Over-Claim Guard

Conservative-wording reference for English papers. The goal is not timid prose — it is to state evidence strength precisely: strong evidence earns strong wording, weak evidence gets weak wording.

Boundary with claim-evidence-contract.md

• claim-evidence-contract.md decides whether the visible evidence supports a claim (the strength ladder unsupported → observed → supported → strong).
• This file decides how to word a claim once its strength is known (the verb/qualifier to pick so the sentence does not outrun the evidence).

Use them together: the contract sets the ceiling, this guide picks the wording under it. Where they appear to conflict, the contract wins (it is about substance, not phrasing).

Certainty ladder (verbs, strongest → weakest)

demonstrate / prove                         ← intervention + controlled experiment
  ↓
reveal / identify / find                    ← strong effect, multi-method or replicated
  ↓
indicate / suggest                          ← significant but single-method
  ↓
support / be consistent with                ← trend, agrees with prior work
  ↓
may indicate / could suggest / appear to    ← marginal or predictive
  ↓
hint at / point toward                      ← very weak signal or hypothesis

Pick the rung that matches the evidence. Do not climb a rung the data cannot reach.

Substitution tables

1. Causal (the most common over-claim: correlation stated as causation)

❌ over-claim	✅ conservative
caused by	associated with / linked to
drives / driving	contributes to / is associated with
determines	influences / shapes
responsible for	implicated in / associated with
results in	is followed by / co-occurs with
proves that	indicates / provides evidence that

Causal wording is allowed only with a controlled intervention (ablation, randomized assignment, A/B test), an instrumental-variable design, or an already-established mechanism your data reproduce. Otherwise use association wording.

2. Novelty / firstness (reviewers verify these in seconds)

❌ over-claim	✅ conservative
the first to	the first, to our knowledge / among the first to
novel (self-labeled)	name what is new; drop the label
unprecedented	substantial / notable
previously unknown	not extensively studied

3. Universality (one setting studied, all settings claimed)

❌ over-claim	✅ conservative
always / never	generally / rarely
in all cases	in the cases studied
universally	across the benchmarks evaluated
any dataset	the datasets sampled

4. Effect size (vague magnitude word with no number)

❌ over-claim	✅ conservative
strong improvement	reduces error by X%
large effect	β = X.XX (95% CI: …)
significant gain	improved from X to Y (p = …)
highly significant	p < 1 × 10⁻¹⁰
robust	consistent across N runs / stable under [perturbation]

If the number itself carries the weight, drop the adjective — the number speaks.

5. Temporal / inferred order (present data, past mechanism)

❌ over-claim	✅ conservative
X drove the change	the change is consistent with X
occurred at time T	estimates suggest ~T (CI: …)
migrated from A to B	the data are consistent with a path A→B

6. Application / impact (downstream uses not demonstrated here)

❌ over-claim	✅ conservative
will revolutionize	has potential implications for
will be widely used	may be useful for / could inform
solves the problem of X	addresses one aspect of X
ready for deployment	provides a candidate approach for [setting]

7. Comparison (disparaging prior work)

❌ over-claim	✅ conservative
previous methods failed to	previous methods were limited by
outperforms all prior work	compares favorably with [specific methods]
resolves the long-standing debate	adds evidence to one side of the debate

High-frequency trap phrases

trap	safe replacement
"Our results demonstrate X." (X causal)	"Our results are consistent with X."
"This is the first work to …"	"To our knowledge, among the first to …"
"X plays a critical role in Y."	"X has been implicated in Y / may contribute to Y."
"These findings have important implications for …"	"These findings provide a basis for further study of …"
"X is a key driver of Y."	"X is associated with Y."
"Strongly supports"	"Is consistent with / provides evidence in line with"

Reverse calibration: when NOT to hedge

Hedging weak evidence is right; hedging strong evidence is timid. Use strong wording when:

• a controlled intervention (ablation / RCT / A-B) gives a causal result → demonstrate;
• multiple methods / datasets / seeds replicate the result → robustly, with the evidence named;
• an established mechanism is reproduced → confirms / validates;
• a large effect with a strong statistic → strong wording plus the number.

Self-check (scan after drafting a paragraph)

• [ ] Used first / novel? Did you actually search the literature, or add "to our knowledge"?
• [ ] Used cause / drive / determine? Is there an intervention? If not → associated with.
• [ ] Used all / always / universally? Is the scope bounded to what you studied?
• [ ] Used significant / strong / substantial? Is a number attached?
• [ ] Listed implications you did not demonstrate? Add may / could.
• [ ] Disparaged prior work? Reframe to "limited by", not "failed".

Script support

deai_check.py flags a focused set of unambiguous over-claim phrases (causal / firstness / universality / application) as [Script] LOW traces and points back to this guide. The script is a safety net for the obvious cases; the tables above cover the judgment calls it cannot make.