Created with high confidence
On round 12, evidence-evaluator scanned the Q1 audited accounts looking
for fragments that satisfied req_capital_minima. It found one. A direct, quotable
line. The figure exceeded the GIBFIN 12.4 threshold by a comfortable margin.
Result: status satisfied, confidence high,
source_verified_at stamped. The Requirement turned green.
A new version arrives
A month later the finance team uploads the Q2 audit. The Document's
pending_upload_at populates. The Evidence row hasn't changed — but its
source_verified_at is now older than the document version it was attached to.
Nothing has happened yet. But on the next round, this row is on the work list.
Re-evaluation
evidence-evaluator picks it up. Re-reads the source — now the Q2 accounts. Finds the same
section, the same shape of claim. Updates fragment_snapshot to the new wording,
bumps source_verified_at.
If the figure had held, the row would stay satisfied and the round would move on.
But this time…
Quiet trouble
Capital dropped. £500k in Q1, £280k in Q2 — below the GIBFIN 12.4 floor of £350k. The
evaluator flips status to concern, drops confidence to medium,
and writes a concern_text that surfaces in the next round note.
The Requirement that was green is now amber. The CFO will see this tomorrow morning.
Why this matters
Evidence isn't a one-shot stamp. Documents change. Numbers move. Regulators amend their guidance. An application that was sound in April can be broken by July.
Confidence isn't decoration. It's a re-check budget: high evidence is sampled less
often, low is revisited every round. The source_verified_at timestamp
and the source document's pending_upload_at tell the evaluator exactly what to
revisit. superseded_by chains stale claims to their fresh replacements.
Drift gets caught. The audit trail is intact.