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 component source retains its independent review. 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. No automatic Research launch, model, ticket, work dispatch, live project commission or resource ETA is qualified.
The optional software-factory/milestone-research entry is a pure parser/preparer for complete caller-authored findings and a first plan. Its types use the existing native M01 and W32 identities; it is not a generic research protocol. software-factory/milestone-research-contract has no Node ambient requirement. A parsed digest is an identity check, never current permission or semantic approval.
import { prepareMilestoneResearch } from 'software-factory/milestone-research';
export function prepareRecordedDraft(link, commission, originalIntent, command) {
// All four values were retained by the caller. No new ID, pin or version is created.
return prepareMilestoneResearch({ link, commission, originalIntent, command });
}
export function reconcileOriginal(milestones, retainedEnvelope) {
// The actual M01 API replays the exact original actor/request under current READ.
return milestones.execute(retainedEnvelope.context, retainedEnvelope.request);
}
inspectResearchHandoff(value) checks the original sf-milestone-research-handoff/1, recorded-draft/1 value and returns a frozen draft with eligibility:'unassessed', researchJob:'not-bound' and forecast:'unknown', or a fixed refused result. Its digest is SHA256 of canonical JSON {protocol:'sf-milestone-research-handoff-digest/1',value:handoff}. Native collection/quote syntax is checked using Discovery's actual parsers and quote rules. No store or capture bytes are read, so a valid shape does not verify attribution, relevance or truth.
prepareMilestoneResearch takes exactly {link,commission,originalIntent,command}. The separate sf-milestone-research-link/1 wraps the unchanged handoff and one exact retained W32 admission/collection/observation per collection. Only replayed is omitted from an admission identity. The link digest uses {protocol:'sf-milestone-research-link-digest/1',value:link}. Existing Product, job, occurrence, collection, brief, source and command grammars remain native. Job IDs are not truncated into SourcePins. Collection completeness and byte-verifiable quotes never become a semantic review.
For a criteria proposal, command must be null; output has no M01 request. The owner can independently review a later explicit revision, keeping the original report bound to its original intent. For a plan, the caller provides command:{requestId,expectedVersion} and complete generation/plan values with the existing M01 canonical digest preimages. The preparer validates a dependency-before-dependent first-generation DAG, complete current criterion coverage, exact commission/intent/value/policy/allowance equality and provenance bounds. Output contains one ordinary admit-plan request, still unassessed. It does not establish that no prior plan exists or that M01 has remaining lifetime capacity: the actual current host and receiver decide those facts.
Retain the entire expanded request and original actor in the caller's existing durable operation owner before invoking M01. After missing acknowledgement or restart, use those same bytes first. Do not rebuild a request with current pins, use another actor, choose a new ID or release the pending operation because a read is absent. Native recorded receipt replay under current READ must precede fresh report/review/source checks; later review expiry, revoked mutation authority or policy changes do not rewrite historical admission.
Qualify the existing caller
The module supplies no report, review, envelope store or effect adapter. Without an existing qualified retained-record owner and current independent-review lookup, effect integration is unsupported; an unassessed draft can still be displayed. RetainedResearchRecords<C,F> describes optional caller capabilities, not an installed service. The two synchronous lookups receive the authenticated context, exact SourcePin and the host's live physical frame. They return a retained record or fixed missing/forbidden/unavailable data, never an allowing token.
The report is an earlier record containing author, original W32 requests/admissions/results, complete native collection identities, citations and classified observations/inferences/questions. The later handoff points to that record; the independently retained review binds the exact link and native plan digests, milestone/criteria, original allowance, current values/policy, evidence-use basis, author-independent reviewer, recorded time and exclusive validity cutoff. Do not create a self-referential digest or derive a record ID from a URL. The caller's real record owner authenticates its immutable identities; a self-generated hash and an approved word supply no authority.
Use the existing actual M01 CurrentAuthorityFence<CommandJournal>. Its common physical lock must cover all relevant original commission, values/Guidance, policy, source, record/review, cache and permission writers. Resolve the original records and current independent reviewer eligibility under that lock. Confirm actual current READ/mutate/evidence-use permission, withdrawal, expiry and exact current native observations before returning M01's existing planAdmission. Reading pins across independent stores is not an atomic fence. Admission-time validity does not promise physical commit-before-expiry under Journal contention.
W32 preflight runs outside that lock: authenticate the complete original JobRef with public inspect, traverse all evidence pages and obtain the exact original admission receipt by calling admit with the actual inspected originalRequest, requiring replayed:true. Do this only after the existing job was found; never submit an arbitrary request copied from a report. Retain these authenticated owner observations separately from the report. There is no W32 already-held-frame API. Inside M01's lock, compare the entire original JobRef/request/admission (including producer/store, command ID, original version and recordedAt; the current replay flag is not an original field) with the retained report/link and current original host records. Also use the caller's captured native Journal and passive Discovery collection/quote readers to recheck the authenticated aggregate version, complete native CollectionRef, whole collection and current citation validity. Missing owner observations, changed original identity or changed version make new admission unavailable; matching collection bytes or a recomputed request digest are insufficient. Never invoke Research methods inside that fence or call collection/transport from Journal decide. Fresh policy must expressly assess parked/late/uncertain evidence; the link retains those observations without turning them into failure or approval. A ten-minute collection deadline is not an evidence-use deadline.
The disposable two-host recipe in the source tests demonstrates co-located SQLite records and a separate local SQLite record owner under the same physical lock. It is a qualification fixture, not a supplied application store or a live Tally commission. Actual UI adoption of a real report/review/command owner remains a separate composition gate. Existing milestone Research/ticket controls remain unavailable.
Bounds and effects
Whole input/link and output are limited to64KiB canonical UTF-8, depth32 and4,096 observation slots; one-over refuses without truncation. There are at most4 collections/jobs,16 citations/findings,8 criteria/generations,8 prerequisites per generation and dependency depth8. Statements/rationale are512 canonical UTF-8 bytes; native generation/request/source limits also apply. Quote checks retain the native15-word/1,024-code-unit rules. Count maxima do not promise all maximal strings fit together. Symbol/accessor/non-JSON/native Promise input refuses with fixed INVALID, LIMIT or CONFLICT; bounded intrinsic observation never invokes getters or user thenables, awaits data, or grants effects.
Imports and calls open no stores and invoke no collector, callback, clock, model, transport or M01 method. Runtime imports include native Discovery/Journal code for canonical parsing; the entry is pure-call, not an I/O-free dependency graph. Objective wording, reports, quotations and source URLs remain private caller data. Diagnostics expose only fixed codes. Maintenance/security/recovery/improvement keeps its existing independent policy and allowance. Native M01 date projections remain separately attributed; this adapter derives no overall or subobjective ETA from capacity counts.
