Seven useful epistemic categories
Your private workflow uses a distinction that should stay visible throughout the writing process. It separates seven kinds of statement:
- Direct observation: what you personally saw, heard, measured, recorded, or experienced.
- Reported fact: information supplied by a source or record.
- Interpretation: what you think an observation or fact means.
- Inference: a conclusion drawn from evidence that does not state it directly.
- Assumption: something the argument requires but has not established.
- Speculation: a plausible possibility worth considering but not claiming.
- Conclusion: the position you are prepared to defend given the present evidence.
Writers routinely collapse these categories. "Sales fell after we changed the pricing" can be an observation. "The pricing change caused sales to fall" is a causal inference. The second sentence requires evidence the first one does not.
Claim strength must track evidence strength
Compare these phrases:
- "This proves…"
- "This suggests…"
- "This is consistent with…"
- "One explanation is…"
- "I suspect…"
- "I do not yet know whether…"
These phrases are not stylistic decoration. They are measurements of confidence.
One common editing error is to remove uncertainty because hedged prose feels weak. Another is to hide uncertainty in vague language until the claim cannot be challenged.
The better standard is precise confidence. State clearly what you know. State clearly what you do not know. Do not use fog to occupy the space between them.
Mechanisms beat labels
Framework language is tempting because it compresses thought. "Confirmation bias," "culture," "incentives," "network effects," "trust," "leadership," and "systems problem" can all be useful terms. They can also stand in for explanation.
When a load-bearing term appears, remove it for a moment and ask:
What is physically, economically, socially, or psychologically happening here?
Instead of "There was a culture problem," you may discover an actual mechanism:
- Supervisors learned that reporting bad news delayed their incentives.
- Errors were corrected privately, so nobody saw recurring patterns.
- Managers were measured on speed, not reversibility.
- The person with authority did not have the information required to use it well.
Once the mechanism is visible, the label may return. Now it names something rather than concealing it.
Find the smallest defensible claim first
Before strengthening an argument, find the smallest version you can actually defend.
If the broad claim is:
"AI makes knowledge workers more productive."
The smallest defensible claim may be:
"For tasks with clear evaluation criteria, AI can reduce the time required to produce a usable first version."
The smaller claim may feel less exciting. It is often more useful because the conditions are visible. Then ask whether evidence supports a stronger version.
Ask which answer you are hoping for
Your Private Thinking Grill contains an unusually important question:
What answer am I hoping the inquiry supports?
That question exposes your stake before the evidence is organised around it. Follow it with these:
- What answer would be inconvenient for me?
- What evidence would I be tempted to dismiss?
- What would a person with different incentives notice?
- What condition in my experience may not exist elsewhere?
- What fact would force me to change my mind?
A position becomes conviction only after you know how it could fail. Module 4: From raw thought to a defensible private position runs the full Private Thinking method.
Every reusable principle needs a boundary
Every reusable principle should answer four questions:
- When does it apply?
- When does it stop applying?
- What enabling conditions does it depend on?
- What stronger claim am I explicitly not making?
This matters especially when you import Western business models, biographies, or management ideas into Indian conditions. The question is not whether an idea is "universal". The question is which mechanism is universal enough to transfer and which conditions are local.
Use the first person to attribute judgment, not to decorate prose
The Howard Marks material sharpens a distinction already present in this course. The first person can tell the reader what kind of statement they are reading:
- Fact: the record shows X.
- Interpretation: I think X happened because Y.
- Expectation: I expect Y to make Z more likely.
- Uncertainty: I do not know whether the effect persists outside these conditions.
- Judgment: in my view, this trade-off is acceptable.
This is different from scattering "I think" through a piece to sound conversational. The first person earns its place when it assigns ownership to an inference or judgment that should not masquerade as fact.
To audit this, remove every "I think" for a moment. If a sentence now sounds like an established fact, ask whether the first person was doing necessary epistemic work. If it was merely softening a claim you are prepared to defend, remove it. Precision is the goal, not timidity.