Concern Hypothesis

IV node standard published
Published 2026-07-11

Type: Node / Bounded Interpretive Signal

Working Definition: A concern hypothesis is a provisional, bounded, low-harm interpretation that a human concern — affecting cognition, agency, participation, or development — may be present, offered as a prompt for human review rather than a conclusion about the person.

Function in the Grammar: Concern Hypothesis is how a governed system may notice something worth a human's attention without pretending to diagnose, judge character, or score risk. It is a specialized Claim type — carrying the same epistemic discipline every claim does, but bounded by an explicit, non-negotiable list of prohibited conclusions. Where an ordinary hypothesis proposes an explanation for observed data and invites testing, a concern hypothesis proposes only that a human concern may be present and invites human review. This is one of the sharpest edges in the whole grammar between assistance and harm: a system that can notice is useful; a system that concludes is dangerous. The governing wiki states the boundary directly — a concern hypothesis must never become a diagnosis, character judgment, hidden profile, stigma engine, risk score, or coercion excuse.


Formal Pattern

ConcernHypothesis(h, subject: p, observed: E, interpretations: I) iff
  h has ObservedEvidence(E): the specific, disclosed basis for the hypothesis
  h has PossibleInterpretations(I): more than one candidate reading, not a single verdict
  h has EpistemicProfile:
    derivation: inferred
    producer: human | model | hybrid
    grounding: source-grounded | partially-grounded
    review_stage: provisional until human review
    warrant_status: unchallenged | challenged
  h has AuthorityBoundary(a): who may act on h, and what they may not do
  h has PrivacyBoundary(pb): what may and may not be exposed, inferred, or retained
  h has DignityImpact(d): an assessed effect of raising h on the subject's dignity
  h has RecommendedSupport(r): low-harm support option(s), if any
  h has EscalationConditions(c): explicit conditions under which h is escalated, if any
  h has ProhibitedConclusions(x): {diagnosis, character judgment, hidden profile,
    stigma engine, risk score, coercion excuse} — always populated, never empty
  h has ReviewPath(v): the human review h must pass through before any action follows

RAISES(Agent(a), ConcernHypothesis(h), subject: p)
REVIEWS(HumanReviewer(r), ConcernHypothesis(h))
ESCALATES(ConcernHypothesis(h), to: Institution(I))   -- only if EscalationConditions(c) are met
RETIRES(ConcernHypothesis(h))                          -- reviewed, found not to warrant action

Required Elements

A concern hypothesis that omits any of these is not properly formed — it is either an unbounded suspicion or a disguised conclusion:

  1. Observed evidence — the specific thing that was actually seen, said, or recorded, not a summary impression
  2. Possible interpretations (plural) — a hypothesis with only one candidate reading has already collapsed into a verdict
  3. Epistemic profile — derivation remains inferred, and review stage remains visibly provisional until human review; institutional acceptance never turns the hypothesis into a diagnosis
  4. Authority boundary — who may act on it, and the explicit limit on what that authority permits
  5. Privacy boundary — what may and may not be exposed, inferred, remembered, or shared further
  6. Dignity impact — an assessment of what raising this hypothesis itself does to the subject
  7. Recommended low-harm support — an offer, not a mandate
  8. Escalation conditions, if any — named in advance, not improvised after the fact
  9. Prohibited conclusions — the explicit list of things this hypothesis must never become
  10. Review path — the human review required before the hypothesis can lead to any action

Examples

Domain Observed evidence Concern hypothesis (bounded) What it must not become
Digital governance / personal recovery app A user's entries show a sudden drop in engagement and shorter, more withdrawn language "This pattern may indicate the person is struggling; a gentle check-in may help" A depression diagnosis, a risk score, or an automatic alert to a third party
Education A student's submissions show a marked change in tone and quality over two weeks "This change may indicate something is affecting the student outside coursework; a conversation may be warranted" A disciplinary character judgment or a permanent file annotation
Workplace / HR An employee's communication pattern shifts noticeably after a reported life event "This may be affecting the employee's capacity right now; support may be appropriate" A performance-risk flag used in review decisions
Ordinary life A friend has been unusually quiet and has canceled plans repeatedly "Something may be going on; it may be worth asking directly" A private diagnosis you never voice, treated as settled fact in how you now relate to them

Distinctions

Concern Hypothesis ≠ Diagnosis. A diagnosis is a conclusion, typically made by a qualified authority, that classifies a person's condition. A concern hypothesis is explicitly not a conclusion — it is a bounded, provisional signal that invites human review. The moment a concern hypothesis is treated as settled, it has become the thing it is constitutionally forbidden to be.

Concern Hypothesis ≠ Ordinary Hypothesis. An ordinary Claim of type "hypothesis" proposes an explanation for observed data and is typically tested toward acceptance or rejection. A concern hypothesis is never tested toward "acceptance" in that sense — its only legitimate resolution is human review that either recommends support, escalates under pre-named conditions, or retires the hypothesis. It does not graduate into a fact about the person the way an ordinary hypothesis may graduate into an accepted claim.

Concern Hypothesis ≠ Risk Score. A risk score is a quantified, typically opaque, output that ranks or classifies a person for downstream automated or semi-automated decisions. A concern hypothesis carries no score, ranking, or classification — only observed evidence, possible interpretations, and a review path. Any implementation that assigns a number to a concern hypothesis has built a risk score wearing a concern hypothesis's name.

Concern Hypothesis ≠ Failure-of-Fit. Both are provisional, bounded signals rather than conclusions, and both must generate a review path rather than dead-ending. But a Failure-of-Fit names a gap in the grammar itself — the categories cannot represent the situation. A concern hypothesis names a possible gap in understanding a person — the categories may be adequate, but something about this specific person may need attention. Confusing the two risks either treating a person as a grammar problem, or treating a genuine grammar inadequacy as if it were a fact about someone's condition.


Common Failure Modes

Mode Description
Diagnosis creep A concern hypothesis is treated, in practice, as if it had already been confirmed — acted upon as settled fact rather than routed for review. Where to look: language or downstream records that state the hypothesis as established rather than provisional.
Single-interpretation collapse Only one interpretation is offered, eliminating the "possible interpretations, plural" requirement and turning the hypothesis into a verdict in every respect but name. Where to look: concern hypotheses with exactly one stated reading and no acknowledged alternative.
Hidden profiling Concern hypotheses accumulate into a durable, retained profile of a person across many instances, without the privacy and retention boundaries the mechanism requires. Where to look: systems that store concern-hypothesis history in a way indistinguishable from a persistent character file.
Scored concern A concern hypothesis is assigned a numeric weight, tier, or ranking used in downstream automated decisions — reintroducing a risk score under a different name. Where to look: any quantification attached to a concern hypothesis.
Missing review path A concern hypothesis is raised, and some action follows, without the required human review ever occurring. Where to look: concern-hypothesis-triggered actions with no reviewer of record.

Cross-References

Cross-References

This entry is AI-assisted. Reviewed by the human author before publication.