Answer
How should evidence be separated from interpretation?
Structurally, so a reader can see which is which without judging. Mixed in prose, the two become one confident paragraph.
By structure rather than by phrasing. Put what was observed in one place with its source, and what it means in another. In continuous prose a reader cannot tell an observation from a conclusion, and both inherit the same confidence.
The problem is not that systems interpret; interpretation is usually what was wanted. The problem is that a sentence reporting what a document says and a sentence concluding what that implies are typographically identical, so a reader assessing reliability has no signal to work from. The inference inherits the credibility of the observation next to it, and the combined paragraph is treated as established.
Separation by wording does not hold. Hedging words are supposed to mark the inference — suggests, indicates, appears — and they are distributed by register rather than by epistemic status, so they appear on solid observations and are absent from speculative leaps. Anything that depends on the writer choosing the right verb will fail at exactly the moment the writer is confident about something they should not be.
Structural separation does hold, because it is checkable. A findings section containing only what was observed, each item with its source. A separate section containing what follows from it. A reader can then verify the first without engaging with the second, and can disagree with the second without doubting the first. This is the arrangement used in any discipline where the distinction matters, and it works for the same reason.
It also produces a useful discipline on the way in. Writing the observation section forces the question of what was actually seen, and that question frequently reveals that a load-bearing statement is a conclusion with no observation under it. That discovery is the main practical benefit, and it happens during production rather than during review.
For automated systems the equivalent is a schema. A field for the value and a field for the source; a separate field for the conclusion. This is stricter than prose sections and it removes the possibility of the two being mixed, which prose does not: a findings section can contain interpretation if nobody is checking, and a field cannot contain two things.
One caveat about how far to take it. A document that separates every clause becomes unreadable, and readability is not a luxury when the point is for someone to check the work. The separation needs to be at the level of the claim rather than the sentence, and the test is whether a reader could list what was observed after one pass. If they could, the structure is doing its job.
An observation and an inference set in the same paragraph become the same kind of statement, and the weaker one is carried by the stronger.
Siddharth Sharma, Context Theory
Related questions
Where does an assumption sit?
In a third place, and it is the one most often left out. An assumption is neither observed nor inferred from what was observed; it is something taken as given so the reasoning could proceed. Listing assumptions separately with what each would change if wrong is the highest-value part of this structure, because that list is what a reader can actually act on.
Does this apply to a short answer?
In compressed form. Even a paragraph can distinguish the recorded value from what it means with a single phrase — the record shows, which suggests — provided the phrasing is consistent rather than decorative. The structure matters more as the output grows, and short answers benefit mostly from citing the specific source of each fact.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Close rate — response under 5 minutes vs over 24 hours | 32% vs 12% | Category-wide |
| AI-cited sources that also rank in the Google organic top 10 | 10% | Category-wide |
Optifai speed-to-lead benchmark · n=939 companies · Q2 2025–Q1 2026 · verified
2026 generative engine citation study · fewer than · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Workflow | An observation and an inference set in the same prose are typographically identical, so the inference inherits the credibility of the adjacent observation and the combined passage is treated as established. | Asking a reader of a mixed passage to mark which statements were observed and which were concluded. |
| Response | Hedging words are distributed by register rather than by epistemic status, appearing on solid observations and absent from speculative conclusions, so any separation depending on verb choice fails where the writer is wrongly confident. | Checking whether hedged statements in a sample of outputs are actually the less well-supported ones. |
| Software | Writing the observation section first frequently reveals that a load-bearing statement has no observation beneath it, which makes structural separation a production discipline rather than only a presentation choice. | Extracting the observed statements from an existing analysis and checking which conclusions lose their support. |
| Constraint | A schema enforces separation in a way prose sections cannot, because a findings section can contain interpretation when nobody checks while a typed field cannot hold two kinds of content. | Inspecting a prose findings section for statements that are conclusions rather than observations. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one