--- id: WF-PFM-ELG-019 title: Evaluate computed attention on schedule trigger: >- At hourly minute twenty the scheduler submits due computed-kind evaluations for each workspace and schedule slot. product_area: platform module: eligibility revision: 1 status: draft design_intent: confirmed origin: >- Responsibilities and attention architecture D-03 and D-13 (Platform evaluates only computed kinds, through the query layer), approved by Andrew 2026-10-18 (PLAT-M15 "Decisions"); the job standard (docs/jobs.md) implementation_evidence: observed inference: [] source_sha: services-platform: ad3f6a6582843008ed0f5a17b131878d15829af1 implements: - CAP-PFM-ELG-016 - CAP-PFM-ELG-015 moves: - STATE-PFM-ELG-007 crosses: - SVC-PFM-ELG-attention - SVC-PFM-ELG-responsibilities idempotency: >- Workspace, kind and slot identify a job submission. Item publication refreshes existing keys; unconfirmed resolution runs only after successful enumeration. concurrency: >- Each job is scoped to one workspace-kind-slot, but records publish in individual transactions and a later failure cannot undo earlier items. hops: - 'n': 1 actor_or_service: Evaluation scheduler action: >- Determine hourly, six-hourly or daily eligibility and submit due workspace-kind jobs. anchor: >- services-platform:modules/attention/services/evaluate/submit-evaluations.ts:submitDueEvaluations seam: schedule on_failure: >- Log each submission failure and continue other pairs; the silence alarm detects missing scheduler invocation. left_inconsistent: A missed pair can remain stale without the global silence alarm firing. observer: >- Scheduler logs, dead-letter handling and invocation silence show different failure classes. provenance: observed source: >- services-platform:modules/attention/services/evaluate/submit-evaluations.ts:evaluationSlot - 'n': 2 actor_or_service: Evaluation state machine and task action: Start a run and evaluate the kind’s criteria or internal source. anchor: >- services-platform:modules/attention/services/evaluate/evaluate-definition.ts:evaluateDefinition seam: state_machine on_failure: Exceptions during evaluation mark failure and trigger execution alarms. left_inconsistent: Items already committed remain; old items are not cleared by a failed run. observer: State-machine failure and timeout alarms expose execution. provenance: observed source: >- services-platform:modules/attention/services/evaluate/evaluation-workflow.ts:attentionEvaluationWorkflow - 'n': 3 actor_or_service: Evaluation task (uncovered areas) action: >- For uncovered responsibilities, orchestrate advisory publication from gap rows and resolve absent gaps after successful publication. anchor: >- services-platform:modules/attention/services/evaluate/evaluate-definition.ts:runInternal seam: state_machine on_failure: Evaluation exceptions follow the run failure path. left_inconsistent: A partial publication remains until later evaluation. observer: Evaluation alarms expose incomplete work. provenance: observed source: >- services-platform:modules/attention/services/evaluate/unassigned-responsibilities.ts:evaluateUnassignedResponsibilities - 'n': 4 actor_or_service: Evaluation task (dataset kinds) action: >- Read dataset pages of one thousand, apply settings to criteria, project item fields, publish matches and resolve unconfirmed items only after finishing. anchor: >- services-platform:modules/attention/services/evaluate/evaluate-definition.ts:itemOf seam: state_machine on_failure: >- Beyond one hundred thousand matches settle over capacity without clearing; exceptions fail, and unsupported eligibility audiences are refused. left_inconsistent: Earlier items remain and later matches beyond capacity are not published. observer: The over-capacity alarm distinguishes this outcome from failure. provenance: observed source: 'services-platform:docs/attention.md#7. The sweep and the evaluator' branches: - >- Daily uncovered-area evaluation is scheduled at 07:20 independently of responsibility reconcile completion. waits: [] evidence: - >- services-platform at ad3f6a6: modules/attention/services/evaluate/; services/persistence/evaluation-runs.ts; aws/logic/attention-evaluation.lambda.ts, attention-evaluation.step-function.ts; docs/attention.md §7; docs/jobs.md. controls: [TC-AVL-005, TC-AZ-001, TC-AZ-002, TC-AZ-003, TC-AZ-004, TC-CHG-001, TC-CHG-002, TC-CHG-003, TC-CHG-004, TC-CHG-005, TC-CHG-006, TC-CHG-007, TC-CRY-001, TC-CRY-002, TC-CRY-003, TC-CRY-004, TC-DAT-001, TC-DAT-002, TC-DAT-003, TC-DAT-004, TC-IAM-001, TC-IAM-002, TC-IAM-004, TC-IAM-005, TC-IAM-006, TC-LOG-002, TC-LOG-003, TC-LOG-004, TC-LOG-005, TC-MON-001, TC-MON-002, TC-PI-001, TC-PI-002, TC-PI-003, TC-PI-005, TC-PRV-001, TC-PRV-002, TC-TEN-001, TC-TEN-002, TC-TEN-003, TC-TEN-004, TC-VEN-001, TC-VEN-002, TC-VEN-003] --- # Evaluate computed attention on schedule ## What It Does At hourly minute twenty the scheduler submits due computed-kind evaluations for each workspace and schedule slot. ## The Flow ```mermaid flowchart TD H1["Evaluation scheduler" ] H2["Evaluation state machine and task" ] H1 --> H2 H3["Evaluation task (uncovered areas)" ] H2 --> H3 H4["Evaluation task (dataset kinds)" ] H3 --> H4 ``` Determine hourly, six-hourly or daily eligibility and submit due workspace-kind jobs. Start a run and evaluate the kind’s criteria or internal source. For uncovered responsibilities, orchestrate advisory publication from gap rows and resolve absent gaps after successful publication. Read dataset pages of one thousand, apply settings to criteria, project item fields, publish matches and resolve unconfirmed items only after finishing. ## When Something Fails Log each submission failure and continue other pairs; the silence alarm detects missing scheduler invocation. A missed pair can remain stale without the global silence alarm firing. Exceptions during evaluation mark failure and trigger execution alarms. Items already committed remain; old items are not cleared by a failed run. Evaluation exceptions follow the run failure path. A partial publication remains until later evaluation. Beyond one hundred thousand matches settle over capacity without clearing; exceptions fail, and unsupported eligibility audiences are refused. Earlier items remain and later matches beyond capacity are not published. ## Open Design Questions Daily uncovered-area evaluation is scheduled at 07:20 independently of responsibility reconcile completion.