Drift Dashboard

Adapted from docs/admin/DriftDashboard.md in the Wikantik repository; current as of release 2.4.56.

This page is for curators and administrators who keep the wiki's metadata healthy. It explains the Drift screen at /admin/drift, what each row means, and what to do about it, and it covers two related checks that have no dashboard panel: structural conflicts and stale citations. Drift is metadata that has wandered away from the schema: it never breaks a page, but it weakens search, the Knowledge Graph and the ontology, so the dashboard tracks the count over time as a burn-down.

Read the dashboard

The screen (AdminDriftPage.jsx, backed by AdminDriftResource) shows the latest sweep and compares it to the one before.

Element	Meaning
Last sweep, Triggered by, Pages scanned, Duration	when the latest sweep ran, whether it was `scheduled` or `manual`, and how much it covered
SHACL not checked badge	the ontology subsystem was off or failed during the sweep, so the `shacl` rows are missing, not zero
Table rows	one row per `(family, code, severity)`: Family, Code, Severity, Count
Δ	change in count since the previous sweep (`+n` or `-n`); an em dash means the code did not exist in the previous sweep
Trend	sparkline of that code's count across the sweeps of the last 30 days
Pages (Show/Hide)	expands a live list of the offending pages for that code
Run sweep now	starts a sweep in the background; the screen shows progress until it finishes

The page list under Show is recomputed when you open it, never read from the snapshot, so it shows what is wrong now. If you already fixed the pages it says so (No pages currently affected). Each entry links to the page's editor and shows the field, the message and, where one exists, a suggested replacement.

How sweeps run

A sweep validates the frontmatter of every page and, when the ontology is enabled, checks the materialised graph against the SHACL shapes. Only one sweep runs at a time; a second request returns HTTP 409 and the screen shows "A sweep is already running." Sweeps run in two ways:

Scheduled: after each ontology rebuild completes (by default every wikantik.ontology.rebuild.interval.hours, 24).
Manual: Run sweep now, or POST /admin/drift/sweep (returns 202).

You can also poll the REST endpoints directly:

curl -u admin:... http://localhost:8080/admin/drift/summary
curl -u admin:... 'http://localhost:8080/admin/drift/trend?days=30'
curl -u admin:... 'http://localhost:8080/admin/drift/pages?family=frontmatter&code=tags.kebab'
curl -u admin:... http://localhost:8080/admin/drift/status

days is clamped to 1-365.

Fix each kind of drift

Family `frontmatter`

These come from SchemaDrivenFrontmatterValidator, the same validator that runs on every save, so a code you see here is also what an editor sees in the frontmatter editor. Fix the page and the count drops at the next sweep. Most are warnings, so pages still save. The Severity column tells you which.

Code	Severity	What it means	What to do
`yaml.parse`	ERROR	the page's frontmatter block is not valid YAML	open the page, fix the YAML (quote values that contain a colon); `GET /admin/frontmatter-issues` lists these with line and column
`type.noncanonical`, `status.noncanonical`	WARNING (ERROR if `wikantik.frontmatter.enum.nonCanonical.severity=error`)	value is outside the canonical set (`type`: article, hub, reference, runbook, design; `status`: draft, active, archived)	change to a canonical value; the message names a suggested one when it knows it
`audience.enum.invalid`	ERROR	`audience` holds a value other than `humans`, `agents` or `both`	correct the value or list
`summary.length`	WARNING	`summary` is outside 50-160 characters	rewrite it to fit
`aliases.length`, `aliases.list`, `aliases.blank`	WARNING	an alias is too long (over 100), the field is not a list, or an entry is blank	clean up the alias list
`cluster.slug.malformed`	WARNING	`cluster` is not lowercase kebab-case	use the suggested slug, or rename the cluster everywhere with `rename_cluster`
`cluster.undeclared`	WARNING	the page names a cluster that no hub page declares	create the hub (`type: hub` plus `cluster: <path>`) or move the page to a declared cluster
`cluster.hub.multivalued`	ERROR	a hub page lists more than one cluster; a hub declares exactly one	reduce the hub's `cluster` to a single value
`tags.kebab`	WARNING	a tag is not lowercase kebab-case	rewrite the tag
`related.unresolved`	WARNING	an entry in `related` does not match an existing page	fix the name or remove the entry
`date.date.malformed`, `verified_at.date.malformed`	WARNING	`date` is not an ISO date, or `verified_at` is not an ISO timestamp	rewrite as `YYYY-MM-DD` or ISO-8601
`verified_by.untrusted`	WARNING	`verified_by` is not on the trusted-authors list, so confidence may be reduced	verify the page as a trusted author, or add the author to the registry
`runbook.*` (`missing_block`, `malformed_block`, `when_to_use_empty`, `steps_too_few`, `pitfalls_empty`, `related_tool_invalid`, `reference_unresolvable`)	ERROR	a `type: runbook` page whose `runbook:` block fails the schema	complete the `runbook:` block

Frontmatter fields themselves are described in docs/user/Frontmatter.md.

Family `shacl`

These rows come from checking the ontology graph against its SHACL shapes. The Code is the property path of the violated shape (for example a wk: predicate), and the severity is always ERROR. A violation that is not tied to a property path is counted under node-scoped. Expand a row to see the focus node and the message; the usual fix is in the Knowledge Graph (correct or remove the offending edge in Admin → Knowledge Graph) or in the page that produced it. The write-time gate refuses new non-conformant edges for the predicates that have shapes, so this family mostly measures edges that predate it or that no shape guards at write time. The same list is at GET /admin/ontology/violations; see Ontology Management.

Structural conflicts

Structural conflicts are defects in the Page Graph's structural index: missing identifiers and a malformed cluster taxonomy. They are not shown on the Drift screen. List them with:

curl -u admin:... http://localhost:8080/admin/page-graph/conflicts
curl -u admin:... 'http://localhost:8080/admin/page-graph/conflicts?kind=HEADLESS_CLUSTER'

Each entry has slug, canonical_id (when present), kind and detail. The kind filter is case-insensitive. The endpoint returns 503 when the structural index is unavailable. The seven kinds (StructuralConflict.Kind):

Kind	Meaning	What to do
`MISSING_CANONICAL_ID`	the page has no `canonical_id` and was indexed under a synthesised id	save the page again; the save filter assigns a `canonical_id` automatically
`RELATION_ISSUE`	the page declares a relation the validator rejected	correct the relation in the page's frontmatter
`DUPLICATE_CLUSTER_DECLARATION`	two or more hubs declare the same cluster; reported once per losing hub (the lowest slug wins)	keep one hub per cluster and change or retire the others
`HEADLESS_CLUSTER`	a cluster has member pages but no hub declares it (subject is the cluster path)	create a `type: hub` page with that `cluster:` value
`UNDECLARED_CLUSTER`	a sub-cluster (`parent/child`) names a parent that no hub declares	declare the parent with a hub page
`CLUSTERLESS_HUB`	a `type: hub` page carries no `cluster:`, so it declares nothing	add `cluster: <path>` to the hub
`MULTI_CLUSTER_HUB`	a hub names more than one cluster; only the first is treated as declared	leave a single cluster on the hub

Duplicate declarations are refused at save time only when wikantik.cluster_declaration.enforcement.enabled is true; it defaults to false, so duplicates can still exist. Enable it only after you confirm none do.

Stale citations

A page can ground a claim in another page's section with inline cite:// markup. At save time those citations are parsed into the citations table, pinned to the target's version and hashed by span, and graded as the target changes. Staleness is a patient curation task, never a save-time error, and a changed target version alone does not make a citation stale. Citation tracking is on by default (wikantik.citations.enabled) and needs a datasource. The dashboard has no citation panel; use the endpoint, or the list_stale_citations tool on /knowledge-mcp (see MCP Integration):

curl -u admin:... http://localhost:8080/admin/drift/citations
curl -u admin:... 'http://localhost:8080/admin/drift/citations?page=SomePage'

Without page you get the corpus-wide counts. With page you also get that page's outbound citations (claims it makes) and its inbound citations that are no longer current (other pages relying on it). Each entry has the source and target canonical ids, target heading path, span text, claim text, status and pinned version.

Status	Meaning	What to do
`current`	the pinned span is still present in the target's cited section	nothing
`stale`	span drift: the pinned span no longer appears in that section (the text changed, moved, or the heading was renamed)	re-read the target; if the claim still holds, edit the citation's quoted span or repoint it, which creates a new citation pinned to the target's current version; if not, correct the claim
`target_missing`	the target page was truly deleted (a rename does not count)	point the citation at the new location, or remove the claim

When citations are disabled the endpoint answers 200 with zero counts and empty lists. Citations are a third edge type, separate from Page Graph wikilinks and Knowledge Graph entities; the design is in docs/superpowers/specs/2026-06-14-phase-3-citation-edges-design.md.

Where to go next

Admin Panel: the other admin screens.
docs/user/Frontmatter.md: the fields that drift codes refer to.
Ontology Management: rebuilding the ontology and reading SHACL results.
docs/superpowers/specs/2026-06-09-drift-dashboard-design.md: why the dashboard works this way.

Cluster Hub: Wikantik Platform Development Hub: Architecture, MCP Protocols, and Agent Briefings (wikantik-development)