Vocabulary Drift Sentinel (VDS) Prompts¶
Six prompts for the Vocabulary Drift Sentinel — the knowledge_graph
persona at Zachman cell ENGINEER/WHAT that watches cross-project
vocabulary usage and raises alerts when a term's meaning begins to
drift between FCC and a consumer project (ATHENIUM, MNEMOSYNE,
PHOENIX, one of the ten constellations).
VDS is the tripwire that keeps 175+ packaged YAML mappings honest as the ecosystem evolves.
Context for Claude¶
When invoked as VDS you load the canonical mappings under
src/fcc/data/objectmodel/, scan consumer-repo term usage (via
vendored registries + federation tracker snapshots), and raise
drift incidents with severity, scope, and remediation owner.
You never fix drift — you file the incident for the owning
persona to resolve.
Prompt 1: Weekly drift scan¶
VDS, run the weekly drift scan. For each of the 8 vendored
consumer projects, compute:
- Term count
- Terms whose FCC mapping changed in the last 7 days
- Terms whose consumer usage diverges > 10% by count
Produce a drift incident log in JSON with at most 20 entries, most-
severe first.
Expected outcome: drift incident log.
Prompt 2: Severity triage¶
For this morning's drift incident log (20 entries), triage each as
P0 (breaks contract), P1 (risk of breaking), or P2 (cosmetic).
Provide:
- Triage label
- 1-sentence justification
- Named owner persona
- Suggested SLA (48h / 1 week / 1 PI)
Expected outcome: triage table with realistic SLAs.
Prompt 3: Sister-project drift report¶
Produce a drift report specific to MNEMOSYNE (the JV L2 memory
library, vendored fcc.vocabulary_providers entry points, pinned
fcc>=1.2.1). Include:
- Terms MNEMOSYNE publishes vs FCC canonical
- Drift direction (FCC-changed-upstream vs MNEMOSYNE-diverged)
- Recommended alignment path (bump version, deprecate, coexist)
Expected outcome: per-project drift report suitable for PR description.
Prompt 4: Drift heatmap for 175 mappings¶
Render an ASCII heatmap for the 175 packaged YAML mappings under
src/fcc/data/objectmodel/. Rows = mapping family (by directory);
columns = consumer project. Cells = drift count over 30 days.
Highlight the row with the most drift and the cell with the single
highest count.
Expected outcome: heatmap + highlight commentary.
Prompt 5: Drift-history dashboard¶
Build a CLI dashboard that shows drift incidents over the last 6
PIs. Include:
- Incidents per PI (bar)
- Top-5 drifting terms cumulative
- Mean time-to-resolve per severity
Commit the dashboard rendering to docs/for-architects/vocabulary-
drift-dashboard.md as a generated section — demarcated by
<!-- VDS GENERATED --> / <!-- END VDS GENERATED --> fences.
Expected outcome: dashboard section with fence markers so CI drift check knows where to regenerate.
Prompt 6: Sentinel false-positive tuning¶
A VDS alert fired yesterday on "cost_center" drift but turned out
to be a legitimate MNEMOSYNE extension, not drift. Produce:
- Post-mortem in 250 words
- Proposed tuning (threshold, exclusion list, or new signal)
- Test case that would prevent re-firing on this specific pattern
Close with a sentence on the cost of over-tuning (false negatives).
Expected outcome: post-mortem + tuning proposal + guardrail.
Related¶
- Persona YAML:
src/fcc/data/personas/vocabulary_drift_sentinel.yaml - Related prompts:
persona-zca-prompts.md,persona-kgse-prompts.md - Tutorial:
docs/tutorials/advanced-capabilities/bidirectional-federation-patterns.md(Pool B)