Skip to content

Sense 24: The Steward — jinflow Operating On Itself

Sense 24 · Forming · Last touched 2026-07-27

  • last_verified: 2026-07-27

Synced from docs/design/sense_24_the_steward.md in the engine repo — that’s the source; this page is a build-time mirror.

The analyst has the Canvas. The CFO has Observations. The Steward is the third native habitat — for the person who operates the estate.

Status: proposed Date: 2026-05-03 Author: the owner + Claude (Santa Pola Sunday morning)


Three native habitats exist in jinflow today.

The analyst opens an entity page, navigates the Canvas, browses signals, drills into findings. Every surface they need was designed for them.

The CFO reads an Observation, asks “show me,” sees the Validation, asks “why,” follows the Explanation through Contributing Factors. Sense 15 gave them their front door.

There is a third person in the room, and we have given them nothing.

The estate operator — the person responsible for the health of many tenants, the freshness of many packs, the integrity of every seam between them — operates today through git log, ad-hoc jinflow ls, file-tree spelunking, and personal memory. When the is_priority regression happened, three forensic days passed before anyone could even see the situation clearly. There was no surface for the question “what is the state of every relationship in this estate, right now?”

The Steward is the missing third habitat.

It is not a dashboard. It is not an admin panel. It is the operator’s home — the place where the estate becomes legible the way a single tenant’s data is legible on the Canvas.


The Steward is jinflow operating on itself.

This is the load-bearing idea. Not “build a separate admin application,” not “add an admin sidebar with new pages.” The architecture is already settled; the domains are open. The Steward is one more domain — and the domain happens to be jinflow itself.

Tenants become entities. Packs become entities. Seam crossings become facts. Drift becomes a metric. Migrations become events. Every alignment, every promotion, every engine handshake is a row in a table somewhere. The same engine that compiles signals on hospital data compiles signals on estate health. Same Canvas. Same Map Salon. Same Ledger.

When a Steward asks “which tenants are drifting?” they are running a signal — a typed observation with polarity (drift is negative unless it’s intentional), score (how far?), direction (worsening or healing), impact (which artifacts affected). The same vocabulary, turned inward.

When a Steward asks “what happened on this seam in the last quarter?” they are reading a Ledger — the same computational provenance, but the numbers being explained are seam-crossing counts and migration events, not money flows.

When a Steward simulates “what if I align inspire to the latest pack?” they are using a Simulation (Sense 19) where the deterministic intervention is a curated afs update and the scenario is the post- alignment AFS.

This framing avoids the trap of building a separate operator app. The architecture has always been ready for it.


The Steward’s habitat is composed of five views, each a composition of existing surfaces over estate-shaped data.

A directory of every tenant, every pack, every consortium, with their bilateral relationships drawn explicitly. Not a hierarchy — a graph. Each edge carries metadata: merge base, last alignment, drift count.

Today this lives in nobody’s head fully. Several people have partial maps. The Estate Map is everyone’s shared anchor.

For a single tenant, a per-artifact-class breakdown of how it differs from each repository it descends from. Pack: 12 specialised, 5 diverged, 3 promoted-back-pending. Engine: 1 minor migration available, 0 hard breakages. Sister tenants: 4 patterns that converged, 2 that diverged.

The Drift View is what the Steward consults before deciding to align. It is what was missing in the days leading up to the is_priority incident.

Every seam crossing as a journal entry. On 2026-04-13, this tenant promoted 4 pieces of Subject Matter to the pack. On 2026-04-20, this tenant pulled pack v2.3. On 2026-04-29, the engine migration v0.46 ran against seven tenants.

The History reads like a ledger because it is one — Sense 18’s machinery, applied to estate events instead of analytical numbers.

A safe space to apply a proposed alignment, observe the result, and decide whether to commit. The Steward picks a tenant, picks a pack version (or a migration), and watches the outcome unfold against a disposable copy. The original is never touched.

The Quarantine is where the conflict simulator we discussed lives. Without it, every alignment is a leap of faith.

A cross-tenant view that respects sovereignty. Show me every classification named ‘severity’ across the estate, with a count of how many values each tenant declares. The Steward sees the shape of agreement and disagreement; they do not see what individual tenant rows say without an explicit drill-down.

The Federation is what makes “this concept exists in three of four tenants” a question you can answer in seconds rather than months.


The Steward does four things. The verbs are ordered by the level of authority each requires.

Observe — read-only, always available. Estate Map, Drift View, History, Federation. No side effects. Counts and names and dates; not file contents.

Understand — drill into specific content (a tenant’s actual Legend, a pack’s actual classifications block, a seam crossing’s actual diff). This crosses from state into content. Audited; the drill-down is logged.

Decide — frame an action without taking it. Quarantine an alignment, simulate a migration, list promotion candidates. Decisions are first-class artifacts: written, named, reviewable. Not yet executed.

Act — execute a decided alignment, migration, or promotion. Authenticated via Identity & Passes. Journaled in the History as a new seam crossing. Reversible where possible (alignment events can be re-run from older merge bases); flagged when irreversible.

The four verbs move from passive to authoritative. Every Steward session begins in Observe; only specific moments require Act.


The four invariants from the Seams (Sense 20 and 23) apply here too, shaped to the Steward’s specific posture.

State before content. The Steward sees counts, names, dates, versions, drift metrics by default. To see actual file content (Legend prose, signal definitions, Subject Matter narratives) they must drill down explicitly. The drill-down is logged. This is the sovereignty invariant: the Steward has more authority than any single tenant’s analyst, and that authority is constrained by visibility discipline.

No silent estate-wide actions. Bulk operations announce themselves. Aligning every tenant to a new pack release is a journaled event, with a per-tenant outcome record. The Steward never “quietly cleans up” anything across the estate.

Stewardship is a capability, not a role. Granted via Identity & Passes. Some humans hold partial Steward capabilities (read-only across one consortium); some hold full ones (act across the estate). The Steward surface adapts to the capabilities the viewer carries.

Read-only by default. The Steward observes more than it acts. The default surface for any session is a read-only one. Acting requires explicit elevation, even when the operator has the capability.


This is the architectural punchline.

The Steward’s data layer is another tenant of jinflow. Its bronze layer extracts state from the engine, the packs, and every tenant the operator can see. Its silver layer validates that state. Its gold layer serves the five views.

Tenants are entities. Packs are entities. Seam crossings are facts. Migrations are facts. Drift is a metric.

Signals exist that score drift. Theses exist that say “this consortium needs alignment.” Verdicts exist that explain which incident pattern matches. Subject Matter exists that captures operator knowledge (“this tenant always lags by one quarter; not a problem”). Observations exist for the C-level of the Steward’s reality (the platform team’s leadership): we observe drift growing in the cardiology consortium.

Same vocabulary. Same machinery. Same Form-follows-Data principle.

What this avoids: a parallel codebase. A separate “ops application.” A different mental model. The Steward is the operator’s view; the view is built from the same engine that builds the analyst’s view.

What this gains: every improvement to the engine improves the Steward. When Sense 19 (Simulation) ships, the Steward gets it for free — they can simulate alignments. When Sense 22 (Legend) ships, the Steward gets it — drift severity can carry icons and rules.

The architecture is settled. The domains are open. The Steward is one more domain.


Not yet manifest in the engine. Status: proposed. See the implementation phases / open questions above for the path from this paper to running code.


  • Sense 13 — The Canvas. The Steward’s pages are Canvas pages. No new component framework.
  • Sense 14 — The Signal. Drift detection runs as signals over estate state.
  • Sense 15 — Observation. The platform team’s leadership has Observations about estate health, just as a CFO has them about hospital finance.
  • Sense 18 — The Ledger. The History view is the Ledger applied to seam crossings.
  • Sense 19 — The Simulation. The Quarantine is a Simulation over alignment outcomes.
  • Sense 20 — The Seam. The Steward observes the pack/tenant seam; promotion runs through here.
  • Sense 21 — The Heartbeat. The Steward listens to all beats across all tenants. Beat 5 (notify) carries estate-level news to the Steward.
  • Sense 22 — The Legend. Steward classifications (drift severity, alignment confidence, migration risk) carry their own Legends.
  • Sense 23 — The Engine’s Seam. The Steward tracks every tenant’s migration status against the engine.
  • sense_25_identity_and_passes.md. Stewardship capabilities live in the pass model. Multiple Stewards may exist with different scopes (consortium-only, full-estate, read-only).

  • No specific page layouts. Whether the Estate Map is a force-directed graph, a treemap, or a sortable table is a Canvas composition decision, not a Sense decision.
  • No data model for seam crossings. The schema for “a tenant promoted four artifacts on this date” is implementation. This Sense names the concept; the table shape is for the next paper.
  • No multi-Steward governance. When two Stewards see the same estate, who acts first? How do they coordinate? Out of scope.
  • No retrofit plan. The estate today has zero seam-crossing history captured (we have git logs across many repos, not unified fact tables). Building the bronze layer that ingests that history is implementation work; this Sense names what it should look like.
  • No specific dashboards. Drift dashboard, promotion candidate scanner, migration status view — these are Canvas pages composed over the Steward’s data. Naming each one is downstream.

Same logic as Sense 15 (Observation). The C-suite needed its own front door because no front door existed for them. The administrator needs the same.

Without naming it:

  • Every operator builds ad-hoc tooling.
  • Every regression takes days to diagnose.
  • Every alignment is a leap of faith.
  • The estate’s history lives in personal memory.

With naming it:

  • The Steward has a habitat.
  • The architecture has a place to put estate-shaped surfaces.
  • The Identity & Passes model has a clear authority bracket to govern.
  • The next regression takes minutes to see and hours to fix.

The Steward is also the place where every other Sense’s value compounds. Sense 19 (Simulation) is more valuable when the Steward can simulate cross-tenant alignments. Sense 21 (Heartbeat) is more valuable when the Steward sees every tenant’s beats at once. Sense 22 (Legend) is more valuable when the Steward’s own classifications (drift severity, migration risk) carry meaning the way analytical ones do.

Naming the Steward is what closes the triangle: analyst, executive, operator. Three habitats; one engine.


For nineteen Senses we built rooms for the analyst. For one Sense (15) we built the room for the C-level. The third person — the one who keeps everything alive — has been working in the hallway.

Sense 24 is the door into their room. The room itself is built from the same materials as everywhere else: signals, theses, verdicts, Subject Matter, Canvas pages, Map Salons, Ledgers, Simulations, Legends. The architecture has been ready. We just hadn’t given the Steward the key.

The estate is a tenant. jinflow operates on itself. The Steward is home.


Numerical neighbors: ← Sense 23: The Engine’s Seam — Where the Engine and the Tenant Disagree · Sense 25 — Identity, Roles, Passes, Capabilities — Concept Paper →

jazzisnow jinflow is a jazzisnow product
v0.65.0 · built 2026-11-13 10:20 UTC