Historical local qualification: Source 7cf4b26e focused Core 22/22 and Dashboard 26/26 checks passed. The historical bounded local record keeps their exact scope. Historical source 7cf4b26e passed the 3,286/3,286 full checks and all 36 installed families. Independent bounded local evidence acceptance passed; exact-final-head CI is recorded separately from this frozen local proof. Module-specific and broader limits remain below.
Status: In progress. Corrected actual-Core binding and plan endpoint passed independent source review; built-browser save/select/reload and narrow-screen checks passed. Historical source 7cf4b26e passed the 3,286/3,286 full checks and all 36 installed families. Independent bounded local evidence acceptance passed; exact-final-head CI is recorded separately from this frozen local proof. This optional entry supplies no listener, credentials, registry, policy, store, model, workflow or Budget.
import {createDashboardMilestones} from 'software-factory/dashboard-milestone-binding';
export function ownerMilestones(options) {
return createDashboardMilestones(options);
}
Pass the resulting backend as milestones to startDashboard from software-factory/dashboard. The existing Dashboard is unchanged without it. Ordinary owner inputs are title, objective and optional achievement guidance. The host expands provenance and policy; the form never asks for physical bindings or allowances. Empty criteria remain an honest draft, with completion and planning admission unavailable. Research and ticket workflows remain unavailable; this interface does not schedule work or predict a completion date.
Commission the caller first
Supply the actual public CommandJournal, an original M01 commission, a fixed RetentionCommission and a host whose binding names those exact objects and values. All Product registry, current values, policy, authority, Guidance and retained-envelope writers must share one non-expiring physical lock. The host supplies current authenticated actor and permission outside browser arguments. Its synchronous run delivers the exact request/invocation/binding references while that same lock is held. The adapter's private Core fence consumes only that live frame; it never reacquires the physical lock. Reentrant, repeated and delayed callbacks refuse. These identity checks do not authenticate a dishonest host or replace physical qualification.
The caller commissions the private originals root before construction. Its exact grammar is RetainedEnvelopes; initialise revision 0, entries [], protocol sf-milestone-ui-originals/1 and bindingDigest equal to sha256: plus SHA-256 of canonical JSON {protocol:'sf-milestone-ui-binding/1',commission,retention}. Use public canonicalJson from software-factory/command-journal. No implicit creation, missing-store repair, expiry, eviction, namespace rotation or quota reset occurs. The retained original capacity is metadata storage, never authority to spend. Fixed original limits are at most 128 entries, 512 writes and 4 MiB; each envelope is at most 32 KiB. A new preparation reserves three writes and 512 bytes for its largest native Core receipt before it can execute. Exhaustion still allows passive reads and existing recorded Core receipt replay.
originals.replace must durably commit its CAS before returning, independently of the outer authority transaction, then read must return those exact bytes. A later outer rollback must not roll back this preparation after the separate Core Journal commits. Use a separate durable store under the common lock. Native synchronous stores must not leave a deferred write after refusal or lock release. Per-envelope canonical checksums reject altered retained fields; they are consistency checks, not authentication of a dishonest store. Persistence integrity, crash durability and physical identity belong to the qualified caller; arbitrary records asserting these properties are not proof.
Original requests and current authority
prepare retains the complete original actor and expanded Core request before any Core write. An identical ID/request finds that original; changed reuse conflicts. Only one unresolved request can exclude replacement per Product. execute first retains unknown state, then invokes actual M01 with the original bytes. It never regenerates pins, actor, ID or expected version. After lost reply/reload/restart, read discovers pending originals; only exact original Core replay can recover a possibly committed command. Current READ remains required, including historical replay; new writes still need current mutation authority and current pins. A retained receipt is historical, not new permission. Missing records never prove no effect.
A synchronous definitive Core refusal may retain not-recorded only after the original invocation settles under this host lock. Future duplicates then return that refusal without calling Core. Timeout, generic host/storage fault, uncertain commit and corrupt history cannot produce that conclusion. Original envelope IDs and terminal records are never reused. Editing milestone guidance preserves its existing completion criteria and desired target; it does not edit Product values, release rules or accepted work.
Revision history uses recordedAt:null: M01 does not expose original revision timestamps. Whole-item pages stay below the response cap and carry captured-version cursors. The owner overview keeps its existing unknown forecast/dependency fields. The optional separate plan view below retains native evidence without claiming a resource ETA.
When guidanceWriter:'same-fence' is commissioned, the adapter constructs actual ProductGuidance on the same Journal and performs its write inside the same current host frame. The host's next values observation must derive from that actual committed Guidance revision. Otherwise choose disabled; no unfenced legacy writer fallback is available.
Honest limits:
- Two disposable real caller arrangements and HTTP tests qualify the stated source boundary only; they are not live deployment, CI, Fable or browser acceptance.
- No Research, implementation ticket, release, model, notification or scheduler dispatch is provided.
- The coordinator owns real app assets and responsive browser proof; installed-package and combined integration evidence are separate.
Passive reviewed plans and source dates. source candidate
The actual binder now exposes optional readPlan(productId, query). Other backends may omit it; the session then omits milestonePlans and the plan route returns UNAVAILABLE. No new authority, allowance, callback or store is created. A caller may supply HostFrame.planningEvidence as current data under the same qualified physical fence. The complete canonical value is limited to 61,440 bytes, leaving space within M01's existing 65,536-byte request bound. Core's native Temporal/Capacity parsers retain scope, freshness and attribution checks. Native data must already be suitable for this owner's disclosure; raw source documents, credentials and response payloads are never needed.
GET /api/products/:id/milestones/plan?query=<JSON> accepts exactly milestoneId, revision (1–8 or current), plan (native PlanPin or latest), limit (1–8), and native cursor or null. JSON is at most 2,048 UTF-8 bytes; duplicate members/parameters, extra fields and write methods refuse. Later pages require the returned exact milestone revision and plan pin. The cursor preserves captured Core version, milestone, plan and last subobjective; a changed durable version refuses as stale. Missing revisions refuse rather than silently switching plans.
The separate factory-milestone-plan/1 envelope preserves the actual MilestoneProjection and identifies this page's supplied planning-data digest or explicit absence. The complete HTTP JSON is bounded to 73,728 bytes, with no split or truncated generation/forecast/dependency. M01 already bounds its whole success response to 65,536 bytes; the additional wrapper fits within the reserved 8 KiB. Each page records its own native asOf, source/evaluated digests and evidence digest: external evidence and time are not cursor snapshots. The owner page replaces pages rather than concatenating them into a fictional single forecast.
The linked Plan and timeline page displays intended subobjective outcomes, exact prerequisites, source scenarios, conflict/expiry/unknown states, separately labelled desired targets, and historical versus current completion. Capacity counts retain their original successive-full-cycle meaning; aggregate resource ETA remains unknown/not derived because the opaque account/profile bridge is unqualified. Research, ticket binding, implementation and dispatch remain unavailable. Current maintenance/security/incident/self-improvement policy stays visible without invented allocation. This source increment still requires distinct review, installed proof and the coordinator's real asset/browser qualification.
