amplifying-an-existing-number-is-a-provenance-recheck-trigger

name	amplifying-an-existing-number-is-a-provenance-recheck-trigger
description	A request to ELEVATE / amplify / feature an existing number that's already in a deliverable — turn a prose figure into a chart, make it a headline, restore "the breakdown we had" — is itself the trigger to re-verify that number's provenance BEFORE amplifying it. Use when: (1) asked to add a visual/chart for a number that already exists as text, (2) asked to make an existing stat more prominent, (3) a stakeholder says "we had this insight, put it back / show it bigger", (4) a number was seeded earlier as "client feedback" / "stakeholder feedback" / "per the reviewer" and is being built on. Pre-existing and especially stakeholder-contributed numbers carry FALSE AUTHORITY — they look already-vetted, so they skip the provenance scrutiny a brand-new number gets, and may trace to nothing. Verify (grep the source, re-derive with triple-probe) before making them louder. Prevents amplifying a fabrication.
disable-model-invocation	true
author	Claude Code
version	1.0.0
date	"2026-07-01T00:00:00.000Z"

Amplifying an existing number is a provenance re-check trigger

Problem

A number is already sitting in a client-facing deliverable as quiet prose. A stakeholder asks you to make it louder — "add a chart for that", "where's the breakdown, put it back / show it bigger", "feature this stat". The natural move is to take the number at face value (it's already shipped / reviewed, someone put it there) and just render it more prominently.

That's the trap. The number's presence is not evidence of its truth. Numbers that were contributed — seeded as "client feedback", "reviewer per-item feedback", "per the stakeholder", or carried over from a prior session — carry false authority: they read as already-vetted, so they skipped the provenance scrutiny a brand-new number would get. Amplifying such a number (chart, headline, PDF) both increases the blast radius if it's wrong and launders it — a chart looks computed even when the figure was never computed.

Context / Trigger Conditions

• "Add a visual / plot / chart for the X finding instead of just words."
• "Where's the <breakdown/number> — I thought we had it? Put it back / make it prominent."
• You're about to elevate a figure that currently lives in a muted note / caption / aside.
• The number's git blame / comment says it was baked in as feedback ("reviewer feedback", "per client", "stakeholder said"), not from an analysis file or a cited query.
• A caveat/QA gate hard-asserts the exact number (e.g. a test requires the literal 70/65/25/15 to be present) — that gate is now a fabrication-enforcer, not a check.

Solution

1. Before amplifying, trace the number to a real source. Grep the repo/analysis dir for the figure and the finding it belongs to. Ask: is there a computed artifact (query, notebook, analysis doc) that produced it — or only prose that asserts it? A note that the split was a "not-yet-done follow-up" is a red flag the number was never computed.
2. If it traces to nothing, re-derive it — don't amplify on faith. Use the project's verified derivation path (mirror the production join, not a hand-rolled one) and triple-probe (≥2 anchors × ≥2 reasonable definitions). See finding-verification-live-bq-triple-probe.
3. Compare re-derived vs published. Often the ordering/story reproduces but the magnitudes don't (the tell of a plausible-but-fabricated figure). Replace the numbers with the derived set; keep the qualitative claim if it survives.
4. Amplify the verified number, with the caveat baked into the visual (small-n, raw/not-adjusted, selection-inflated) — see preserve-caveats-when-refactoring-annotated-content.
5. Fix any gate that hard-asserted the old number so it asserts the derived truth, not the fabrication (update the literal it checks).
6. Widen the blast check: if this number was seeded in a batch ("baked stakeholder feedback across cards 01/04/08/…"), the siblings likely share the gap — flag them for the same spot-check, don't assume only the one you touched is wrong.

Verification

• The amplified figure now traces to a named query/analysis file (or an inline table.column
  • SQL), not to prose that merely repeats it.
• Re-derived numbers are recorded (probe table + method) in an analysis doc.
• Any test/gate that referenced the old literal now references the derived one.

Example (a decision-support dossier)

Card 03 carried, as a muted note, "among accepted prospects, deposit rates ≈ Info Session 70 / Open House 65 / Webinar 25 / non-attender 15" (illustrative). The stakeholder asked to elevate it to a chart. Instead of charting it, I traced it: the figures had been baked in as reviewer per-item feedback in a prior commit, appeared in no computed analysis, and the by-type split was explicitly logged elsewhere as a not-yet-done follow-up. Re-derived live (triple-probe, 2 anchors × 2 denominators, mirroring the production event-attendance join) → 85 / 80 / 50 / 25. The ordering held (in-person ≫ webinar ≫ non-attender) but the magnitudes were far off — classic plausible-but-uncomputed. Charted the verified numbers with the raw/small-n caveat baked into the PNG; updated the caveat gate's literal 70/65/25/15 → 85/80/50/25 (it had been enforcing the fabrication); and flagged the sibling cards seeded by the same commit for a spot-check. Had I taken the "just add a chart" brief at face value, I'd have laundered a fabricated number into a client-facing plot.

Notes

• The elevation request is the gift: it's the moment a stale/unsourced number gets fresh eyes. Treat "make X prominent" as "prove X first."
• Distinct from dashboard-redesign-gated-on-finding-revalidation (a known label-fix in flight during a redesign) — here the number's unsourced-ness is undiscovered until the amplify request forces the check.
• Distinct from preserve-caveats-when-refactoring-annotated-content (that governs keeping caveats attached; this governs verifying the number itself before making it louder).

See also

• finding-verification-live-bq-triple-probe — the re-derivation mechanics (triple-probe, anchors).
• data-provenance-verifier, verify-plan-constants-against-data — provenance/constant checks.
• preserve-caveats-when-refactoring-annotated-content — bake the caveat into the amplified visual.
• dashboard-redesign-gated-on-finding-revalidation — the redesign-time sibling.
• handoff-fabricated-locked-decision-vs-open-roadmap — fabricated-authority in handoffs.