PART OF 1-1-3: PEACE · STEP ONE1-1-3: PEACE · THE PROPOSED PATH · RUNG ONE1-1-3: PEACE · BOOTSTRAP TOOLS · RUNG ONE

The Ladder

Smart tools that help people get smarter. They make guesses, and the real world tells us if they were right. Tools that are not alive, raising the person who uses them. Narrow, goal-free tools whose predictions are sealed in advance and scored against reality and stronger baselines.

The big idea: we don't build a robot to be smart for us. We build helpers that make us smarter, one step at a time, like climbing a ladder. “The bootstrap is a set of tools, not a creature. Each is superhuman at one job: designing, predicting, checking. None has goals of its own, and none acts unless a person asks.” “The bootstrap is a set of tools, not a creature. Each is superhuman at one job: designing, predicting, checking.” Rung one tests that claim with pre-registered forecasts and a tamper-evident record.

…
I

Meet the helpersThree tools and a LockerThree tools and a Locker

There are three helpers and one safe box. Each helper is great at just one job. None of them is alive, and none of them does anything unless a person asks.

Each tool is a first, working rung toward a power from your thesis. None of them acts on its own. Each one does one job, and the Locker keeps them honest.

Each component is narrowly capable, has no goals, and runs only when a person asks. The Locker makes every claim tamper-evident, so no result can be quietly adjusted after the fact.

Blueprints and Gifts power art
Designer
Tries about a thousand paper helicopter shapes in a blink and picks the best one.Searches thousands of shapes and picks the one that best meets your goal.Exhaustive search over ~1,000 stability-constrained geometries; maximizes predicted fall time or minimizes distance to a target time.
RUNG ONE OF · BLUEPRINTS & GIFTSToday it designs paper helicopters.Designs a paper helicopter that hits the flight you ask for.Desk proving ground; generalizes to any simulate-then-verify design problem.
What-If Sight power art
Predictor
Guesses what will happen before it happens.Says what will happen before it happens, with an honest range.Point forecast plus an 80% interval; versions are frozen and sealed before use.
RUNG ONE OF · WHAT-IF SIGHTToday it guesses space weather and how long a paper helicopter falls.Calls Earth's space weather and the helicopter's fall time in advance.Kp 3-hour blocks (Newell coupling + persistence) and descent time (one-parameter drag model).
Bias Check power art
Checker
Checks every guess against what really happened. No fudging allowed.Scores every guess against reality and against simpler rivals.Scores against observed outcomes and against baselines: NOAA SWPC's forecast, persistence, and the human.
RUNG ONE OF · BIAS CHECKToday it grades our helper against real scientists, and against you.Grades the tool against NOAA, plain repetition and you.Mean absolute error per source; interval coverage for desk rounds.
Perfect Memory power art
The Safe BoxThe LockerThe Locker
Once a guess goes in, nobody can change it. Not you, not us, not anyone.Seals every guess with a fingerprint. Nothing can be quietly changed later.Append-only log; each entry's SHA-256 covers its content and the previous hash, so any edit or deletion fails verification.
RUNG ONE OF · PERFECT MEMORYA memory that can't be changed or faked.A record where history can't be rewritten.Internal integrity only; public anchoring of the head hash is the next step.
II

Sky: guessing space weatherSky: calling space weatherSky: geomagnetic Kp forecasting

The Sun is always blowing a wind out into space. A spacecraft far away feels it first, about an hour before it reaches Earth. Every 3 hours, our helper guesses how much that wind will shake Earth's magnetic field. The scale goes from 0 (calm) to 9 (a huge storm with bright northern lights). Scientists at NOAA, America's weather and space-weather agency, make their guess too. Then the real number comes in, and we see who was closest.

Right now, a spacecraft 1.5 million km away is measuring the solar wind before it reaches Earth. Every three hours the Predictor seals a guess for how stirred-up Earth's magnetic field will get (the Kp scale, 0 calm to 9 extreme storm). Then NOAA's measurement scores it. Its rivals are NOAA's own forecasters, and a lazy guess that just repeats the last value.

Real-time solar wind (plasma and IMF in GSM) from the active L1 spacecraft via NOAA SWPC RTSW. For each 3-hour UT block, a Kp forecast is sealed before the block starts. It blends the Newell et al. (2008) coupling relation, averaged over the last 60 minutes, with persistence. The SWPC forecast and persistence are sealed at the same moment as baselines. Everything is scored by absolute error against SWPC's preliminary planetary Kp.

The Sun's wind over the last dayThe last day of solar wind24-hour driver vs. observed Kp

Every locked-in guessEvery sealed callForecast ledger

III

Desk: the paper helicopterDesk: the paper helicopterDesk: rotor descent forecasting

You can do this one at home! All you need is paper, scissors, a paper clip and a phone timer. The helper picks a shape. You guess first: how many seconds will it take to fall? Then the helper's guess gets locked in too. Drop it, time it, and see who was closer. Every time you play, the helper learns and gets a little better.

A test you can run at your desk with paper, scissors, a paper clip and your phone's stopwatch. The Designer picks the shape. You guess first, then the Predictor seals its own guess. You drop it and the Checker scores you both. After every drop the tool learns, so its next guess should get closer. That's the ladder you can watch.

A physical forecasting task with a human comparison arm. The model is steady descent: mg = ½ρCdAv² with blade area A, and t = h/v + v/2g. It has one free parameter, ln Cd, with a normal prior that each round's implied Cd updates. The human forecast is sealed before the model's output is shown.

Every roundEvery roundRound ledger

IV

The Ladder: are we getting smarter?The Ladder LogThe Ladder Log

This is the big question: do these helpers really help a person do things they couldn't do before? We only write a step here once the real world proves it.

The thesis claim, measured: with the tools, can a person do what they couldn't before? Rungs are only written from sealed, scored results, plus your own sealed notes.

The thesis's central empirical claim: does tool-assisted human performance exceed unassisted performance on held-out, pre-registered tasks? Rungs are derived only from resolved, sealed records.

Steps we've provenRungs, from the recordDerived rungs

Add your own stepAdd a rung of your ownAdd a rung

Write down something you can do now that you couldn't do before. It gets locked in with the date and time.Something you can now do that you couldn't before. It gets sealed in the Locker with a timestamp.A self-reported capability claim, sealed with timestamp and optional evidence pointer.

V

The Safe BoxThe LockerThe Locker

Every guess and every answer goes in here, in order. Each one is locked to the one before it, like links in a chain. If anyone changes even one letter, the chain breaks and this box turns red.

Every model, guess and result in one chain. Each entry carries a fingerprint of itself and of the one before it. Change a single character anywhere and the check below turns red.

Append-only JSONL; hash = SHA-256(canonical JSON of {seq, type, at, body, prev}); verification recomputes every hash and link in your browser. External anchoring via OpenTimestamps establishes when.

The public recordThe public recordPublic chain (sky)

Your own recordYour own recordPersonal chain