Current status

Checkpoint date: 2026-07-14 America/New_York
Task: 019f522e-385b-7623-a657-1915dd7bf0a7

Outcome#

The semantic operating build now covers all eight non-respite GUIDE service pillars in task-level detail and joins them into one longitudinal service system. Comprehensive Assessment and Ongoing Monitoring and Support are explicit first-class pillars:

1. Comprehensive Assessment performs the applicable initial, annual, or bounded reassessment and establishes the source-attributed baseline, unknowns, watch items, and next Assessment horizon.
2. Person-Centered Care Plan converts applicable findings and the person's choices into accepted goals, actions, owners, and contingencies.
3. Ongoing Monitoring and Support carries meaningful promises, observations, results, and changes forward between Assessments without becoming a repetitive reassessment or a status-only billing contact.
4. 24/7 Access provides the live human response and opens only the exact follow-on work supported by the event.
5. Care Coordination and Transitional Care pursues outside records, responses, transitions, and practical dependencies to their truthful endpoints.
6. Medication Management preserves source-separated medication facts and routes protected reconciliation and medication judgment to the authorized U.S. clinician.
7. Referral and Coordination of Services and Supports distinguishes choice, transmission, receipt, intake, service start, usefulness, and truthful non-success.
8. Caregiver Education and Support distinguishes a bounded same-event teaching service, a full skills-training route, application support, and other caregiver support.

Program Operations supports these eight services; it is not treated as a ninth service pillar.

All material remains a provisional project draft until the named clinical, compliance, privacy, and operational owners approve it. Open law, contract, clinical-authority, timing, evidence, and payment choices remain open decisions rather than invented requirements.

Exact denominator#

Every task/card was reconciled to its authoritative name or stable source title. The integration work did not create replacement task identifiers.

Longitudinal operating loop#

1. Complete the applicable initial, annual, or bounded Assessment work and preserve each fact's source, date, speaker, scope, uncertainty, and professional authorship.
2. Send the current baseline, explicit unknowns, watch items, open owners, and next Assessment horizon to Ongoing as permitted context. Context availability alone is not an Ongoing service or a human-acceptance task.
3. Create or revise only the affected Care Plan elements through the beneficiary-led planning route.
4. Furnish the selected service through the responsible pillar and preserve that pillar's independent result.
5. Return completed, failed, corrected, refused, and late results to the affected Care Plan action and to Ongoing when a useful longitudinal promise, observation, or person-facing result remains.
6. If a result materially changes a baseline domain, open a bounded affected-domain Assessment. Do not rebuild all 49 Assessment tasks merely because one fact changed.
7. Keep the annual Assessment due separately. A bounded reassessment does not silently replace the annual cycle, and Ongoing does not absorb Assessment ownership.

Cross-pillar feed rules#

The canonical integration manual binds 31 compressed source-feed rows to 48 exact receiving-service routes:

No downstream work opened is not a fourth feed outcome. It is the source- and authority-bounded disposition used before a feed is created when no local SOP trigger, unresolved protected issue, person request, promise, or plan action requires receiving work. The source fact, reason, remaining unknowns, any context destination, and re-entry condition remain visible.

For every feed:

• capture a shared event and its facts once;
• preserve each source and conflict rather than replacing it with an AI summary;
• reuse the packet instead of making the person repeat the history;
• credit a pillar only when its distinct service content or result was actually furnished and evidenced;
• count shared human minutes once, then add only genuinely new human acts;
• keep downstream work open until its own truthful endpoint; and
• return the material result, barrier, or non-success to the affected longitudinal owners.

Preparation before judgment#

The integrated manuals apply settled decision D-021:

1. deterministic retrieval, bounded AI, and permitted lower-cost staff gather and organize available facts;
2. conflicts, missing facts, source limits, and uncertainty remain visible;
3. the system prepares the exact choice set, question, or action packet;
4. a clear authenticated non-protected choice may proceed without purchasing navigator time merely to bless it;
5. a Puerto Rico certified care navigator enters for an actual required/requested human GUIDE service, ambiguity, disagreement, distress, accessibility failure, warm introduction, relationship-dependent work, or other named human condition; and
6. protected clinical work routes directly to the beneficiary-location-authorized U.S. professional, who performs any clinically necessary history/examination and the judgment itself.

AI may collect, normalize, summarize, draft, and communicate information. It does not make clinical, legal, operational, eligibility, payment, routing, closure, or escalation decisions.

Natural episodes covered#

The shared manual includes three end-to-end examples:

• a 24/7 call about a person refusing to bathe, which may furnish limited approved CG12 reinforcement through a permitted L2 human now and separately open full training, Care Plan, referral, clinical, or follow-up work only when actually indicated;
• post-discharge medication confusion, which separates immediate safety response, transition coordination, source pursuit, clinician reconciliation, medication execution, teaching, and monitoring; and
• caregiver exhaustion or possible breakdown, which separates live safety action, caregiver support, referral, planning, and follow-up without treating a transfer as completion.

These examples demonstrate how one conversation can legitimately furnish more than one GUIDE service while avoiding duplicate contacts, duplicate fact gathering, duplicate labor, and false completion.

Canonical semantic work completed#

• 25_GUIDE_Eight_Pillar_Service_Integration.md is the shared integration manual.
• 12_SOP_Comprehensive_Assessment.md explicitly covers initial, annual, and bounded Assessment, independent Assessment closure, and the Assessment-to-Ongoing return loop.
• 03_SOP_Care_Plan.md carries cross-pillar action/result integration and transparent provider-choice boundaries.
• 13_SOP_Ongoing_Monitoring_and_Support.md consumes the current baseline, follows exact promises/results, routes material changes, and keeps the Assessment due view separate.
• 14_SOP_24_7_Access.md preserves live human first answer, direct protected routing, and precise cross-pillar feeds.
• 04_SOP_Care_Coordination.md receives exact external obligations and returns source-specific results without mandatory navigator tollbooths.
• 05_SOP_Medication_Management.md keeps medication-source states distinct and does not close a transition medication episode until an authorized result exists.
• 15_SOP_Referral_and_Services.md returns each actual referral milestone or barrier separately.
• 16_SOP_Caregiver_Education_and_Support.md returns furnished, tried, helped, failed, and still-unknown learning/support results without treating AI content alone as completed full skills training.

Detailed working analyses remain preserved under reviews/working/2026-07-14/codex-main-analysis/. They cover literal procedure, required facts/provenance, deterministic and AI preparation, residual human role, co-joining opportunities, low/ordinary/high human minutes, failure behavior, and truthful closure for all 359 tasks.

Review truth#

• Codex and isolated internal agents produced and mechanically reconciled the eight task-level analyses.
• Grok 4.5 completed an adversarial review; accepted corrections were adjudicated into the canonical manuals.
• Opus 4.8 completed an additional Program Operations analysis; useful items were adjudicated without turning Program Operations into a ninth pillar.
• Fable Runs 10, 12, and 13 ended in HTTP 429 errors with no usable prose; Run 14 was never launched. Fable lane 3 remains a scarce coadvisor rather than the main analyst.
• Fugu Ultra Run 19 was attempted as an independent long-running judge but produced no usable judgment. The corrected attempt ran for 10,799.866 seconds and ended exit 3 with diagnostic empty_response, 720 created/in-progress-only events, zero output characters, no final event, and no usage/orchestration evidence. Nothing from Run 19 was integrated; the verified failure disposition is reviews/working/2026-07-14/codex-main-analysis/40-fugu-ultra-eight-pillar-judge-disposition.md.
• Two bounded independent Codex reviews replaced the missing Fugu judgment across five operating seams. Root adjudication accepted a targeted OM-14/OM-15 correction, an explicit truthful no-downstream-work disposition, a permitted-L2 CG12 wording correction, and a 24/7 clinical-transfer-refusal test. The review and adjudication artifacts are 41a-codex-operating-seams-review.md, 41b-codex-longitudinal-seams-review.md, and 41-codex-five-seam-replacement-adjudication.md.

Open decisions preserved#

Manual 25 does not settle the open decisions recorded in 08_Decision_Log.md, including executed Participant Agreement obligations; clinical and medication authority; urgency, retries, correction, refusal, takeover, and contact-attempt policy; Assessment timing; Care Plan change/distribution; caregiver authority and no-caregiver safeguards; service/referral/transition outcomes; evidence and payment treatment; external-data entitlement; and AI participation in calls or outreach.

Verification state#

Verified locally before this checkpoint:

• all eight pillar inventories reconcile to 359 / 359 tasks/cards;
• the canonical matrix contains 48 receiving routes: 40 downstream, four same-interaction, and four context-only;
• all four context-only routes avoid a mandatory human-acceptance gate;
• the manual has no insertion placeholder;
• Initial, annual, and bounded Assessment remain explicit;
• Ongoing remains the longitudinal return engine;
• receiving acceptance remains distinct from downstream completion;
• a stable Ongoing contact uses an open invitation and only relevant follow-up rather than a mandatory all-domain mini-assessment;
• no-downstream-work retains a source-linked reason and re-entry condition instead of dropping a fact or manufacturing a receiving task;
• the bathing micro-service requires the permitted L2 human CG12 act, and 24/7 refusal of clinical transfer does not prove safety or resolution;
• the three worked episodes preserve independent service results and count shared labor once; and
• no new task ID or numeric SLA was introduced by the integration manual.

After the five-seam corrections, the local reviewer website resynchronized 28 Markdown documents. Website lint passed, the production build passed, and the full rendered-page and review-element suites passed 11 / 11 tests. A fresh section-11 parse still returns 48 rows: 40 downstream, four same-interaction, and four context-only. Fugu Ultra Run 19 ended without usable output and therefore supplied no validation or correction.

Exact continuation after this unit#

The next business review is owner adjudication of the eight-pillar service model: clinical authority and safety by authorized U.S. clinicians, privacy/compliance requirements by the named owner, operational feasibility and minutes by service operations, and beneficiary/caregiver burden and value by product/operations. Apply resulting decisions to the affected pillar, rule, edge-case, and decision-log layers together.

Do not begin Fault Gate 2, Fault Gate 17, schemas, manifests, hashes, or automation implementation during that semantic owner-review unit.