stack/docs/requirements/onboarding-wizard.md

# Onboarding Wizard Contract (D4)

Status: DRAFT — awaiting ratification (webui-audit S2, contract 3 of 9).
Authority: PRD D4 (Part I §6), D11 v1 slice (Part I §9), D8 architecture
gate (Part I §8), the identity account-lifecycle contract
(`docs/requirements/identity-lifecycle.md`, §§2–4), and the RBAC grant
model contract (`docs/requirements/rbac-grant-model.md`, §4.3). This
document turns D4 into a concrete contract: what the wizard is
architecturally, its v1 step model, mode branching, re-run/idempotency
semantics, and what it seeds under whose authority.

Revision 2 (sol review F1–F9): first-company authority repaired — the
bootstrap writer creates only the admin and closes the epoch; step 4
runs as the new admin through the ordinary top-level company command
(F1). Mode is an install-time input the wizard reads, never writes
(F2). The initial user's `name` restored from the PRD step list, and
enrollment skippability flagged as a §12.1 drafting addition (F3).
Explicit-choice obligations restated as wizard-run presentation rules
that leave identity's defaults untouched, with a submitted-vs-canonical
witness (F4). The D8 witness bound to contract 5's mapping and
authorization-parity probes (F5). Witness coverage extended to the mode
branch, reconfigurability, canonical-state drift, second-company actor
semantics, exact seed sets, tab inheritance, and enrollment (F6). Step
4's at-most-once semantics grounded in same-transaction audit
reconciliation with per-mutation failure injection (F7). The custody
note reduced to a cross-reference (F8). Creation eligibility defined in
identity terms with an actor matrix (F9).

Revision 3 (sol re-review: residual F2/F3/F5/F6/F7/F9, new N1/N2): the
wizard is now explicitly a client-side composition with no server-side
orchestrator, spanning six named command families, every operation an
ordinary outbound Gateway call inside contract 5 §6.1's inventory scope
(F5). First runs collect steps 1–2 without writing; the collected
settings are applied immediately after the epoch transition under the
first admin's platform authority, so no pre-admin write authority
exists or is invented (N1). Pre-epoch mode branching reads the mode
value from the unauthenticated bootstrap-status response, a disclosed
one-field addition; post-epoch runs use contract 6's authenticated read
(F2). The SSO display-name source is bound to identity §5's mapped name
claim, with a canonical-name witness (F3). At-most-once seeding is
fenced by a caller-supplied idempotency key recorded under a unique
constraint in the same transaction as the mutation; correlation ids
revert to pure trace linkage (F7). Re-run witnesses now cover every
mutable family, assert the second-company grant is same-operation, and
extend tab inheritance to re-run and explicit-choice machinery (F6).
Eligibility is restated in identity §7.1's terms — deactivation is the
better-auth ban, with no separate state (F9). Example-data seeding is
mandatory; the skippability policy is removed (PRD §6.4) (N2). All
drafting additions are consolidated in §7.

Revision 4 (sol re-review 2: residual F2/F5/F7/N1, new N3/N4): the
post-epoch application phase is replaced by a single **bootstrap
finalize command** — step 3 submits the first-admin fields and every
collected value (settings, registration mode, applicable JIT choices,
seed parameters) in one command on identity §3's bootstrap surface,
executed server-side in one transaction that creates the admin, closes
the epoch, and commits the carried values under the new admin's
authority; after it, every collected value is canonical state, so an
interrupted run resumes from canonical state alone (N1, N3). The v1
first admin is password-only; identity §3.6's SSO variant is not
composed by the v1 wizard (N1's SSO ordering). The bootstrap-status
response is a closed two-field schema (epoch state plus a
pre-epoch-only mode field), with the contract 6 §2.2 amendment
disclosed (F2). The idempotency envelope is completed: fence rows
record operation, actor, scope, and payload digest; replay is
re-authorized; changed-payload collision is refused; only committed
outcomes are recorded; the concurrent loser receives the winner's
recorded outcome (F7). §7 now names the contract 5 §3.1 rank-6
mapping rows the composition expansion amends (F5, N4).

Revision 5 (sol re-review 3: residual F2/F7/N1/N3, new N5/N6): the
finalize command is now honestly scoped as the wizard's **single
disclosed server-side composed transaction** — a bounded, severable
amendment to the §1.1 client-side-composition rule, reachable only
while the epoch is open, executing under the bootstrap writer's
amended authority (never by authenticating the caller as the created
admin), with a dedicated finalize-write inventory witness proving its
internal write set equals the §3.3 carried-value list; §1.2 and §5.1
no longer claim identity §3 binds unchanged (N1, N5). The epoch field's
wire values are pinned to the closed token pair `open`/`completed`
(identity §3 names the states but no wire enumeration; the pin is
disclosed) (F2). The company-name seed parameter gains an owning
record: the declared `bootstrap.seed-company-name` settings value,
committed by finalize and read through the ordinary settings surface
(N3). The §6.7 fence witnesses add the different-actor collision, the
different-scope collision, and the aborted-winner concurrency case
(F7). The password-only first admin is restated as a disclosed PRD
deviation (deferment) of the initial-user SSO option, not as its
satisfaction (N6).

Revision 6 (sol re-review 4: residual F7, new N7): the seed parameter
becomes **immutable seed provenance** — `bootstrap.seed-company-name`
is written exactly once by the finalize transaction and any later
write to it is refused, so a re-run's re-derived seed sequence is
byte-stable and replays cleanly against the §4.3 fence instead of
colliding with it; the current company name's canonical home is the
company row (renamed through the ordinary hierarchy command), and the
wizard renders the row, never the provenance value (N7). §6.3 gains
the changed-name re-run and rename-then-re-run witnesses plus the
immutability refusal probe, with a mutated-digest collision control
(N7). §6.7 adds the two missing envelope witnesses: the same raw key
under two different operations executes both independently
(pair-uniqueness, not global key-uniqueness), and a fresh
authorization refusal records no fence row so an authorized retry
executes afresh (F7).

Revision 7 (sol re-review 5: residual F6, new N8): the fence gains a
**replay mode** — each keyed submission declares `actor-bound`
(default) or `shared`, recorded in the fence row; a shared row replays
for any freshly authorized actor whose scope and payload digest match,
while an actor-bound row keeps the revision-6 actor-equality rule, and
a declared-mode mismatch is a collision. Every §3.4 seed submission
declares shared, reconciling actor-bound replay with re-runnability:
the seed set is a per-epoch singleton, so a fresh-authorized second
admin's no-change re-run replays the first run's recorded outcomes
instead of colliding (N8). §6.7's different-actor witness splits into
the actor-bound refusal and the shared replay, plus a mode-mismatch
collision and a shared-row lost-eligibility refusal; §6.3 adds the
second-admin re-run witness (N8). §6.3's changed-name case now
continues through step 4 with the bookkeeping cache deleted — original
digests asserted, every seed outcome replayed, zero new nodes, run
completed — proving the dirty form value never enters derivation and
the renamed row resolves cache-independently (F6).

Revision 8 (sol re-review 6: NEW-9): shared replay gains a
**target-result authorization boundary** — returning a recorded
outcome is a read of the canonical records it references, authorized
separately from fresh-mutation authorization: a shared replay
requires the submitter to hold, at replay time, read authority on
every record the recorded outcome references, and a submitter who
passes the fresh-mutation check but lacks target read authority is
refused with the same constant-shape single bounded conflict as a
§4.3 collision, identifying no record — the response confirms no
tenant target's existence, preserving RBAC §7's no-existence-oracle
rule (§4.3). `shared` becomes server-verified seed-only policy: the
command layer re-derives the epoch's **canonical seed tuple set**
from canonical state alone and refuses any shared declaration whose
(operation, key, scope, digest) tuple is not a member, and a missing
seed fence is originated only by a submission passing the full
fresh-mutation authorization for that seed command — a wrong actor
cannot win an unrecorded seed key, and no shared row can exist
outside the seed sequence (§4.3). An authorized shared replay appends
a **replay access event** — a distinct non-mutation audit event class
recording the replaying actor, the current request's correlation ids,
and the fence row whose outcome was returned — so every access to a
recorded outcome is attributed without a duplicate semantic mutation
event, while the fence row and the mutation event immutably retain
the original actor (§4.3, §7.4). §6.7 adds the unauthorized-target
replay refusal, the non-seed shared declaration refusal, and the
unrecorded-seed-key race; §6.3's second-admin witness grants B
explicit read authority on the recorded seed targets and asserts B's
replay access events; replay-affected count equalities are scoped to
the mutation event class (NEW-9).

Revision 9 (sol re-review 7: NEW-9 residual, NEW-10): the seed
boundary becomes server-complete and oracle-free. The **canonical
seed key set** is fence-independent (epoch id plus the fixed
seed-role list), while the full tuple set is **prefix-derived**: each
position's scope and payload digest derive from the committed
predecessors' recorded fence outcomes, the derivable members at any
state are the committed prefix plus the next unrecorded position,
the pre-first-company state derives exactly the seed-company tuple,
and an out-of-order shared declaration is refused recording nothing
(§4.3, §3.4). The epoch gains one immutable **seed-origin account**
— the first admin the finalize transaction created — and a
**seed-boundary gate** evaluated after fresh-mutation authorization
and before fence presence can affect the response: a submitter on a
canonical seed key who is neither the seed-origin account nor holds
target-result read authority on the position's recorded targets
receives the constant-shape bounded conflict, identical across the
recorded and unrecorded worlds, creating nothing in either — so
§5.2 eligibility alone can never originate a seed fence and the
eligible-wrong-actor race on the top-level seed-company key is
closed (NEW-9; witnesses §6.7: the two-world control and the
top-level origination race). Target-result authorization extends to
**every replay mode**: an `actor-bound` replay requires the recorded
actor to hold live read authority on every record the recorded
outcome references, actor equality remaining an additional
condition, never a substitute — a creator whose grants were revoked
is refused, not replayed (NEW-10; §6.7 lost-target-grant witness).

Revision 10 (sol re-review 8: NEW-11): the seed-origin boundary becomes
recoverable without breaking the no-oracle result. The epoch's
seed-origin is a **designation** the finalize fixes to the first admin,
changed through exactly one disclosed command — **seed-origin
succession** (§4.3, §7 item 12): an eligible platform admin holding
target-result read authority on the committed seed prefix succeeds only
while the current origin fails identity §7.1 eligibility; refusals keep
the constant conflict shape whatever condition failed and whatever is
recorded, concurrent successions serialize on the epoch record, the
change is one audited epoch-record mutation, and recorded fence rows
immutably retain their original actor. The seed-boundary gate and the
origination rule read the epoch record's current designation, so a
banned or deleted origin no longer strands the unoriginated suffix —
the successor resumes and completes it with no factory reset and no new
epoch, restoring D4 re-runnability and §4.4's strand-nothing rule
(NEW-11; witnesses §6.7: succession recovery, origin-available refusal,
succession race).

Scope: the Gateway-backed product onboarding wizard. Out of scope: the
host-local install wizard (`mosaic wizard`, which drives host install and
gateway bootstrap and is not this artifact — audit REPORT.md layer 3);
Enterprise mode conversion (contract 6); custody semantics (contract 7);
the deferred-beyond-v1 steps themselves (connectors, comms
integrations, voice-matching, M365 — D11 defers them; they bind here only
through the extensibility rule §2.4).

## 1. Placement and architecture

1. The wizard is a **product surface over the Gateway command API** — a
   web UI flow (and equivalently scriptable command sequence) that
   composes Gateway commands. It is subject to the D8 hard rule like
   every other webUI surface: no wizard operation reaches the database
   or filesystem directly, and no wizard-only privileged write path
   exists. Concretely, the wizard is contract 5's rank-6 family, and it
   is a **client-side composition with exactly one disclosed
   exception**: every wizard operation except step 3's finalize is an
   ordinary outbound Gateway call issued from the wizard's web modules
   (or the equivalent scripted sequence), and the exception is the
   §3.3 **bootstrap finalize command** — one server-side composed
   transaction on identity §3's bootstrap surface, reachable only
   while the bootstrap epoch is open, disclosed and ratified as the
   §7.3 architecture amendment. No other server-side module composes
   wizard operations. Contract 5 §6.1's outbound-call inventory,
   scoped to the wizard modules, therefore sees the wizard's complete
   outbound call set; the finalize command's internally composed
   writes are proved complete by the dedicated §6.1 finalize-write
   inventory, not by the outbound inventory.
   The composed operations span six existing families: the rank-1
   hierarchy commands and the rank-4 enrollment command (contract 5
   §3.1 — built first), the settings command family (steps 1–2), the
   identity bootstrap and registration surface (step 3, including the
   §3.3 finalize command), the mode reads (contract 6 §2.2 post-epoch;
   the §2.3 bootstrap-status field pre-epoch), and the ordinary
   content commands used for example seeding (step 4). Two of these
   families (ranks 1 and 4) are in contract 5 §3.1's live rank-6
   composition row; the other four are added by the disclosed §7.8
   mapping amendment, without which their operations are unmapped and
   blocked under contract 5 §5.
2. **The wizard introduces no new mutation surface.** Every state change
   it performs is an existing command with its own contract: user and
   epoch writes under the identity contract §3, hierarchy writes under
   contract 1 §5, grant writes under contract 2 §4, settings writes under
   their owning command family. The one exception is the bootstrap
   surface, where the wizard drives the bootstrap writer defined by
   identity §3 **as amended by the disclosed §7.3 finalize
   extension**: the writer's constraints (§3.1–§3.6) bind, and this
   contract's single change to them — extending the epoch-closing
   command to carry the §3.3 value set inside the same transaction —
   is exactly the §7.3 amendment, severable and ratified with this
   contract. Beyond that amendment, nothing is added to identity §3,
   and no undisclosed authority exists.
3. **Wizard state derives from canonical state.** Each step renders the
   system's current configuration (read through the same commands) and
   applies deltas; the wizard does not keep an answer file whose contents
   can drift from reality. Values a first run has collected but not yet
   committed by the §3.3 finalize command are transient run state, not
   a persisted answer file; after finalize every collected value is
   canonical state, and no transient value is needed to resume
   (§4.4). The wizard MAY persist run bookkeeping
   (started/completed timestamps); bookkeeping is a cache, never the
   source of truth for any decision. The §4.3 idempotency fence is
   command-layer canonical state, not wizard bookkeeping — losing
   bookkeeping never affects it.
4. If a wizard-completion marker is stored, it is presentational only
   (which entry screen to show). No authorization or gating decision may
   read it: gating state lives where its owning contract puts it
   (bootstrap epoch in `bootstrap_state`, registration mode in settings,
   mode in the contract-6 mode record).

## 2. Modes and extensibility

1. The wizard **differs by mode** (D4): Standalone and Enterprise share
   one skeleton; Enterprise makes personal data optional and moves focus
   to business structure, RBAC, and external systems.
2. **v1 ships the Standalone flow only** (D11). The mode branch point and
   skeleton MUST still exist in v1 — mode is a property of the flow, not
   a fork of it — but no Enterprise-only step ships, and mode conversion
   is contract 6.
3. **Mode is an install-time input, not a wizard output.** The
   deployment mode is recorded canonically at bootstrap and owned by
   contract 6 (§2 there): the wizard reads the recorded mode and
   branches on it; it never writes mode and never derives it from
   feature state. During the bootstrap epoch no authenticated account
   exists, so contract 6 §2.2's authenticated read surface is
   unreachable there; the pre-epoch flow instead branches on the mode
   value reported by the **unauthenticated bootstrap-status response**
   of the bootstrap surface (identity §3), which evaluates the mode
   record server-side. The bootstrap-status response is a **closed
   two-field schema**: `epoch` — one field whose value set is exactly
   the bootstrap epoch states identity §3 defines for
   `bootstrap_state`, closed to the exact wire-token pair **`open`**
   and **`completed`** — identity §3 names the two epoch states in
   prose but defines no wire enumeration, so this contract pins the
   tokens, as the second clause of the disclosed §7.9 amendment — and
   `mode` — the recorded mode value, present **only while the epoch
   is open** and absent from the response schema once the epoch has
   completed. No other field exists in the response. The unauthenticated pre-epoch
   mode disclosure is a disclosed amendment to contract 6 §2.2's
   authenticated-read rule (§7.9); post-epoch, §2.2's
   authenticated-only rule holds unchanged and the bootstrap-status
   response carries no mode field. Post-epoch wizard runs read mode
   through contract 6 §2.2 unchanged. Changing
   mode later is conversion (contract 6), not a wizard re-run. In v1 a
   recorded mode of `enterprise` is refused at bootstrap (contract 6
   §5), so the wizard's Enterprise branch is unreachable in v1; a
   wizard invoked against an unsupported or unreadable mode record
   produces a single bounded refusal (contract 5 §4.2 precondition
   class), never a partial flow.
4. **Extensibility:** new wizards attach as tabs (D4). Attaching a wizard
   tab is a registration of additional steps against the same skeleton,
   inheriting this contract's rules (§1 architecture, §3 step contract,
   §4 re-runnability). A tab cannot opt out of them: tab steps are
   subject to the same §6 witnesses as the built-in steps.

## 3. v1 step model (Standalone)

The v1 wizard consists of exactly these steps, in order, each backed by
the named authority:

1. **System and company name.** System name and the seed company name
   are both settings values: the company name is recorded under the
   declared settings key **`bootstrap.seed-company-name`** — the §3.4
   seed parameter, a disclosed drafting addition (§7.11) — committed
   by step 3's finalize command among the ordinary settings writes and
   readable through the ordinary settings read surface. The company
   name feeds step 4's company creation. **The seed parameter is
   immutable seed provenance:** it is written exactly once, by the
   finalize transaction, and every later write to the key — through
   any surface, any actor — is refused by the settings family (a
   disclosed immutable-key rule, §7.11). It records what the seed
   sequence was derived from, not what the company is currently
   named: the current name's canonical home is the company row, and
   renaming the seed company is the ordinary hierarchy rename
   command against that row, never a settings edit. On a first run
   this step collects and validates only — the collected values are
   committed by step 3's finalize command (§3.3). On a re-run (an
   authenticated admin exists) the system name and other settings
   apply directly through the settings commands; the company-name
   field renders the company row's current canonical name and
   applies a change as the hierarchy rename.
2. **Component choices.** Mosaic Comms/Matrix vs external; Mosaic
   SSO/Authentik vs external; Mosaic DB/PostgreSQL vs external; vector
   DB (D4). v1 records the choices as settings (first run: collected,
   committed by step 3's finalize command §3.3; re-run: applied
   directly) and
   configures what is installable in-product; component provisioning
   beyond that is host tooling, out of scope here.
3. **Initial user and finalize.** Email, password, and display name
   (PRD §6 step list: "email/password/name/SSO"), per identity §3: the
   first-admin transition is atomic and one per bootstrap epoch. **The
   v1 first admin is password-only.** A step-2 SSO choice cannot yield
   a canonically configured provider before the epoch closes, so an
   SSO-authenticated first admin is not constructible in this flow;
   the v1 wizard does not compose identity §3.6's SSO variant. The
   PRD step list's SSO option **for the initial user is not
   implemented in v1 — a disclosed PRD deviation (deferment, §7.10)**:
   no v1 step configures a provider before finalize and no v1 step
   links the first admin to a provider afterwards, so the initial user
   cannot be SSO-created or SSO-linked in v1. SSO first becomes
   available to accounts created post-epoch (providers configured and
   JIT-enabled after finalize; identity §5 governs every SSO-created
   account) — those are different users, not a satisfaction of the
   initial-user option. The password path's canonical `users.name` MUST equal the
   submitted display name; any SSO-created account's name is bound to
   identity §5's mapped name claim (witness §6.5); no field of the
   PRD's step list is dropped. The wizard also presents the applicable
   mandatory choices — registration mode (identity §2.2) and, for each
   SSO provider canonically configured at finalize time, that
   provider's JIT enablement with its allowlist warning (identity
   §4.1, §4.3); a provider configured after finalize receives its JIT
   choice at configuration time under identity §4.1's default. The
   wizard MUST obtain an explicit submission for each applicable
   mandatory choice and MUST refuse finalize while one is unsubmitted.
   This is a presentation-and-submission obligation on wizard runs
   only: identity's own defaults (registration `closed`, JIT off)
   continue to govern everything a wizard run never touches, and this
   contract creates no new stored-record obligation beyond those
   identity itself defines. **Finalize rule (first runs):** step 3
   ends in a single **bootstrap finalize command** on identity §3's
   bootstrap surface — a disclosed extension of that surface (§7.3) —
   carrying the first-admin fields and every collected value: the
   step-1 system name, the step-1 company name (recorded as the §3.4
   seed parameter), the step-2 component choices, the
   registration-mode submission, and the applicable JIT submissions.
   The command executes **in one transaction**: it creates the admin
   and closes the epoch exactly as identity §3 defines, then commits
   every carried value through the ordinary settings and identity
   configuration write implementations — the same code paths the
   post-epoch commands use; no parallel write path. The executing
   authority is the **bootstrap writer's own authority as amended by
   §7.3**, never an authentication of the caller: the finalize
   request begins and ends unauthenticated, and creating the admin
   row does not authenticate the caller as that admin. Each carried
   write is attributed in audit to the newly created admin as the
   accountable principal the transaction establishes; the admin
   authenticates afterwards by ordinary login.
   Either the whole transaction commits or none of it exists (identity
   §3.5's atomicity extends over the carried writes). Before finalize
   the run holds collected values only: no canonical write of any kind
   occurs pre-epoch, and no pre-admin write authority exists or is
   invented (witness §6.11). After finalize, every collected value is
   canonical state; no transient value is needed to resume (§4.4,
   witness §6.7).
4. **Initial hierarchy and seeding.** Runs strictly after step 3's
   finalize transaction, authenticated as the newly created admin. The
   bootstrap writer (identity §3) creates only the first account,
   closes the epoch, and commits the §3.3 carried values; it has no
   hierarchy authority and creates no hierarchy node. The seed
   sequence is derived entirely from canonical state: the seed
   parameter the finalize command recorded (the company name, read
   from the `bootstrap.seed-company-name` settings value, §3.1) and
   the fixed example set (§4.3). A resumed run — including a fresh
   client holding none of the original run's transient state —
   reconstructs the same ordered sequence and the same deterministic
   §4.3 keys from that canonical state alone, prefix-wise: each
   position's scope and payload derive from the committed
   predecessors' recorded outcomes, so the run walks the order and
   never precomputes a tuple past the next unrecorded position
   (§4.3 shared-declaration boundary). Because the seed
   parameter is immutable (§3.1), the re-derived sequence is
   byte-stable across every re-run and resume: the same keys carry
   the same payload digests, so already-committed mutations replay
   (recorded outcomes) rather than collide, regardless of any
   hierarchy rename performed since — and, because every seed
   submission declares §4.3's `shared` replay mode, they replay for
   whichever currently eligible admin holding §4.3 target-result
   read authority on the recorded seed targets performs the re-run,
   not only the actor the fence rows record. The §4.3 fence, not the
   parameter's presence or absence, is what prevents re-seeding; the
   parameter is provenance the derivation reads, never a value any
   later step may change.
   The first company is created by the ordinary top-level company
   command under §5.2's eligibility policy, with the new admin — the
   epoch's §4.3 seed-origin account — as actor,
   naming the admin as initial `owner` in the same audited operation
   (contract 2 §4.3); the §4.3 seed-boundary gate reserves
   origination of the canonical seed tuples to the epoch's current
   seed-origin account — initially this admin, thereafter changed
   only by §4.3 seed-origin succession — while
   the same command outside the canonical seed key set follows §5.2
   unchanged. The initial estate, initial project, and initial
   workspace with seeded example data (D4, D11) follow through hierarchy
   commands (contract 1 §5.1) under the admin's `owner` authority
   (contract 2 §4.3). Seeded examples are ordinary workspace content
   created by ordinary commands, attributable in audit to the acting
   admin, carrying the run's trace correlation ids (contract 5 §4.3)
   and the §4.3 idempotency keys.
5. **Minimal agent enrollment.** One harness, API-key login, agent name
   and persona (D11). Enrollment specifics belong to the rank-4
   agent-enrollment command family; this contract binds only that the
   step exists, uses that family, and is skippable. Skippability is a
   drafting addition under PRD §12.1 (the PRD step list does not mark
   the step optional); it is disclosed in §7 and ratified with this
   contract.

Steps deferred beyond v1 (user onboarding profile, connectors, comms,
voice-matching) appear in `docs/ROADMAP.md` per D11. The deferred
profile step's custody semantics are contract 7's (D14); the v1 wizard
collects no sensitive category, so v1 ships no custody surface.

## 4. Re-runnability and idempotency (D4 "no lock-in")

1. **Re-run is a first-class operation.** After completion, running the
   wizard again re-opens every step against current state (§1.3) for
   reconfiguration. Nothing about completion locks the wizard.
2. **The bootstrap epoch does not re-open** (identity §3.4). On re-run,
   step 3 shows the existing admin/registration/JIT configuration and
   allows changing the mutable parts through their normal commands;
   "setup already completed" is a stable state, and factory reset — a
   future, explicitly destructive operation — is the only path to a new
   epoch.
3. **Idempotent seeding, fenced at the command layer.** Step 4 on
   re-run MUST NOT duplicate: the initial company/estate/project/
   workspace and the example data are created at most once per bootstrap
   epoch. The at-most-once mechanism is an **idempotency key**, a
   disclosed drafting addition to contract 5's §4 command envelope
   ratified with this contract (§7): a mutating command MAY carry a
   caller-supplied idempotency key. The complete envelope:
   - **Fence row.** The command layer records, in a
     uniqueness-constrained fence table **in the same transaction**
     as the mutation and its audit event: the key, the operation
     identifier, the acting principal, the mutation's authorization
     scope (the target company, or the platform scope for top-level
     operations), a digest of the canonicalized request payload, the
     submission's declared **replay mode** — `actor-bound` (the
     default) or `shared` (§7.4) — and
     a reference to the committed outcome. Fence uniqueness is the
     pair (operation identifier, key).
   - **Replay.** A submission whose (operation, key) pair is recorded
     is first authorized exactly as a fresh submission would be — an
     actor who is not authorized receives the authorization refusal,
     never the recorded outcome. If authorization passes, the
     submission's declared replay mode equals the recorded row's, and
     the recorded scope and payload digest equal the submission's,
     the submitter must pass **target-result authorization** — in
     every replay mode. Returning a
     recorded outcome is a read of the canonical records that
     outcome references, and neither fresh-mutation authorization
     nor recorded-actor identity is
     read authorization on those records: every replay, `shared` or
     `actor-bound`, requires the submitter to
     hold, at replay time, read authority on every canonical record
     the recorded outcome references, under each record's owning
     contract's read-authorization rules (for the seed targets, the
     contract 2 grant model — RBAC §1.1 platform-admin standing
     confers none of it). Actor equality is the additional
     `actor-bound` condition, never a substitute for target-result
     authorization: an `actor-bound` row replays only for the
     recorded actor, and only while that actor holds live
     target-result read authority — an original actor whose grants
     on the referenced records were since revoked is refused, not
     replayed (witness §6.7). A submitter who passes the fresh-mutation
     check but fails target-result authorization is refused with the
     same single bounded conflict error as a collision below —
     constant in shape, identifying no record, disclosing nothing of
     the recorded outcome or its targets — so the refusal confirms
     no tenant target's existence or identity and RBAC §7's
     no-existence-oracle rule is preserved. On replay the command
     executes nothing and returns the recorded outcome. A shared
     replay adds no **mutation** audit event and leaves the fence
     row untouched — audit attributes each mutation to the actor who
     performed it, and the recorded acting principal is immutable —
     but it appends a **replay access event**: a distinct
     non-mutation audit event class recording the replaying actor as
     accessing principal, the current request's correlation ids
     (contract 5 §4.3), and a reference to the fence row whose
     outcome was returned. The current submission's correlation
     travels only in that access event and the trace layer; the
     recorded outcome and its original mutation event are returned
     and retained unchanged. Every access to a recorded outcome is
     thereby attributed to its accessing actor without a duplicate
     semantic mutation event ever existing.
   - **Collision.** A submission whose (operation, key) pair is
     recorded but whose scope, payload digest, or declared replay
     mode differs — or, against an `actor-bound` row, whose actor
     differs — is
     refused with a single bounded conflict error (contract 5 §4.2);
     it executes nothing and discloses nothing of the recorded
     outcome.
   - **Shared-declaration boundary.** `shared` is a server-verified,
     seed-only policy, never a caller privilege. The boundary has a
     fence-independent part and a prefix-derived part. The
     **canonical seed key set** — the (operation identifier, key)
     pairs of §3.4's ordered seed sequence — is derived from the
     epoch id and the fixed seed-role list alone: the keys are
     deterministic and depend on no generated id, so membership is
     decidable before any seed mutation has run and without
     consulting the fence table. The full **canonical seed tuple
     set** is prefix-derived, because later seed tuples embed
     generated ids: the tuple at seed position k — its
     authorization scope and payload digest — is derived from the
     epoch id, the immutable `bootstrap.seed-company-name`
     provenance, the fixed example set, and the canonical recorded
     outcomes of positions 1 through k−1 (the ids the committed
     predecessor fence rows reference). At any canonical state
     exactly these members are derivable: every committed-prefix
     tuple (read back from its fence row) and the next unrecorded
     tuple in order. Before the first company exists, the derivable
     set is exactly the seed-company tuple. A fresh client derives
     the same way — §3.4's derivation is this walk: submit the
     sequence in order, learning each generated id from the
     returned recorded outcome or the submission's own execution,
     never precomputing a tuple past the next position. A
     submission declaring `shared` whose (operation, key) is
     outside the canonical seed key set, whose tuple does not equal
     its position's derived tuple, or whose position lies past the
     next unrecorded position (out of order) is refused with a
     single bounded refusal (contract 5 §4.2 validation class) that
     executes nothing and records no fence row. No other operation
     can carry a shared declaration, so no shared fence row can
     exist outside the seed sequence — the seed-only rule is
     enforced by the command layer, not by wizard convention.
   - **Seed-boundary gate and origination.** The epoch record holds
     one **seed-origin designation**: at finalize it names the
     account the §3.3 finalize transaction created as the epoch's
     first admin (identity §3), and afterwards it changes through
     exactly one path — the **seed-origin succession** command below
     — never by any other write. The epoch's **seed-origin account**
     is the account the designation currently names. Every mutating submission
     whose (operation, key) is in the canonical seed key set —
     whatever replay mode it declares — passes, after
     fresh-mutation authorization and before the fence table is
     consulted, the **seed-boundary gate**: the submitter is the
     seed-origin account, or holds §4.3 target-result read
     authority on the canonical records the position's recorded
     outcome references. A submitter satisfying neither is refused
     with the same single bounded conflict error as a collision —
     and because the gate is evaluated without consulting fence
     presence, the refusal is byte-shape-identical whether the seed
     fence and its targets exist or not: the recorded and
     unrecorded worlds are indistinguishable to that submitter, no
     mutation or fence row is created in either, and RBAC §7's
     no-existence-oracle rule holds at the seed boundary itself,
     not merely at an existing fence (witness §6.7). For an
     unrecorded position no recorded outcome exists to hold read
     authority on, so only the seed-origin account can proceed to
     origination: §5.2 eligibility alone never originates a seed
     fence, which closes the eligible-wrong-actor race on the
     top-level seed-company key. **Origination.** A seed fence row
     not yet recorded is originated only by the seed-origin account
     executing its seed mutation in order, passing the full
     fresh-mutation authorization for that seed command (§5.2
     eligibility plus the hierarchy authority the command itself
     requires); a refused submission records no row (no-error
     replay below). The recorded actor of every shared row is
     therefore the account that was the epoch's seed-origin at that
     position's origination, authorized for the mutation the row
     fences. Recorded positions are replayable by any admin holding
     target-result read authority (§6.3).
   - **Seed-origin succession.** Loss of the seed-origin account
     does not strand the epoch (D4, §4.4). The designation changes
     through exactly one mutating command, an amendment to identity
     §3's epoch surface disclosed in §7 item 12: an eligible
     platform admin — passing §5.2 fresh-mutation authorization in
     full — submits succession naming itself the epoch's
     seed-origin. The command succeeds only when, evaluated against
     canonical state inside the succession transaction itself:
     (a) the current seed-origin account fails identity §7.1
     eligibility (banned, deleted, or disabled) — succession while
     the current origin remains eligible is refused — and (b) the
     submitter holds §4.3 target-result read authority on every
     canonical record referenced by the recorded outcomes of the
     committed seed prefix (vacuously satisfied while no position
     is committed). A submission failing either condition is
     refused with the same single constant-shape bounded conflict
     as the seed-boundary gate, byte-shape-identical whichever
     condition failed and whether any seed fence exists — so
     succession adds no existence oracle (witness §6.7) — executes
     nothing, and appends no event. Successful succession updates
     the designation in the epoch record and appends one ordinary
     mutation audit event recording the prior designation, the new
     designation, and the acting principal; it rewrites no fence
     row and no recorded outcome — rows already recorded immutably
     retain their original actor. Concurrent successions serialize
     on the epoch record: exactly one submitter commits and becomes
     the seed-origin, and the loser, re-evaluated against the
     committed winner, fails condition (a) — the now-current origin
     is eligible — and receives the constant-shape refusal. The
     seed-boundary gate and the origination rule always read the
     epoch record's current designation: after succession the
     successor originates the remaining suffix in order under its
     own full fresh-mutation authorization, and the fence rows it
     originates record the successor. Factory reset (§4.2) remains
     the only path to a new epoch; it is never required to complete
     an interrupted seed sequence.
   - **No error replay.** The fence row commits only with its
     mutation, so only committed outcomes are ever recorded. A failed
     or refused submission records no fence row; a retry executes
     afresh. There is no recorded-error state.
   - **Concurrency.** Two submissions with the same (operation, key)
     pair serialize on the fence's unique constraint: exactly one
     executes. The loser waits for the winner's transaction to
     resolve; if it committed, the loser is handled as a replay
     (fresh-mutation authorization, the seed-boundary gate where
     the key is a seed key, and target-result authorization first —
     then the recorded outcome, or
     the collision refusal on mismatch); if it aborted, no fence row
     exists and the loser executes. The loser never performs a second
     mutation and is never left without a defined response.

   The wizard submits deterministic keys derived from the bootstrap
   epoch and the seed role (e.g. epoch id + "seed-company") for every
   seed mutation, and every seed submission declares the `shared`
   replay mode: the seed set is a per-epoch singleton (§4.4), derived
   from immutable provenance (§3.4), so whichever currently eligible
   admin holding target-result read authority on the recorded seed
   targets performs a re-run or resume must replay the recorded
   outcomes rather than collide on actor identity — re-runnability is
   not restricted to the original acting admin (witness §6.3). All
   other submissions default to `actor-bound`; nothing in this
   contract declares `shared` outside the seed sequence, and the
   shared-declaration boundary makes that exclusivity
   server-enforced: a shared declaration outside the canonical seed
   tuple set is refused, not merely unconventional. Correlation
   ids (contract 5 §4.3) remain pure trace
   linkage and carry no idempotency semantics; a shared replay's
   current correlation is carried by its §4.3 replay access event,
   never by rewriting the recorded outcome or its mutation event. Wizard bookkeeping of
   seed node ids remains a cache
   (§1.3): losing it cannot cause duplication, because the fence is
   canonical command-layer state. A re-run offers to create
   _additional_ hierarchy nodes (PRD: users can create N companies, N
   estates, N projects) but never re-creates or resets the originals,
   and never touches user content added since. Example-data seeding is
   likewise at-most-once per epoch; the example set is fixed and its
   seeding is not skippable (PRD §6.4: a completed first run contains
   the fixed example set).

4. **Interrupted runs strand nothing.** The wizard is resumable at the
   granularity of a single command: each mutation either commits
   through its command with its same-transaction audit event and fence
   row, or leaves nothing (identity §3.5, extended over the carried
   writes, for the §3.3 finalize command). A run interrupted before
   finalize has written nothing; its collected values are re-entered.
   A run interrupted after finalize resumes from canonical state
   alone: every collected value was committed by the finalize
   transaction, and the remaining work — the step 4–5 command
   sequence — is derivable from it (§3.4). Step 4 is a sequence of
   individually atomic commands, not one transaction: an interruption
   between them leaves a prefix of committed seed nodes, and the §4.3
   fence makes the resumed run complete exactly the remaining suffix —
   the run re-derives the full ordered seed sequence from canonical
   state and re-submits it with the same deterministic keys,
   already-committed mutations return their recorded outcomes, and
   only the remainder executes, without duplication and without
   compensating rollback of completed commands. There is no
   wizard-level transaction spanning steps. Loss of the seed-origin
   account mid-sequence is likewise recoverable without a new epoch:
   §4.3 seed-origin succession designates an eligible successor, and
   the resumed run completes the remaining suffix under the successor
   — interrupted runs strand nothing even across origin-account loss.

## 5. Seeding authority (resolves contract 2 review NEW-1)

1. During the bootstrap epoch, the bootstrap writer's authority
   (identity §3, as amended by the §7.3 finalize extension) covers
   creating the first account, closing the epoch, and committing the
   §3.3 carried values inside the finalize transaction — nothing
   else. It creates no hierarchy node. The
   first company is created post-epoch by the first admin through the
   ordinary top-level company command (§5.2), naming that admin as
   initial `owner` (contract 2 §4.3). No operator-designated third
   party, service actor, or wizard-privileged writer exists in this
   flow.
2. **Post-bootstrap top-level company creation** — the "N companies" flow
   — is decided by the ruling below: any **eligible platform user** MAY
   create a top-level company and MUST name an initial `owner` grant in
   the same audited operation (contract 2 §4.3); the creator naming
   themselves is the default. Eligible means, in identity-contract
   terms: an authenticated account (identity §2) that is not banned
   (identity §7.1 — deactivation on this platform IS the better-auth
   ban; no separate deactivated state exists). No further role or grant
   is required. Until that ruling, deny-by-default holds (contract 2
   §3.1): no implicit creation authority exists.
3. Child-node creation inside the wizard (estate, project, workspace
   under the seeded company) follows contract 2 §4.3 unchanged: parent
   `owner` authority, no automatic grant needed.

## 6. Verification requirements

Binding on the implementing PRs:

1. **D8 mapping witness:** every Gateway call the wizard makes resolves
   to a contract 5 mapping row, asserted by contract 5 §6.1's
   outbound-call inventory scoped to the wizard modules — an inventory
   that is complete for the wizard's outbound calls because §1.1
   forbids server-side wizard composition outside the single declared
   §3.3 finalize handler, and the witness statically asserts that
   scoped prohibition (no server-side module other than the declared
   finalize handler composes wizard operations). The inventoried
   operation set is asserted equal, in both directions, to the §1.1
   declared family list. **Finalize-write inventory:** the finalize
   handler's internal write set is statically enumerated and asserted
   equal, in both directions, to the §3.3 carried-value list — the
   admin creation and epoch transition plus the settings writes
   (including `bootstrap.seed-company-name`), the registration-mode
   write, and the applicable JIT writes, nothing else; each internal
   write is asserted to invoke the owning command family's ordinary
   write implementation (no parallel write path); and the handler is
   asserted refused once the epoch has completed (identity §3.4). Wizard modules appear in no
   class-table writer allowlist (contract 1 §6.3b) and hold no direct
   database or filesystem access (static assertion, plus a runtime
   probe that a wizard-context filesystem/database access attempt is
   refused). **Authorization parity:** for each mutating wizard
   operation, the same actor invoking the underlying command directly
   receives the same authorization outcome as through the wizard —
   probed for at least one allowed and one refused actor per operation.
2. **Mode witnesses:** the branch point exists — a pre-epoch flow
   resolves its step set from the bootstrap-status mode field (§2.3),
   a post-epoch run from contract 6 §2.2's authenticated read; the
   bootstrap-status response matches §2.3's closed two-field schema
   exactly while the epoch is open — the epoch field carrying one of
   §2.3's two exact wire tokens — and after the epoch completes
   contains the epoch field only — the mode field absent from the
   response (closed-field assertion on the response schema in both
   phases); post-epoch, an unauthenticated mode read through any
   surface is refused (contract 6 §2.2, as amended by §7.9, holds); v1 with mode `standalone` yields the §3 step set; a
   simulated unsupported or unreadable mode record yields one bounded
   precondition refusal (§2.3) and no partial flow; wizard sources
   contain no mode write and no feature-state mode derivation.
3. **Re-run witnesses:** complete the wizard, then re-run it to
   completion making no changes — assert zero new hierarchy nodes, zero
   new users, zero new grants, zero duplicated example content, and the
   bootstrap epoch still completed (identity §8's epoch witnesses cover
   the transition itself). **Reconfigurability:** a re-run changes one
   value in each mutable family — the system name (settings), a
   component choice (settings), registration mode, and a configured
   provider's JIT enablement — and each canonical value reflects the
   change; a run that skipped enrollment enrolls an agent on re-run
   through the rank-4 family and the enrollment exists canonically.
   **Canonical-state drift:** mutate state outside the wizard (a
   rename through the hierarchy command), then re-run — the wizard
   renders the current canonical value, not a stored answer.
   **Seed-provenance witnesses (§3.1, §3.4):** rename the seed
   company through the hierarchy command, delete the wizard's
   seed-node bookkeeping cache, then re-run to completion —
   the re-derived seed sequence carries the same keys and payload
   digests as the first run (asserted digest equality), every seed
   submission replays with the recorded outcome, zero new nodes are
   created, the run completes, and the company keeps its renamed
   name — the deleted cache proving the renamed row was resolved
   from canonical state alone (§1.3), not from a stored id. A re-run
   submitting a different company name renames the company row and
   commits no settings write to the seed key, **and that same run**,
   with the bookkeeping cache likewise deleted beforehand, continues
   through step 4 deriving the seed sequence from the immutable seed
   parameter — asserted to carry the first run's keys and payload
   digests, not digests of the newly submitted name — replays every
   seed outcome, creates zero new nodes, zero new users, and zero
   new grants, and completes; a direct post-epoch
   settings write to `bootstrap.seed-company-name`, by any actor, is
   refused; and a control run that force-mutates a seed payload
   digest receives the §4.3 collision refusal, proving the fence
   still discriminates.
   **Second-admin re-run witness (§4.3 shared replay):** after admin
   A's completed first run, create a second account B and grant it
   platform-admin, the hierarchy authority the seed commands
   require, and **explicit target-result read authority on every
   recorded seed target** — a contract 2 grant conferring read on
   the seed company and its seeded children (B is freshly
   authorized with explicit target authority, and no text restricts
   re-runs to A); delete the bookkeeping cache; B re-runs the
   wizard to completion making no changes — every seed submission
   replays with A's recorded outcome (asserted by mutation-count
   and mutation-audit-event-count equality across the run, the
   fence rows still recording A as acting principal), exactly one
   §4.3 replay access event exists per replayed submission, each
   naming B as accessing principal and carrying B's run's
   correlation ids while referencing A's fence row, zero new
   hierarchy nodes, users, grants, or example content exist, and
   the run completes.
   **Second company:** a re-run creating a second company succeeds with
   the running user as actor and names the creator as `owner` in the
   same audited operation — asserted on the single audit event naming
   both the creation and the grant (contract 2 §4.3–§4.4), not merely
   on the grant row's existence — and leaves the first company's exact
   node and content set untouched (row-set and content-digest
   assertions).
4. **Seed-set witness:** after a first completed run, the created seed
   set is exactly the declared set — one company, one estate, one
   project, one workspace, the example content (fixed digest), the one
   §4.3 `owner` grant, and no other node or grant attributable to the
   run's idempotency keys or trace correlation ids. Every seeded node,
   grant, and example item has its command audit event (contract 1
   §5.2, contract 2 §4.4) committed in the same transaction as its
   mutation and §4.3 fence row, attributed to the acting admin; the
   first-company event names the initial owner grant in the same
   audited operation.
5. **Explicit-choice witness:** a run submitting explicit values
   completes, and each canonical value equals the value the run
   submitted (asserted against the captured submissions, not by reading
   the canonical store twice); canonical `users.name` equals the
   declared source — for the first admin (password path), the
   submitted display name; for an SSO-created account (a post-epoch
   JIT-provisioned user), the identity §5 mapped name claim (each
   asserted against the captured submission or captured assertion); a
   run with
   an applicable mandatory choice unsubmitted is refused at step 3
   (§3.3); identity's defaults are asserted for a provider the run
   never touched (registration `closed`, JIT off — identity §2.2,
   §4.1).
6. **Gating-independence witness:** with any wizard-completion marker
   deleted (§1.4), authorization and registration gating behave
   identically — proving no gate reads wizard state.
7. **Interrupted-run witness:** inject failure at each point of a
   run — **inside the finalize transaction** (after admin creation but
   before the carried writes commit: assert nothing exists — no
   account, the epoch still open, canonical settings untouched, no
   fence row, no audit event), after the committed finalize
   transaction, after each individual seed command (company, estate,
   project, workspace, each example item), and after the bookkeeping
   cache write — and for each injection point assert: no stranded
   partial outcome (every committed command has its same-transaction
   audit event and fence row; nothing else exists), and a resumed run
   completes exactly the remaining mutations with no duplication
   (identity §8 covers the epoch transition; §4.3–§4.4 define the
   fence the witness exercises). The resume after the committed
   finalize MUST be performed by a fresh client holding none of the
   original run's transient state, deriving the remaining seed
   sequence from canonical state alone (§3.4). Deleting the
   bookkeeping cache between failure and resume MUST NOT change the
   outcome. **Idempotency-fence witnesses (§4.3):** re-submitting a
   committed seed command with its same key, actor, and payload
   executes nothing and returns the recorded outcome (asserted by
   mutation-count and mutation-audit-event-count equality); the same
   key with a
   changed payload digest is refused with the single bounded conflict
   and executes nothing; the same key and payload submitted by a
   different authorized actor against an **actor-bound** fence row (a
   control row recorded outside the seed sequence) is refused with
   the same single bounded conflict (actor mismatch) and executes
   nothing, while against a **shared** row (every §3.4 seed fence) a
   different authorized actor holding target-result read authority
   replays with the recorded outcome, executing nothing, adding no
   mutation audit event, and appending exactly one replay access
   event naming the replaying actor and its correlation ids, the
   fence row still recording the original actor; a
   submission whose declared replay mode differs from the recorded
   row's is refused with the single bounded conflict and executes
   nothing; the same key,
   actor, and payload submitted against a different authorization
   scope is likewise refused (scope mismatch) and executes nothing; a
   replay by an actor who has since lost
   eligibility (identity §7.1 ban) receives the authorization refusal,
   not the recorded outcome — asserted against a shared seed fence,
   proving shared replay never bypasses fresh authorization.
   **Target-authorization and boundary witnesses (§4.3, NEW-9):**
   an **unauthorized-target shared replay** — a platform admin
   holding §5.2 fresh-create eligibility but no contract 2 grant on
   the seed company derives the canonical seed tuple per §3.4 and
   submits it with the recorded `shared` mode — is refused with the
   single bounded conflict, and the refusal response is asserted
   byte-shape-identical to the changed-digest collision refusal
   above (same error class, same fields, no record identifier, no
   scope detail beyond the caller's own submission), executes
   nothing, changes no fence row, and appends no mutation audit
   event — proving the recorded target is neither returned nor
   confirmed to exist and RBAC §7's no-existence-oracle rule holds
   at the fence; the **two-world seed-boundary control (§4.3
   seed-boundary gate)** — the SAME eligible non-originator actor
   submits the exact canonical seed-company tuple in two prepared
   worlds: one where the seed fence and its targets exist (the
   recorded world) and one freshly finalized epoch where they do
   not (the unrecorded world, before any seed command has run) —
   and in both worlds receives the identical bounded conflict
   refusal (asserted byte-shape-identical across the two worlds and
   to the collision refusal), executes nothing, and creates no
   mutation, fence row, or hierarchy node in either — proving the
   response is computed without consulting fence presence, the
   recorded and unrecorded worlds are indistinguishable to that
   actor, and no eligible wrong actor can originate the top-level
   seed fence; a **non-seed shared declaration** — an ordinary
   hierarchy or content command whose tuple is outside the
   canonical seed tuple set, submitted with `shared` declared by an
   actor fully authorized for the mutation — is refused with the
   §4.3 validation refusal, executes nothing, and records no fence
   row (asserted absent); an **out-of-order shared declaration** —
   with the seed prefix committed through position k, the
   seed-origin account submits the position k+2 tuple (past the
   next unrecorded position), constructed by the test harness from
   ids it obtained out of band — and is refused with the §4.3
   validation refusal, executes nothing, and records no fence row,
   proving the prefix-aware derivation rejects positions the
   canonical walk cannot yet derive; an **unrecorded-seed-key
   race** in two variants — (child variant) with one child seed
   tuple deliberately left unrecorded, an actor lacking that seed
   command's hierarchy authority races the seed-origin account's
   resume for the same (operation, key): the unauthorized
   submission receives the authorization refusal and records no
   fence row, the origin account's submission executes afresh, and
   the resulting fence row records the seed-origin account;
   (top-level variant, NEW-9) with the seed-company tuple
   unrecorded, an eligible platform admin who is not the
   seed-origin account — passing §5.2 fresh-mutation authorization
   in full, since the top-level command requires no hierarchy
   authority — races the seed-origin account for the seed-company
   (operation, key): the non-originator receives the constant
   bounded conflict refusal from the seed-boundary gate and
   records no fence row and no company, the seed-origin account
   executes afresh, and the resulting fence row records the
   seed-origin account — proving origination of a missing seed
   fence is bound to the seed-origin account, not to eligibility
   alone; an **actor-bound lost-target-grant replay (NEW-10)** — an
   actor creates a non-seed top-level company under an
   `actor-bound` key, grants a second account `owner`, and the
   second account then revokes every grant the creator held on the
   company; the creator, still authenticated and eligible,
   resubmits the exact recorded (operation, key, scope, payload,
   mode) tuple — actor equality holds but live target-result
   authorization fails, so the submission is refused with the
   single bounded conflict, executes nothing, returns nothing of
   the recorded outcome, and appends no access or mutation event;
   re-granting the creator read authority and resubmitting returns
   the recorded outcome — proving actor equality is never a
   substitute for live target-result read authority on any replay
   mode; **seed-origin succession witnesses (§4.3, NEW-11):** a
   **succession recovery** — the seed-origin account commits a
   proper seed prefix and is then banned (identity §7.1); an
   eligible platform admin holding target-result read authority on
   the committed prefix submits succession, the epoch record's
   designation changes to the successor with exactly one mutation
   audit event recording the prior designation, the new
   designation, and the acting principal, and the successor's
   resumed run replays the committed prefix (its access attributed
   by §4.3 replay access events) and originates the remaining
   suffix in order — the new fence rows record the successor, the
   pre-succession rows immutably retain the original origin, and
   the full seed set completes with no factory reset and no new
   epoch; an **origin-available succession refusal** — the same
   eligible admin submits succession while the current origin
   remains §7.1-eligible — is refused with the single
   constant-shape bounded conflict (asserted byte-shape-identical
   to the collision refusal), changes no epoch record, and appends
   no audit event; a **succession race** — with the origin banned,
   two eligible, prefix-authorized admins submit succession
   concurrently: the epoch record serializes them, exactly one
   commits and becomes the designated origin with one audit event,
   the loser receives the constant-shape refusal, and the seed
   sequence completes exactly once under the winner; a submission that failed before commit
   leaves no fence row and its retry executes; two concurrent resumed
   runs executing the seed sequence yield exactly one seed set — per
   key, exactly one mutation and one mutation audit event exist, and
   the losing
   submission received the winner's recorded outcome (its access
   attributed by the §4.3 replay access event); and with the
   winner's transaction forced to abort, the waiting loser finds no
   fence row, executes, and commits exactly one mutation and one
   mutation audit event. Two envelope-shape witnesses complete the §4.3
   coverage: the same raw key submitted to two DIFFERENT operations
   executes both independently, each committing its own mutation,
   audit event, and fence row — proving uniqueness is the
   (operation, key) pair, so a globally key-unique fence fails this
   witness; and a fresh submission refused by authorization records
   no fence row — the same actor, made eligible, retries the same
   (operation, key, payload) and the command executes afresh,
   distinguishing an authorization refusal from a pre-commit failure
   and proving refusals record nothing.
8. **Actor-matrix witness (§5.2):** post-bootstrap top-level company
   creation succeeds for an ordinary authenticated non-admin user
   (positive), and is refused for an unauthenticated caller and for a
   banned account (identity §7.1's deactivation-as-ban) (negatives),
   with contract 2's error classes.
9. **Tab-inheritance witness:** a registered test tab's steps are
   subject to the same assertions — its modules fail the §6.1 static
   assertion if they access the database directly, its mutations
   appear in the §6.1 mapping inventory, a tab re-run making no
   changes produces zero new mutations (§6.3 style), and a tab
   declaring a mandatory choice is refused while it is unsubmitted
   (§6.5 style) — proving §2.4 is enforced by machinery, not
   convention.
10. **Enrollment witness:** the enrollment step exists, invokes only the
    rank-4 family, and a run that skips it completes with zero
    enrollment-family mutations.
11. **Pre-epoch authority and finalize witness (§3.3):** during a
    first run, before finalize, canonical state is untouched
    (settings-state comparison against the pre-run capture), and an
    injected settings-write attempt from the unauthenticated pre-epoch
    flow is refused with contract 5 §4.2's **authentication-failure
    class** (the caller is unauthenticated, not an
    authorized-but-refused actor); the finalize command commits the
    admin, the epoch transition, and every carried value in one
    transaction, each carried write attributable in audit to the new
    admin; the §6.7 inside-finalize injection covers the atomicity
    negative.

## 7. Drafting additions (PRD §12.1 disclosure)

Proposed drafting additions, visible here for ratification, each
severable; the step list, mode branching, re-runnability, and seeding
obligations themselves are traced to PRD D4/D11 and the named sibling
contracts and are not additions:

1. Enrollment-step skippability (§3.5) — the PRD step list does not
   mark the step optional.
2. The unauthenticated bootstrap-status mode disclosure — the closed
   two-field response schema (epoch state plus the recorded mode
   value, the latter present during the bootstrap epoch only) (§2.3).
3. The **bootstrap finalize command** (§3.3) — an extension of
   identity §3's bootstrap surface: one transaction carrying the
   first-admin fields, the collected settings, registration-mode, and
   JIT values, and the §3.4 seed parameter, committing the carried
   writes under the bootstrap writer's amended authority (never by
   authenticating the caller as the created admin; §3.3) with
   identity §3.5's atomicity extended over them. This item is also a
   disclosed, bounded amendment to the §1.1 client-side-composition
   rule: the finalize handler is the wizard's single server-side
   composed transaction, reachable only while the epoch is open,
   witnessed by §6.1's finalize-write inventory. Severable from the
   rest of this contract.
4. The idempotency-key field and its same-transaction uniqueness
   fence, with the §4.3 replay, collision, no-error-replay, and
   concurrency rules — an addition to contract 5 §4's command
   envelope, proposed and ratified here, severable from the rest of
   this contract. Including the **replay mode**: each keyed
   submission declares `actor-bound` (default) or `shared`, the
   declaration is recorded in the fence row, and every replay — in
   either mode — requires the submitter, freshly authorized with
   matching scope and payload digest, to pass the §4.3
   target-result authorization: read authority, live at replay
   time, on every canonical record the recorded outcome
   references, refused otherwise with the constant-shape conflict
   that preserves RBAC §7's no-existence-oracle rule. An
   actor-bound row additionally requires recorded-actor equality —
   never as a substitute for target-result authority — and a
   declared-mode mismatch is a collision. Including
   the **shared-declaration boundary**: `shared` is server-verified
   against the epoch's canonical seed key set (fence-independent,
   derived from the epoch id and the fixed seed-role list) and its
   prefix-derived canonical seed tuple set — each position's scope
   and digest derived from the committed predecessors' recorded
   outcomes — with a shared declaration outside the key set, off
   its position's derived tuple, or past the next unrecorded
   position refused recording nothing (§4.3). Including the
   **seed-boundary gate**: the epoch's seed-origin designation
   initially names the first admin the finalize transaction
   created and changes only through item 12's succession command;
   every submission on a canonical seed key must be the currently
   designated account
   or hold target-result read authority on the position's recorded
   targets, refused otherwise with the constant-shape conflict
   evaluated before fence presence — so recorded and unrecorded
   worlds are indistinguishable to the refused submitter — and
   origination of a missing seed fence is reserved to the
   seed-origin account passing the full fresh-mutation
   authorization for that seed command (§4.3). Including the **replay access event**: a
   distinct non-mutation audit event class, appended on every
   shared replay, recording the accessing actor, the current
   request's correlation ids, and the fence row returned — the
   disclosed mechanism by which a replay's access is attributed
   without a duplicate semantic mutation event, the fence row and
   mutation event immutably retaining the original actor. This
   contract declares `shared` for exactly the §3.4 seed sequence —
   deterministic, epoch-scoped singletons any eligible,
   target-authorized admin must be
   able to re-run (§4.1) — and for nothing else.
5. The presentation-and-submission obligation for applicable mandatory
   choices on wizard runs (§3.3).
6. The no-server-side-orchestrator architectural constraint (§1.1),
   scoped by the single disclosed §3.3 finalize exception (item 3).
7. The wizard-completion-marker presentational-only rule (§1.4).
8. The contract 5 §3.1 rank-6 mapping expansion (§1.1): the rank-6
   composition row, which today names only the rank-1 hierarchy and
   rank-4 enrollment families, is amended to name all six composed
   families — adding the settings command family, the identity
   bootstrap and registration surface (including the §3.3 finalize
   command), the mode reads (contract 6 §2.2 and the §2.3
   bootstrap-status field), and the ordinary content commands used for
   example seeding — with one mapping row per newly named family.
   Without this amendment those operations are unmapped and blocked
   under contract 5 §5.
9. The contract 6 §2.2 amendment permitting the single unauthenticated
   pre-epoch mode disclosure through the §2.3 bootstrap-status field;
   post-epoch, §2.2's authenticated-only rule is unchanged. Second
   clause: the pin of the `epoch` field's wire enumeration to exactly
   the tokens `open` and `completed` (§2.3) — identity §3 names the
   two epoch states in prose but defines no wire values, so this
   contract fixes them.
10. The password-only v1 first admin (§3.3) — a disclosed **PRD
    deviation (deferment)**: PRD Part I §6's initial-user SSO option
    is not implemented in v1. Identity §3.6's SSO variant is not
    composed by the v1 wizard, the initial user cannot be SSO-created
    or SSO-linked in v1, and SSO first becomes available to accounts
    created post-epoch (different users, not a satisfaction of the
    initial-user field). The deviation narrows the PRD field for v1
    and is severable.
11. The **`bootstrap.seed-company-name`** settings key (§3.1) — the
    owning record of the §3.4 seed parameter: written exactly once by
    the finalize transaction among the ordinary settings writes, read
    through the ordinary settings read surface, and **immutable
    thereafter** — the settings family refuses every later write to
    the key, from any surface and any actor. The key is seed
    provenance for §3.4's derivation; the current company name lives
    on the company row and changes only through the ordinary
    hierarchy rename (§3.1, witnesses §6.3).
12. The **seed-origin succession command** (§4.3) — an amendment to
    identity §3's epoch surface, proposed and ratified here,
    severable: one mutating command that redesignates the epoch's
    seed-origin to an eligible platform admin holding target-result
    read authority on the committed seed prefix, valid only while
    the current origin fails identity §7.1 eligibility; refusals
    are the constant-shape §4.3 conflict regardless of which
    condition failed, the change is one audited epoch-record write
    serialized on the epoch record, and recorded fence rows are
    never rewritten. Without this amendment a banned or deleted
    seed-origin account strands the unoriginated seed suffix,
    contradicting PRD D4's no-lock-in requirement (§4.4).

## Ruling request

Ratify sections 1–7 as written, with one decision embedded:

- Decision (§5.2): post-bootstrap, any eligible platform user
  (authenticated, not banned — identity §2/§7.1) may create a top-level
  company, naming an initial `owner` grant (default: self) in the same
  audited operation. Basis: PRD Part I §4 "Users can create N
  companies, N estates, N projects" read as end-user capability, not
  admin-only. Alternative if rejected: top-level creation stays
  deny-by-default and becomes a platform-admin-granted capability in a
  later contract — nothing in this contract or contract 2 breaks
  either way, because deny-by-default is the resting state (contract 2
  §3.1).