AI & Research Framework · Lesson 04 of 05
Writing a thesis you can score
What it is
Lesson 01 gave you the loop. This lesson gives you the artifact — the short written document that makes the loop real.
A scoreable thesis is one page with seven fields:
- The claim. One sentence: asset, direction, and the mechanism — why this should happen, who gets forced to buy or sell.
- The horizon. Days, weeks, or months. Pick one.
- The evidence. Both readings, with sources — what supports it, and what's against it right now.
- The invalidation. The observable condition that kills it. A level, a close, a behavior — something a stranger could check.
- Confidence. A word — low, medium, high — plus one sentence of why. (Not "73%." You'll see why below.)
- The counter-case. The strongest argument you're wrong, written like you mean it.
- The grade. Left blank. Reality fills this one in.
If those fields look familiar, they should — it's the anatomy of every publication on this site, taught as a personal practice. And the reason it's written matters as much as what's written: a vague idea can hide comfortably in your head forever. On paper, with a date on it, vagueness has nowhere to stand. Writing is thinking made checkable.
Why it matters
Unwritten theses mutate. Memory quietly rewrites them after the fact — the entry gets earlier, the doubt gets smaller, the near-miss becomes "basically called it." (Next lesson is entirely about why your brain does this.) A written, dated thesis is the only version of your past thinking that can't be edited by your ego.
Stack twenty of them, graded, and you own something priceless: a private track record. It will teach you things no course can — that your High-conviction ideas hit twice as often as your Lows (or embarrassingly, don't), that your macro reads outperform your alt picks, that your invalidations are chronically too tight. That's calibration, and it's the actual difference between year-one and year-five researchers. Twenty graded theses teach more than two hundred vibes.
Two practical bonuses. Risk falls out for free: the distance to your invalidation is your risk per unit — sizing becomes arithmetic instead of mood. And critique becomes possible: a one-page thesis can be handed to a friend, or to the AI red team from last lesson, and attacked productively. Nobody can usefully critique a feeling.
The two readings, always taught together
Read bullish when
- Marks of a thesis built to survive contact with reality: The claim fits in one sentence and names its engine. "SOL resolves this three-week compression upward because spot is leading, funding is flat, and a raid on the lows already failed." Asset, direction, mechanism — attackable, therefore useful.
- The horizon matches the evidence's timeframe. A weekly-structure thesis gets weeks-to-months, not "by Friday" (Market Structure 06's rule, now in writing).
- The invalidation is observable and well-placed. Beyond the structure, outside the pools (Market Structure 02), phrased so a machine could check it: a daily close below X, not "if it starts looking heavy."
- Confidence is a word with a reason. "Medium — the setup is clean but the macro dial is against it." Honest, calibratable, human. The counter-case has real teeth. If your strongest opposing argument sounds like a strawman, you haven't found it yet — you've decorated your confirmation.
Never alone — confirm with the counter-case & the scoring sheet
Read bearish when
- Smells of a thesis that was doomed at the keyboard: No horizon. "Eventually" is not a timeframe; it's an escape hatch. Unfalsifiable by construction (Lesson 01's horoscope, in business casual). A missing or renegotiable invalidation. If wrongness has no address, the thesis can never die — it can only be abandoned quietly, teaching nothing.
- One-sided evidence. Five confirming signals, zero opposing ones. Markets never look like that; only evidence-shopping does.
- Certainty theater. Expressed as leverage, all-caps, or fake-precise percentages. Real conviction is a word and a reason; everything louder is usually compensation.
- Scope creep. "…and ETH will follow, and macro will turn, and alts will…" One claim per thesis. A document that predicts everything grades nothing. Written after the entry. Then it's not research — it's a press release for a position you already hold. The loop runs forward or it doesn't run.
Never alone — confirm with the counter-case & the scoring sheet
Visual explanation
Real market example
July 2021, written as it could have been written. A retrospective exercise — labeled as such — on ground you know from Market Structure 03. Put yourself at July 26, 2021: the $30,000 floor of the summer range was raided days earlier, the most telegraphed breakdown in crypto failed, and price has reclaimed the range.
The template, filled honestly with what was knowable that day: Claim: BTC travels toward the top of the range ($40,000+) because the breakdown trapped sellers whose covering now fuels upside, and negative funding shows the crowd still leaning short. Horizon: weeks. Evidence for: failed breakdown reclaimed within a day; funding negative (Derivatives 02's July squeeze); the range floor absorbed its raid. Against: summer volume thin; the broader "bull market over" mood. Invalidation: a daily close back below the reclaimed low — if the raid resumes, the reclaim was a trap. Confidence: medium — structure clear, sentiment hostile. Grade — filled in by history: hit, and then some; the range top fell within weeks and new highs printed by November.
The honest coda: the same template, applied all through 2021, would have graded plenty of theses invalidated — the point of the practice isn't that it produces winners. It's that it produces answers. Every filled grade column, hit or miss, is tuition actually collected. The vibes-trader who lived that same summer collected none.
How RIX Intel uses this signal
Directly — this template is the publication format. The Read is the claim plus the confidence word and its reason. The horizon is a required field. The recorded plan freezes entry, invalidation, and targets. The counter-case is mandatory. And the Track Record is the grade column, filled in public, hits and misses in identical type. The publish checklist mechanically rejects anything missing a field — the craft in this lesson, enforced by software.
Which makes this the curriculum's graduation assignment: write one scoreable thesis this week. Not to trade it — to grade it. Use any signal from any shelf. Date it, file it, and let reality mark it. The private version of what this site does in public — that's the practice, and it starts with one page.
Common mistakes
Where this signal ruins people
Writing the thesis after entering the position. That's a justification with a template on. The loop runs forward: thesis, then action. Order is the whole integrity of the exercise.
Fuzzy invalidations. "If it starts looking weak" is a mood, not a condition. If a stranger — or a script — couldn't check it, it isn't an invalidation yet.
Percent-confidence theater. "73% sure" is false precision wearing a lab coat — you don't have the distribution, and the decimal is a costume. A word and a reason is honest, and honesty calibrates; theater doesn't.
Never filling the grade column. The sheet only teaches if it's scored — including, *especially*, the losers. An ungraded thesis archive is scrapbooking. The grades are the tuition receipts.