Order of exposure: a criterion authored after the evidence is shaped by it

[h4x0r]

The single largest omission in a tool aimed at expert witnesses: nothing in peira knows about
order of exposure.

Every enforced gate examines a claim's content — does it declare a warrant, a falsifier, an
extension. None examines the order in which the examiner learned things, and that order is where
forensic bias has been repeatedly demonstrated. The mechanism the veil of ignorance names — judge the
rule without knowing which side you land on — is the same one Linear Sequential Unmasking applies to
casework: fix the criteria before seeing which answer they favour.

The failure has a signature. A criterion authored after the examiner has seen the evidence is a
criterion shaped to fit it, and it is indistinguishable in the finished record from one authored
before. The claim looks identical; only the sequence differs.

peira is unusually well placed here, because Criterion and Examination are already node kinds and
the 立極 gate already requires a judged_by: edge. What is missing is any notion of when.

Proposed PEIR-CRITERION-BACKDATED — a claim whose judged_by: criterion was authored after the
examination that relies on it. Requires a declared: field on Criterion and Examination, which
is an ordinary frontmatter key.

Design questions worth settling in the issue rather than in code:

• What is the timestamp's warrant? A self-declared date is self-asserted evidence, and the
  claim-grading standard says self-asserted claims need the direction of the assertion stated. A
  hand-typed declared: is exactly as trustworthy as the author. That is still worth having — it
  makes the sequence visible and falsifiable rather than absent — but the gate must not imply
  more than it checks.
• Absent dates must be Unassessed, never a pass. With no declared: on either node there is
  nothing to compare, and the honest verdict is that the ordering was never examined.
• The stronger form, deliberately out of scope for a first cut: masking — a mode that renders a
  claim's evidence without the conclusions reached elsewhere in the vault. That is issue Refutation mode: fresh context is a control, not a personality #17
  (refutation mode) reaching the same place from the other direction.

Source: forensic bias literature on contextual influence and sequential unmasking; the same
mechanism the veil of ignorance names in ethics. See docs/method/README.md §"the six structures"
— this is a seventh, and one the taxonomy does not currently carry.