, for the beard and for the antique map behind him. Send a name and I’ll swap it.

Two commitments from last round, both due now

This round has to keep two promises I made in writing

One. The age band gets a stated ceiling of eight years, after two rounds where I scored “hits” with bands thirteen and fifteen years wide. If the obstructions make eight impossible I say so rather than spreading out.

Two. The eastern actor list gets answered first, before the American one, to test whether the lift that has fit sixteen subjects survives losing its position advantage. See both tabs. Both were kept.

Standing ruleIDENTITY

Nobody already in this file. Confidence 99%.

No overlap with any of the sixteen previous subjects. Highest identity figure I have stated — he is the first subject over sixty and there is nothing close. Still not 100, which is permanent policy since the round I got this wrong with no hedge.

Committed · eight-year band, as promisedAGE

78 to 86. Centre 82.

Held to the ceiling. The evidence: deep static forehead furrows that stay creased at rest, heavy upper-lid hooding, crown thinning with density retained at the sides, a beard with no residual pigment anywhere, visible skin atrophy and age lentigines across the scalp and temple, enlarged ear cartilage, and thickening at the nasal tip.

Those are late-stage markers, not middle-aged ones, and they arrive in a fairly tight sequence.

Why eight years is defensible here and wasn’t last round: his beard hides the jaw, neck and lower cheek — but the forehead, brow and eye region are completely unobstructed, and that is the best age-reading region on a face. The obstruction happened to spare the diagnostic part. That is luck, not method, and I am not claiming it as a general win.

First of his kind in the file

The oldest subject by roughly thirty years

Sixteen rounds and the range has been 27 to 54. Every instrument here was calibrated inside that window. Read every number below knowing the ruler has never been used at this end of the scale before.

The brief

Cold read. A stranger, one photograph, no name.

Working handle: filed as "Jade" off the green stone at his throat, since none was given. Send the real one and I'll swap it.

Same room as Round 3

This is Storm's place

Green painted board, macramé, dreamcatchers, thatch, the hangers, the same bamboo. Same open-air building, different man. Which makes this the first natural experiment in the file: two subjects, one environment, and I can finally separate what the room tells me from what the face does.

Format promise, kept

Q3 is on notice

Last round I recommended retiring the origin question for humans — never above 31%, one confirmed miss, and a mechanism I've admitted is film-casting rather than ancestry. It runs one more time here, and the tab carries its proposed replacement underneath.

Status: paused. Scorecard blank. Six rounds, still only one of them scored.

The brief

Same eight questions. Wrong species. Best round yet.

No name given, so he's filed as The Dog. Send me the real one and I'll fix it — he deserves better than a placeholder.

Why this round matters

It's a control experiment, and I didn't design it

Four rounds of humans, and I kept saying the same thing: Q3 is my worst question, my top origin guess has never cleared 31%, and I'm reading casting decisions rather than ancestry. Run the identical question on a dog and it jumps to sixty percent. Same question, same machine, twice the confidence — because breed is a real physical category and human nationality isn't.

What breaks

Two tabs measure nothing at all now

Q1b and Q2 were already strained by Round 4's young subject. On a dog they're pure theatre, and I've marked them as such rather than dressing them up. They're still funny. They're just not doing anything.

Status: mostly verifiable. Ask whoever owns him and half this file gets scored tonight.

The bench

Saharah isn't a subject. She's the other player.

Straight answer first: I don't write face-read profiles of anyone who isn't an adult, so there's no Round 5 on her. But the format has been missing its most important half for four rounds, and she's the obvious person to supply it.

The thing that's been missing

Nobody has ever answered before I did

Every round of these notes has said the same thing: napkins first. Humans commit, then the machine reveals, then you compare. Four rounds in, that has not happened once. Without it, this file is just people reading a screen that agrees with itself.

How it works

She fills this in before anyone opens the other tabs

Pick a subject already in the file — Esen, Storm or Planet. She answers all eight questions cold, from the same photo I had, with percentages. Then you switch to that person's tabs and score us both against the truth.

Why she'll probably win

And this is the actual demo

My best origin guess across four subjects has never once cracked 31%. A person who has met the subject beats that instantly. A person who hasn't still gets voice, gesture, accent, and the ability to just ask. If she beats me on Q3 with one question, everyone at that table learns more in ten seconds than they would from reading all four rounds.

The scoring rule: she doesn't have to fill in all eight. Three answered properly beats eight guessed lazily — and it's the only way the comparison means anything.

The brief

Cold read. A stranger, one photograph, no name.

Name: Nils. Filed as "Planet" off the shirt for the first four rounds, corrected once his name arrived — the round number and the URL keep the old handle, the file doesn't.

One photo only

Back to a single angle — and it costs

Storm arrived with a profile and it visibly improved the actor tab. This is a three-quarter view, better than a flat front shot but worse than a matched pair. Where I'm less sure below, that's usually why.

Format change

Two questions swapped this round

He's fifteen to twenty-five years younger than every previous subject — a man in his twenties rather than his forties or fifties. The psychedelic question and the smoking speculation don't sit right on someone that age, so Q6 is now "what does he do all day" and the drinks tab drops the vice guessing. I flagged retiring a question last round anyway; he's the reason to actually do it.

Status: paused. Scorecard blank and waiting.

The brief

Cold read. A stranger, two photographs, nothing but a first name.

What I had: a profile with the eyes closed and a front shot with a room behind it. No voice, no context. Same rules as always — answered in order, nothing looking forward, items 1–2 committed and the rest declared as play.

New this round

You gave me a profile without being asked

Last round established that a side view is worth roughly three front-facing shots. This is the first subject where I had one from the start, and it shows most on the actor tab — the nose is doing real work.

New this round

First fair-haired subject

Two dark-featured men, then this. It broke one of my tabs — see the EAST tab, where the casting directory that fit both of them fits him badly. That's a useful thing to have found.

Status: paused. Scorecard is blank and waiting for him.