OKF-DWP

Independent experimental publication. Not an official DWP service, benefits advice or an entitlement calculator.

View this version’s Markdown source

On this page

Delivery and acceptance work packages

Generated from the public backlog register. Delivery, independent human review and external permission are separate. A completed proposal or model trial does not establish specialist acceptance.

Backlog overview · Team handover

DWP-BL-001: Extensive persona, journey and question review

Priority P0; aggregate status needs_domain_review.

Work package Kind Status Executor Next action or retained result
DWP-BL-001.delivery implementation recorded_complete agent Delivered bounded increment; use the linked evidence and named remaining obligations for the next reviewed expansion.
DWP-BL-001.acceptance independent_review waiting human Validate projected personas and intended tasks with staff and user researchers.

Evidence: docs/next-stage/stakeholder-needs-and-evaluation.md · evaluation/staff-questions/cases.json · evaluation/staff-needs/matrix.json · evaluation/staff-needs/README.md

DWP-BL-002: Benefits engine feasibility

Priority P2; aggregate status not_started.

Work package Kind Status Executor Next action or retained result
DWP-BL-002.delivery implementation not_started agent Produce one non-executable rule packet after the evidence dependencies are reviewed.
DWP-BL-002.acceptance independent_review waiting human Approve legal conditions and operational authority before any calculator.

Evidence: docs/next-stage/semantic-contract.md

DWP-BL-003: Application and change-of-circumstances journeys

Priority P2; aggregate status not_started.

Work package Kind Status Executor Next action or retained result
DWP-BL-003.delivery implementation not_started agent Design separate synthetic apply and change journeys after evidence dependencies.
DWP-BL-003.acceptance independent_review waiting human Validate service ownership, privacy and assisted access with users.

Evidence: docs/next-stage/stakeholder-needs-and-evaluation.md

DWP-BL-004: Frozen full DMG and ADM capture

Priority P0; aggregate status recorded_complete.

Work package Kind Status Executor Next action or retained result
DWP-BL-004.delivery implementation recorded_complete agent Preserve and verify the frozen 513-PDF, 19,090-page acquisition.

Evidence: source/full-dmg-2026-09-15/inventory.json · source/adm-2026-09-19/inventory.json · context/corpus/manifest.json

DWP-BL-005: Broader semantic modelling: neutral domain concepts and relationships

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-005.delivery implementation recorded_complete agent Delivered bounded increment; use the linked evidence and named remaining obligations for the next reviewed expansion.
DWP-BL-005.domain-expansion research in_progress agent The current source increment has 57 authored concepts and 74 conditional support dependencies, including connected international qualifications. Continue benefit/variant, ADM, legal and temporal/territorial coverage. This is incomplete implementation as well as pending specialist review; graph size is not answerability.
DWP-BL-005.abroad-connectivity implementation recorded_complete agent Bounded source implementation and independent agent source review complete: reused international IDs, overseas aliases and whole pages 8–14; 24 assemblies and 10 negative controls pass. At 512 KiB four Pension Credit cases retain 7/7 pages and 14/14 paths. At 32 KiB the general question loses its source page and PC cases refuse on metadata. Public-service admission, capacity repair, other-benefit coverage and specialist acceptance remain separate.
DWP-BL-005.cross-benefit-coverage research in_progress agent Map missing benefit and variant concepts, including Universal Credit and other ADM international routes. For each, identify territory, date, purpose, duration, dependencies and conflicts from frozen sources before authoring assertions. Require source-backed definitions and held-out specialist evidence checks; lexical page presence is not semantic coverage. The first parallel Data Analytics review returned unreviewed duration, purpose and qualifying-young-person proposals; import its complete artefacts, verify the reported hashes and inspect every condition before considering integration. This frozen-file exercise did not call Ask OKF.
DWP-BL-005.acceptance independent_review waiting human Review definitions, confusions, qualifications and assertion meanings; publication of proposals is not acceptance.
DWP-BL-005.logical-units implementation recorded_complete agent Retain the verified additive implementation and its explicit limits; broader boundary review, profile migration, compact delivery and specialist acceptance remain separate unfinished packages.
DWP-BL-005.remaining-unit-boundaries research in_progress agent The additive reviewed source slice now contains 75 authored boundaries (116,202 exact source bytes); 49,551 machine candidates remain uncertain. Follow the 40-case coverage ledger to prioritise Pension Credit core conditions, household and carer dependencies, then further ADM topics. Source-byte preservation and candidate overlap are not semantic or specialist acceptance.
DWP-BL-005.source-bound-navigation implementation in_progress agent The original notional-capital miss is repaired by opt-in scoped navigation. Preserve the 5/6 development and 3/4 confirmation results, both with 0/5 negative selections, and the transfer-wording miss; broader independently reviewed wording coverage remains open.

Evidence: docs/next-stage/semantic-contract.md · docs/ontology-use.md · evaluation/semantic-expansion/build.json · evaluation/semantic-expansion/catalogue.json · docs/semantic-expansion.md · evaluation/semantic-expansion/profiles.json · docs/monday-delivery-work-log.md · docs/household-evidence-expansion.md · validation/qualification-context/2026-09-21/README.md · docs/carehome-component-dependency-review.md · docs/partner-addition-qualification-review.md · validation/partner-context/2026-09-21/README.md · validation/partner-context/2026-09-21/protocol.json · validation/partner-context/2026-09-21/attempt-01/comparison.json · docs/ignored-person-qualification-review.md · validation/ignored-person-context/2026-09-21/README.md · validation/ignored-person-context/2026-09-21/protocol.json · validation/ignored-person-context/2026-09-21/attempt-01/comparison.json · docs/abroad-semantic-audit.md · docs/data-agent-semantic-review.md · domain-profile/staff-semantic/concepts.yamlld · scripts/test_abroad_semantic.py · validation/abroad-context/2026-09-21/README.md · docs/logical-evidence-units.md · docs/logical-units-work-log.md · logical-units/manifest.json · evaluation/logical-units/run/summary.json · scripts/test_logical_context.py · docs/semantic-closure-and-compact-delivery.md · evaluation/semantic-coverage/closure-2026-09-22/audit.json · domain-profile/logical-units/authoring-registry.json · docs/evidence-workbench.md · docs/evidence-workbench-work-log.md · domain-profile/evidence-connect/profile.json · structured-context/evidence-connect-manifest.json · evaluation/evidence-workbench/manifest.json · scripts/build_evidence_connect.py · scripts/build_evidence_workbench.mjs

DWP-BL-006: Legislation, regulations and case-law reconciliation

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-006.delivery research recorded_complete agent Delivered bounded increment; use the linked evidence and named remaining obligations for the next reviewed expansion.
DWP-BL-006.legal-body-evidence research in_progress agent Twenty selected statutory units and 43 navigation relationships are implemented with dated sources and retained failures. Next resolve remaining statutory dependencies, 12 unknown extents, commencement/effects and the missing judgment bodies; do not promote extraction to applicability. For the new UC/DLA case, obtain and verify the exact operative Schedule 1 paragraph 31 and UC regulation 24 bodies, amendments, territory and effective dates; then test whether A4361 governs this disabled-child addition, rather than inferring it from the carer-element example. The reading-help bridge separately retains three Chapter 60 targets as four observed geographical variants, with the original failed attempt preserved. The optional context passes structural and governance admission, but the fixed-budget ordinary query does not select the bridge. Public dated-navigation checks now pass; substantive applicability and ordinary-query selection remain unresolved. The separate UK-legislation format candidate needs a successor release contract and current assurance before publication; it is not a prerequisite for the DWP presentation increment.
DWP-BL-006.acceptance independent_review waiting human Review applicability, commencement, amendments and completeness for each substantive conclusion.

Evidence: docs/next-stage/legislation-integration.md · knowledge/legislation-resources.yamlld · evaluation/legal-reconciliation/coverage.json · evaluation/legal-reconciliation/assertions.json · docs/legal-reconciliation.md · docs/monday-delivery-work-log.md · docs/legal-body-evidence.md · domain-profile/legal-bodies/coverage.json · evaluation/out-of-sample/uc-disabled-child-supersession-001.json · docs/uc-disabled-child-supersession-follow-up.md · docs/reading-help-legislation-bridge.md · domain-profile/reading-help-law/bridge.json · docs/reading-help-rollout-work-log.md · domain-profile/reading-help-law/admission.json · evaluation/reading-help-rollout/law-context-probe.json · reading-help-ch60-law.json · docs/reading-help-release-2026-09-27.md · evaluation/reading-help-rollout/release-20260927/manifest.json · evaluation/reading-help-rollout/release-20260927/public-attempt-03/receipt.json

DWP-BL-007: Broader semantic modelling: task-specific evidence profiles

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-007.delivery implementation recorded_complete agent Delivered bounded increment; use the linked evidence and named remaining obligations for the next reviewed expansion.
DWP-BL-007.evidence-closure implementation in_progress agent Continue all 43 open source-closure obligations. The ignored-person extension exposes a larger requirement set that exceeds 512 KiB despite 177/177 candidate overlap. Preserve all missing requirements and distinguish source coverage, context capacity, legal applicability and specialist acceptance.
DWP-BL-007.qualification-retention implementation recorded_complete agent Bounded increment verified: final DWP 7f9feb96 and Explorer c4f2de0a retain 433/433 declared path occurrences and 177/177 candidate occurrences at both budgets; Staff 012/013 retain all seven household pages. All 320 assemblies and exact replay pass, with 14 offline controls. The 64 KiB Staff 012 metadata refusal, optional temporary dependency at 512 KiB and all 203 obligations remain explicit. Publication and larger component sets are separate packages; no historical answer is regraded.
DWP-BL-007.qualification-publication live_observation in_progress agent Qualification source 7f9 passed protected-main validation, Pages and actual public Chrome checks; learning revision 91b also passed its actual 123-response byte/link audit. Disability PR 18 and its actual immutable-candidate Chrome observation are separate. Complete protected publication and fresh public/source/service checks for later increments; preserve every earlier receipt and live 0.5.0 identity.
DWP-BL-007.component-qualification-expansion implementation recorded_complete agent The bounded overview/no-partner increment is authored and independently reviewed: 903 records, 1,464 assertions, 29 support dependencies, 46 semantic and 20 Reader controls pass. All 320 comparison assemblies and exact replay pass: 497/497 declared path occurrences at 512 KiB, 407/497 at 256 KiB. Required-support gaps remain explicit; all 203 obligation IDs/statuses remain open. Partner and ignored-person extensions are separate work packages; public service and specialist acceptance remain separate.
DWP-BL-007.partner-qualification implementation recorded_complete agent The bounded partner increment is independently source-reviewed: ten dependencies, conditional Staff 012/013/014/017 paths and unchanged 018. Combined Reader, 40-case evaluation, 52 semantic and 20 Reader controls pass. Frozen 320-assembly comparison and exact replay retain 585/585 declared path occurrences at 512 KiB and 449/585 at 256 KiB. All 203 obligations remain open; publication and specialist acceptance are separate.
DWP-BL-007.ignored-person-qualification implementation recorded_complete agent Bounded source modelling is independently reviewed: two concepts, 17 whole support pages (eight newly selected), 913 records, 1,526 assertions, 61 requires, and 61 focused controls. Existing evidence and all 203 obligations are unchanged. Larger context retention fails to preserve all declared paths at 512 KiB; that capacity problem, public delivery and specialist acceptance remain separate.
DWP-BL-007.qualification-capacity implementation in_progress agent Investigate staged component context or reversible representation to retain the larger ignored-person closure without raising budgets, dropping required paths or disguising omissions. Compare required record bytes, repeated path metadata and delivered selected evidence. Keep Monday service/trial source fixed while evaluating this separate regression. The abroad increment also loses the general question’s source page at 32 KiB, while the PC task refuses on metadata. Preserve these observations and test a generic selection/representation repair before changing the public default.
DWP-BL-007.abroad-qualification implementation recorded_complete agent Bounded source implementation and independent agent source review complete: reused international IDs, overseas aliases and whole pages 8–14; 24 assemblies and 10 negative controls pass. At 512 KiB four Pension Credit cases retain 7/7 pages and 14/14 paths. At 32 KiB the general question loses its source page and PC cases refuse on metadata. Public-service admission, capacity repair, other-benefit coverage and specialist acceptance remain separate.
DWP-BL-007.task-discrimination implementation in_progress agent Trial 06 applies reviewed conjunctive guards and activates source-read selection for 37/40 unchanged staff occurrences. Retain missing-benefit and ambiguous-SDA cases 001/008/018; do not infer legal applicability from lexical matches or a guard. Scope and specialist review remain open.
DWP-BL-007.uc-supersession-classification implementation in_progress agent Keep the failed UC/DLA replay intact. Classify the change as a conditional other-benefit supersession candidate only after BL006 checks the dated law; map F1123 and child-responsibility conditions to A4352’s explicit A4361 exception and A4361’s full examples with source-bound dependencies. Freeze independent positive phrasings and negative controls for lower DLA rate, changed entitlement date, child outside the UC family, within-period notice and ordinary late change. Replay selection, exact spans, task activation and bounded delivery separately; leave assessment-period dates unknown and evidence_status insufficient until complete source closure and specialist review.
DWP-BL-007.acceptance independent_review waiting human Accept the evidence requirements and qualifications independently before marking broader answers sufficient.
DWP-BL-007.unit-qualification implementation recorded_complete agent Retain the verified additive implementation and its explicit limits; broader boundary review, profile migration, compact delivery and specialist acceptance remain separate unfinished packages.
DWP-BL-007.remaining-unit-profiles research in_progress agent Source-read State Pension, IIDB/PIP/overlap and carer increments bring source-read activation to 37/40, retaining 47/47 inherited, 181/181 previous and 208/208 expanded paths at 512 KiB. Complete independent trial review and publication. Resolve the three remaining underspecified task families only with explicit scope. All 203 earlier obligations remain open.
DWP-BL-007.discovery-cards implementation in_progress agent Trial 06 retains source and relationships in 40/40 large packages and 436/436 declared source-selection path incidences. Preserve 32 KiB zero evidence and location trade-offs: 469/765 restored and 34/87 inferred. Original custody delivery has a separate failure; complete its bounded repair and public adoption. Cards remain literal previews, not legal summaries.
DWP-BL-007.discovery-ranking-probes research recorded_complete agent Both diagnostic probes and independent checks remain frozen. Token alias expansion regressed two locations; fixed BM25 improved four candidate locations but did not establish relevance. The later source-led implementation has extractive discovery cards and its own tests; do not relabel this earlier probe as a model-summary or answer-quality result.
DWP-BL-007.manual-guide implementation in_progress agent Local implementation and independent review complete: 21 scoped conventions, 30 source-page and four PDF-tree excerpts, all 513 documents accounted for. Complete contract/CI/publication acceptance; preserve the 25 untagged observations and one output-limit fallback. This guide does not certify legal applicability.
DWP-BL-007.structured-corpus implementation in_progress agent All 53,727 units and original text from 513 documents are accounted for. Global check 04 passes 1,355 input bindings. The Reader alias/index repair loads 54,578 records and 19,193 relationships within existing limits. Complete exact CI, publication and browser verification; machine boundaries remain uncertain.
DWP-BL-007.remaining-structural-review research in_progress agent The isolated successor and 40-case equal-budget replay are implemented: 15 target boundaries corrected, 12 partial and one unresolved. Matched declared candidate-page coverage and requirements do not improve; all packages remain insufficient. Review the residual structural findings and source-bound suggestions independently before any separate adoption. The broad parked proposal, frozen sources and earlier replay are preserved. The 27 September non-regression recheck reproduces all 55 review files, passes four tests, and confirms exact hashes and recorded spans for the nine previously reviewed substantive chapters. Case 014 remains unresolved; no new visual or specialist review is implied.
DWP-BL-007.original-acceptance-migration implementation in_progress agent Preserve the original imprisonment failure and the separate continuation-allocation patch retained after the 23 September freeze. This work is outside the supplied forty-question brief and must pass its own new fixed-source evaluation before adoption.
DWP-BL-007.capital-pilot implementation recorded_complete agent Bounded source repair, 21 Python controls and fixed-question attempt-03 delivered: ten groups, two boundary repairs and typed Appendix 1. Larger packages retain all declared spans; the baseline misses only six whitespace bytes. No answer gain is established. Additive adoption and bounded delivery are recorded separately; 11 dependency groups and specialist review remain open.
DWP-BL-007.capital-delivery implementation recorded_complete agent All 40 retained public contexts reconstruct exactly through 673 parts of at most 32 KiB. The deployed version 3 site matches the merged commit byte for byte; this is bounded delivery, not evidence sufficiency or answer quality. Recheck after any source or consumer change.
DWP-BL-007.capital-dependencies research in_progress agent Four captured dependency groups now have exact source-entry links in the additive overlay. Eleven declared dependency groups remain open overall because source, scope or specialist review is incomplete; do not infer closure from lexical selection or public navigation.
DWP-BL-007.capital-adoption implementation recorded_complete agent The ten repaired capital passages are adopted in the additive full-DMG/ADM overlay and the live opt-in Reader-to-Ask descriptor. The public capital journey selects U07, the DMG 84861 entry and their direct link, but remains insufficient and truncated. Frozen corpora, pilot and service defaults are unchanged; specialist review stays open.
DWP-BL-007.capital-specialist-review independent_review not_started human Review proposed groups, summaries, dependencies and typed table for legal applicability; engineering checks do not establish complete benefits answers.
DWP-BL-007.passage-review-workbench implementation recorded_complete agent Merged Explorer PR 155 and DWP PR 44 deliver the generic 28-case inspection and isolated-preview workflow. Public checks verify all 34 app files, all 28 case previews, PDF rendering/page synchronisation, explicit unresolved status, successor reference table and local suggestion export. Full document/corpus impacts of an edited correction remain unknown without wider replay. Structural residuals and separate reviewed adoption remain open.
DWP-BL-007.liteparse-source-pilot research recorded_complete agent The frozen 12-window local LiteParse 2.14.6 comparison is complete: 8/12 literal anchor checks pass; both complete table-association checks fail. Preserve the failed gate, punctuation/spacing diagnosis and unchanged corpus defaults. The 24 fresh windows and 40-question replay were not admitted. A spatial sidecar with explicit header, partial-table and reversible character alignment needs a new reviewed protocol; no retrieval or answer-quality gain is claimed.

Evidence: evaluation/context-assembly/imprisonment-case.json · evaluation/staff-questions/cases.json · evaluation/semantic-expansion/profiles.json · evaluation/semantic-expansion/evaluation.json · docs/semantic-expansion.md · docs/monday-delivery-work-log.md · docs/household-evidence-expansion.md · docs/context-performance.md · docs/household-qualification-budget-review.md · domain-profile/staff-semantic/concepts.yamlld · domain-profile/staff-semantic/profiles.yamlld · scripts/test_staff_semantic.py · docs/carehome-component-dependency-review.md · validation/qualification-context/2026-09-21/README.md · validation/qualification-context/2026-09-21/attempt-01/comparison.json · validation/qualification-context/2026-09-21/protocol.json · docs/disability-addition-qualification-review.md · validation/disability-context/2026-09-21/README.md · validation/disability-context/2026-09-21/protocol.json · validation/disability-context/2026-09-21/attempt-01/comparison.json · validation/household-reader-public/7f9feb96-c4f2de0a/README.md · docs/partner-addition-qualification-review.md · validation/partner-context/2026-09-21/README.md · validation/partner-context/2026-09-21/protocol.json · validation/partner-context/2026-09-21/attempt-01/comparison.json · docs/ignored-person-qualification-review.md · validation/ignored-person-context/2026-09-21/README.md · validation/ignored-person-context/2026-09-21/protocol.json · validation/ignored-person-context/2026-09-21/attempt-01/comparison.json · docs/abroad-semantic-audit.md · docs/data-agent-semantic-review.md · scripts/test_abroad_semantic.py · validation/abroad-context/2026-09-21/README.md · docs/logical-evidence-units.md · docs/logical-units-work-log.md · logical-units/manifest.json · evaluation/logical-units/run/summary.json · scripts/test_logical_context.py · domain-profile/logical-units/profiles.json · domain-profile/logical-units/reference-review.json · docs/semantic-closure-and-compact-delivery.md · evaluation/semantic-coverage/closure-2026-09-22/audit.json · domain-profile/logical-units/authoring-registry.json · domain-profile/logical-units/closure-dependencies.yamlld · domain-profile/logical-units/closure-pip-transition.yamlld · domain-profile/logical-units/closure-profile-dependencies.yamlld · docs/discovery-cards-experiment.md · evaluation/semantic-coverage/discovery-card-diagnosis.json · evaluation/discovery-probe/README.md · evaluation/discovery-probe/protocol.json · evaluation/discovery-probe/independent-review.md · evaluation/discovery-probe/length-ranking.md · evaluation/discovery-probe/length-protocol.json · evaluation/discovery-probe/length-independent-review.md · scripts/probe_discovery_terminology.mjs · scripts/probe_discovery_length.mjs · docs/manual-structure-process.md · docs/source-led-manual-guide.md · docs/manual-structure-evaluation.md · evaluation/manual-structure/runs/candidate-expanded-04/report.json · evaluation/manual-structure/context-probe/protocol.json · manual-guide/manifest.json · pdf-structure/bounded-v2/manifest.json · scripts/test_manual_guide.py · scripts/test_pdf_structure.py · structured-units/manifest.json · scripts/test_manual_structure_review.py · evaluation/manual-structure/context-probe/runs/attempt-01/README.md · evaluation/manual-structure/runs/candidate-expanded-06/report.json · evaluation/manual-structure/auxiliary-review/README.md · evaluation/manual-structure/context-probe/runs/attempt-02/README.md · evaluation/manual-structure/context-probe/runs/attempt-02/comparison.json · domain-profile/structured-units/pc-core.yamlld · domain-profile/structured-units/household-routing.yamlld · domain-profile/structured-units/legacy-location-policy.yamlld · scripts/test_structured_context_profiles.py · scripts/test_structured_location_migration.py · evaluation/manual-structure/runs/candidate-expanded-07/report.json · evaluation/manual-structure/final-source-review/large-paragraph-triage.json · evaluation/manual-structure/final-source-review/manifest-review.json · evaluation/manual-structure/context-probe/runs/attempt-05/README.md · evaluation/manual-structure/context-probe/reports/attempt-05/README.md · docs/source-led-demonstration.md · docs/source-led-results.md · evaluation/manual-structure/context-probe/runs/attempt-06/report.json · evaluation/manual-structure/context-probe/analysis-06/comparison.json · evaluation/manual-structure/context-probe/corpus-check-04.json · domain-profile/structured-units/state-pension-routing.yamlld · domain-profile/structured-units/benefit-interactions.yamlld · domain-profile/structured-units/custody-restoration.yamlld · docs/original-acceptance-restoration.md · evaluation/manual-structure/large-paragraph-review-2026-09-23/README.md · evaluation/manual-structure/large-paragraph-review-2026-09-23/findings.json · evaluation/manual-structure/context-probe/analysis-06/independent-reading.md · docs/pc-capital-bounded-repair.md · evaluation/capital-pilot/protocol.json · evaluation/capital-pilot/runs/attempt-01/report.json · capital-pilot/repair-receipt.json · capital-pilot/groups.json · evaluation/capital-pilot/runs/attempt-03/report.json · scripts/test_capital_pilot.py · docs/evidence-workbench.md · docs/evidence-workbench-work-log.md · domain-profile/evidence-connect/profile.json · structured-context/evidence-connect-manifest.json · evaluation/evidence-workbench/manifest.json · scripts/build_evidence_connect.py · scripts/build_evidence_workbench.mjs · validation/evidence-workbench/2026-09-24/public-release/site/observation.json · validation/evidence-workbench/2026-09-24/public-release/browser/receipt.json · docs/evidence-workbench-handover.md · validation/evidence-workbench/2026-09-24/local-final/reader-ask.json · evaluation/out-of-sample/uc-disabled-child-supersession-001.json · docs/uc-disabled-child-supersession-follow-up.md · scripts/replay_uc_supersession_case.mjs · docs/unmerged-work-review-2026-09-24.md · docs/passage-boundary-review.md · domain-profile/passage-boundary-review/protocol.json · evaluation/passage-boundary-review/manifest.json · evaluation/passage-boundary-review/identity-migration.json · scripts/build_passage_boundary_review.py · scripts/test_passage_boundary_review.py · evaluation/passage-boundary-candidate/question-replay.json · domain-profile/passage-boundary-review/replay-protocol.json · validation/passage-boundaries/2026-09-25/README.md · docs/liteparse-pilot.md · domain-profile/liteparse-pilot/protocol.json · domain-profile/liteparse-pilot/cases.json · validation/liteparse-pilot/source-review-2026-09-26/observation.json · evaluation/liteparse-pilot/stage-1-001/report.json · evaluation/liteparse-pilot/stage-1-001/REPORT.md · validation/liteparse-pilot/README.md · scripts/test_liteparse_pilot.py · docs/reading-help-release-2026-09-27.md · evaluation/reading-help-rollout/release-20260927/manifest.json · evaluation/reading-help-rollout/release-20260927/public-attempt-03/receipt.json · evaluation/reading-help-rollout/release-20260927/boundary-recheck-01/receipt.json

DWP-BL-008: Progressive evidence manifests and exact reads

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-008.delivery implementation recorded_complete agent Retain exact compact catalogue/read reconstruction and public deployment receipts; host console work is BL023.
DWP-BL-008.engine-replay implementation recorded_complete agent Delivered explicit engine identity, source-engine allowlists and historical expected-context replay. The public 0.6.0 SDK run passed 121 bounded requests across 11 cases; runtime and verifier identities remain distinct. Preserve all earlier receipts and require a fresh check for future changes.
DWP-BL-008.client-connection live_observation in_progress agent The existing native task now calls compact tools: after metadata refresh with permissions unchanged, eleven contiguous diagnostics reads reconstruct the returned hash. Preserve the earlier stale-schema rejection and initial catalogue. Test the intended fresh Work/Data Agent task and Voice separately; a transport pass is not an evidence-completeness or answer-quality result.
DWP-BL-008.retained-evidence-resources implementation in_progress agent The independently reviewed generic exporter and DWP publisher now preserve bounded, hash-checked complete packages and small resource reads. Eleven exporter and 23 publisher controls pass, with a separate local Chrome observation. Publish the three explicitly approved examples, verify their public bytes and retain the observed boundary; anonymous questions are not recorded.
DWP-BL-008.logical-compact-admission implementation recorded_complete agent Service 0.7.0 admits the immutable Evidence Connect source with its compatible frozen engine. The exact merged build passed twelve public SDK cases in 135 requests across ten permitted pairs, preserving historical replay and bounded exact reconstruction. Keep inline assembly omissions, source boundaries and insufficient status explicit. Client-specific acceptance remains in client-connection; specialist acceptance is not implied.
DWP-BL-008.workbench-delivery implementation recorded_complete agent The approved 40 public staff-question occurrences and exact hash-bound packages are delivered through the deployed workbench. The public browser run checked the 40-entry catalogue, three representative cases, deep links and local proposal export. No private question logging or specialist acceptance is claimed.

Evidence: docs/compact-evidence-demo.md · validation/compact-delivery/v0.3.1/deployment.json · validation/compact-delivery/v0.3.1/sdk-receipt.json · validation/compact-delivery/v0.3.1/browser/full-corpus/chrome-receipt.json · validation/compact-delivery/v0.3.1/browser/historical/run-summary.json · validation/compact-delivery/deployment.json · validation/compact-delivery/sdk-receipt.json · validation/compact-client/verification.json · validation/compact-delivery/v0.4.0/deployment.json · validation/compact-delivery/v0.4.0/sdk-receipt.json · validation/compact-delivery/v0.5.0/deployment.json · validation/compact-delivery/v0.5.0/sdk-receipt.json · validation/compact-delivery/v0.5.0/browser/staff/run-summary.json · validation/compact-delivery/v0.5.0/artifacts.json · docs/household-service-verification.md · validation/compact-delivery/v0.5.0/README.md · validation/qualification-context/2026-09-21/README.md · docs/methodology.md · docs/retained-evidence-publication.md · validation/compact-delivery/v0.6.0/README.md · docs/chatgpt-connection.md · validation/compact-delivery/v0.6.1/README.md · validation/client-connection/2026-09-21/README.md · evaluation/logical-units/run/summary.json · docs/logical-evidence-units.md · docs/semantic-closure-and-compact-delivery.md · evaluation/semantic-coverage/closure-2026-09-22/audit.json · domain-profile/logical-units/authoring-registry.json · docs/evidence-workbench.md · docs/evidence-workbench-work-log.md · domain-profile/evidence-connect/profile.json · structured-context/evidence-connect-manifest.json · evaluation/evidence-workbench/manifest.json · scripts/build_evidence_connect.py · scripts/build_evidence_workbench.mjs · validation/evidence-workbench/2026-09-24/public-release/site/observation.json · validation/evidence-workbench/2026-09-24/public-release/browser/receipt.json · validation/evidence-workbench/2026-09-24/public-release/files.json · docs/evidence-workbench-handover.md · validation/compact-delivery/v0.7.0/README.md · validation/compact-delivery/v0.7.0/deployment.json · validation/compact-delivery/v0.7.0/sdk/attempt-01/observation.json · validation/client-connection/2026-09-25/remote/README.md · validation/client-connection/2026-09-25/remote/observation.json · validation/client-connection/2026-09-25/remote/after-refresh.json

DWP-BL-009: Conceptual classification and DMG Reader navigation

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-009.lexical-version-competition research not_started agent Measure relevant current chapter and historical amendment/memo competition with fixed source versions and questions. Investigate date, source-family and topic signals without silently discarding history or inferring legal supersession from similar text. Report relevance and qualification retention separately from page overlap.
DWP-BL-009.delivery implementation recorded_complete agent Retain both-manual classification accounting and verified DMG Reader journeys; combined Reader work is BL024.
DWP-BL-009.logical-corpus implementation recorded_complete agent Retain the verified additive implementation and its explicit limits; broader boundary review, profile migration, compact delivery and specialist acceptance remain separate unfinished packages.

Evidence: full-dmg/okf-review-context.json · validation/navigation/build.json · validation/navigation/browser/README.md · validation/navigation/browser/public/README.md · validation/navigation/browser/public/observation.json · validation/navigation/browser/public/deployment.json · evaluation/staff-review/manifest.json · docs/abroad-semantic-audit.md · docs/data-agent-semantic-review.md · docs/logical-evidence-units.md · docs/logical-units-work-log.md · logical-units/manifest.json · evaluation/logical-units/run/summary.json · scripts/test_logical_context.py

DWP-BL-010: Fixed-evidence claim-level model trials

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-010.delivery implementation recorded_complete agent Delivered bounded paired subscription-client experiment with identical authored inputs, retained failures, usage observations and per-claim model critique. Specialist assessment and a controlled model/cost benchmark remain separate.
DWP-BL-010.household-trials implementation in_progress agent Four frozen direct-v4 attempts and independent agent review are retained: both controls and the paired Staff 012 answers pass mechanical checks, but one answer omits selected exceptions and overstates an evidence gap. Preserve all outputs; define a separate qualification-repair experiment without editing these answers. Specialist assessment and a controlled model/cost benchmark remain separate.
DWP-BL-010.acceptance independent_review waiting human Assess entailment, exceptions, dates and abstention independently; model review is not human acceptance.
DWP-BL-010.demo-one-pair research recorded_complete agent Four authorised answers are retained with exact citations, format failures and claim-review concerns. No more calls in this freeze. A later experiment must separately test smaller model-facing diagnostics and complete quotation/qualification retention; no general accuracy or cost claim is established.

Evidence: evaluation/answer-review/README.md · evaluation/answer-review/cases.json · docs/staff-model-trials.md · evaluation/model-comparison/staff-2026-09-20/protocol.json · validation/model-comparison/staff-2026-09-20/results.json · validation/model-comparison/staff-2026-09-20/claim-level-model-critique.json · docs/monday-model-trials.md · evaluation/model-comparison/household-2026-09-21/protocol.json · evaluation/model-comparison/household-compact-v2-candidate/frozen-manifest.json · validation/model-comparison/household-compact-v2/results.json · validation/model-comparison/household-compact-v2/claim-level-model-critique.json · validation/model-comparison/household-compact-v2/README.md · validation/qualification-context/2026-09-21/README.md · validation/model-comparison/household-direct-v3/README.md · validation/model-comparison/household-direct-v4/README.md · validation/model-comparison/household-direct-v4/independent-claim-review.md · validation/model-comparison/household-direct-v4/independent-claim-review.json · docs/demo-one-pair-results.md · validation/demo-1-freeze/comparison/observations.json · docs/demo-one-pair-protocol.md

DWP-BL-011: Source dates and provenance presentation

Priority P1; aggregate status recorded_complete.

Work package Kind Status Executor Next action or retained result
DWP-BL-011.delivery implementation recorded_complete agent Preserve verified source and capture date roles and precision.

Evidence: docs/evidence/cpag-publication-date-review.json · validation/corpus-questions/public-explorer-observation.json

DWP-BL-012: CPAG substantive content access

Priority P1; aggregate status needs_external_permission.

Work package Kind Status Executor Next action or retained result
DWP-BL-012.delivery implementation not_started agent Acquire substantive CPAG content only within separately established rights.
DWP-BL-012.acceptance external_permission waiting external Obtain applicable written permission for intended acquisition, AI processing and redistribution.
DWP-BL-012.permission-scope research recorded_complete agent Owner may review and send the prepared request; no permission or message is implied.

Evidence: docs/cpag-handbook.md · docs/cpag-permission-request-draft.md

DWP-BL-013: Tribunal decision discovery

Priority P1; aggregate status needs_domain_review.

Work package Kind Status Executor Next action or retained result
DWP-BL-013.delivery research recorded_complete agent Delivered bounded increment; use the linked evidence and named remaining obligations for the next reviewed expansion.
DWP-BL-013.acceptance independent_review waiting human Assess precedential weight and question-specific applicability; search matches are not legal conclusions.

Evidence: docs/future-sources.md · evaluation/legal-reconciliation/tribunal-discovery.json · docs/legal-reconciliation.md

DWP-BL-014: Calculator comparison

Priority P2; aggregate status not_started.

Work package Kind Status Executor Next action or retained result
DWP-BL-014.delivery implementation not_started agent Prepare versioned calculator comparisons with controlled synthetic scenarios.
DWP-BL-014.acceptance independent_review waiting human Resolve differences using source evidence and domain review.

Evidence: docs/future-sources.md

DWP-BL-015: CASA framework assessment

Priority P2; aggregate status not_started.

Work package Kind Status Executor Next action or retained result
DWP-BL-015.delivery research not_started agent Assess CASA only against an established service-flow need.
DWP-BL-015.acceptance independent_review waiting human Review accessibility and integration trade-offs for the actual operational service.

Evidence: docs/future-sources.md

DWP-BL-016: ChatGPT Voice and room audio rehearsal

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-016.delivery implementation recorded_complete agent Use the prepared five-minute desk check and explicit browser/recorded fallback. Keep actual Voice tool invocation and room audio acceptance separate.
DWP-BL-016.acceptance live_observation waiting human Rehearse the intended account, microphone, Mac output, PA and projection with the presentation owner.

Evidence: docs/next-stage/seminar-webmcp-and-audio.md · docs/remote-mcp-demo.md · docs/voice-rehearsal.md · docs/monday-handover-2026-09-21.md

DWP-BL-017: Portable discovery-first departmental workflow

Priority P1; aggregate status recorded_complete.

Work package Kind Status Executor Next action or retained result
DWP-BL-017.delivery implementation recorded_complete agent Retain the discovery-first methodology and HMRC brief; no HMRC domain discovery is claimed.

Evidence: docs/methodology.md · docs/templates/hmrc-discovery-brief.md

DWP-BL-018: Fair model and affordability benchmark

Priority P1; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-018.delivery implementation recorded_complete agent Delivered bounded paired subscription-client experiment with identical authored inputs, retained failures, usage observations and per-claim model critique. Specialist assessment and a controlled model/cost benchmark remain separate.
DWP-BL-018.controlled-benchmark research not_started agent Establish actual single-model identity and comparable full invocation conditions, then extend the fixed-input sample. Retain unknown identity and observed provider accounting without guessing affordability.
DWP-BL-018.acceptance independent_review waiting human Review claims independently before cost per accepted result or comparative accuracy is reported.

Evidence: evaluation/model-comparison/README.md · docs/staff-model-trials.md · evaluation/model-comparison/staff-2026-09-20/protocol.json · validation/model-comparison/staff-2026-09-20/results.json · validation/model-comparison/staff-2026-09-20/claim-level-model-critique.json

DWP-BL-019: Broader accessibility and cross-browser review

Priority P1; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-019.delivery implementation recorded_complete agent Delivered additive both-manual Reader and bounded local browser, keyboard, narrow-screen and targeted accessibility checks; continue the separate acceptance work.
DWP-BL-019.acceptance independent_review waiting human Obtain assistive-technology and representative-user observations before claiming broader accessibility conformance.

Evidence: validation/corpus-questions/public-explorer-observation.json · docs/combined-reader.md · validation/combined-reader/README.md · validation/combined-reader/browser/artifacts.json · validation/combined-reader/build.json · validation/combined-reader/public/README.md · validation/combined-reader/public/artifacts.json · docs/household-reader-public-verification.md · validation/household-reader-public/3ef0e786/artifact-manifest.json · docs/household-reader-verification.md

DWP-BL-020: Source refresh and drift process

Priority P1; aggregate status needs_domain_review.

Work package Kind Status Executor Next action or retained result
DWP-BL-020.delivery implementation recorded_complete agent Delivered bounded increment; use the linked evidence and named remaining obligations for the next reviewed expansion.
DWP-BL-020.acceptance independent_review waiting human Approve impact and source-version changes before switching a published evidence binding.

Evidence: docs/adm-acquisition.md · evaluation/source-refresh/observed-comparison.json · evaluation/source-refresh/synthetic-controls.json · docs/source-refresh.md

DWP-BL-021: Canonical contracts and protected publication

Priority P0; aggregate status recorded_complete.

Work package Kind Status Executor Next action or retained result
DWP-BL-021.delivery implementation recorded_complete agent Preserve current schemas, private-input guard and protected reviewed publication.

Evidence: okf.semantic.json · okf.publication.json · docs/repository-governance.md

DWP-BL-022: Retrospective and visible multi-agent change history

Priority P1; aggregate status recorded_complete.

Work package Kind Status Executor Next action or retained result
DWP-BL-022.delivery implementation recorded_complete agent Keep a visible changelog, dated work logs and evidence-backed retrospective in lockstep.

Evidence: CHANGELOG.md · docs/retrospective.md · docs/work-log-2026-09-19.md · docs/work-log-2026-09-20.md

DWP-BL-023: Hosting and Content Security Policy integration

Priority P1; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-023.delivery implementation in_progress agent The supported HTML no-transform change is deployed. Investigate remaining Firefox hosting-cookie behaviour through supported host controls without weakening CSP or hiding errors.
DWP-BL-023.acceptance live_observation in_progress agent Retain passing three-engine staff functional checks and Firefox strict-console failure; require a genuinely clean observation before closing hosting acceptance.

Evidence: validation/compact-delivery/v0.3.1/browser/historical/run-summary.json · validation/compact-delivery/v0.3.1/browser/full-corpus/chrome-receipt.json · validation/compact-delivery/browser/public/run-summary.json · validation/compact-delivery/browser/public/chrome-receipt.json · validation/compact-delivery/browser/public/firefox-receipt.json · validation/compact-delivery/deployment.json · validation/compact-delivery/v0.4.0/deployment.json · validation/compact-delivery/v0.4.0/sdk-receipt.json · validation/compact-delivery/v0.4.0/browser/staff/run-summary.json · validation/compact-delivery/v0.4.0/browser/staff-native/run-summary.json · docs/staff-service-browser-verification.md · validation/compact-delivery/v0.4.0/README.md · validation/compact-delivery/v0.4.0/artifacts.json · validation/compact-delivery/v0.4.0/browser/historical/run-summary.json · validation/compact-delivery/v0.5.0/deployment.json · validation/compact-delivery/v0.5.0/sdk-receipt.json · validation/compact-delivery/v0.5.0/browser/staff/run-summary.json · validation/compact-delivery/v0.5.0/artifacts.json · docs/household-service-verification.md

DWP-BL-024: ADM Reader and cross-manual navigation

Priority P1; aggregate status recorded_complete.

Work package Kind Status Executor Next action or retained result
DWP-BL-024.delivery implementation recorded_complete agent Delivered additive both-manual Reader and bounded local browser, keyboard, narrow-screen and targeted accessibility checks; continue the separate acceptance work.
DWP-BL-024.acceptance live_observation recorded_complete agent Retain exact published identities and scoped three-browser/facet observations; later application or source changes need fresh publication checks.

Evidence: full-dmg/okf-review-context.json · full-dmg/context/navigation/manifest.json · docs/backlog.md · docs/combined-reader.md · validation/combined-reader/README.md · validation/combined-reader/browser/artifacts.json · validation/combined-reader/build.json · validation/combined-reader/public/README.md · validation/combined-reader/public/artifacts.json

DWP-BL-025: Evidence workbench and all-question inspection

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-025.implementation implementation recorded_complete agent Retain the reviewed implementation, all-40 exact reconstruction and protected CI results. Later source, engine or UI changes require a new bounded replay; specialist acceptance remains separate.
DWP-BL-025.public-verification live_observation recorded_complete agent The merged 7eeded763 site passed 1,141 exact public requests and a live browser journey across all 40 cases, representative traces, deep links and local export. The capital U07 and DMG 84861 direct link were observed with 44 records and 68 relationships; context was insufficient and truncated. Preserve the version-bound receipts.
DWP-BL-025.model-inspection implementation recorded_complete agent Bounded implementation delivered in DWP PR 39 and Explorer PR 146: source-bound interaction and blocked calculation-inspection views are available through the reviewed manifest. Preserve exact source bindings and the observed sidebar developer-connection route; specialist acceptance, universal client support and award calculation are not established.
DWP-BL-025.specialist-review independent_review not_started human Use the workbench to review interpretation, source boundaries and open obligations; engineering acceptance must not be relabelled as legal or specialist acceptance.
DWP-BL-025.sidebar-repeatability live_observation recorded_complete agent Retain the 24 September Requirements and Interactions observation and the separate 25 September fresh-session success after an initial failed attempt. The authorised CDP route called native page tools and changed the visible staff-016 Requirements view; only UI/tool trace was retained, not complete payloads or an independent package hash check. Recheck the intended session before a live demonstration. Automatic discovery through a dedicated WebMCP connection, universal client support, Voice and room audio remain separate gates.

Evidence: docs/evidence-workbench.md · docs/evidence-workbench-work-log.md · domain-profile/evidence-connect/profile.json · structured-context/evidence-connect-manifest.json · evaluation/evidence-workbench/manifest.json · scripts/build_evidence_connect.py · scripts/build_evidence_workbench.mjs · domain-profile/workbench-models/inspection.json · evaluation/evidence-workbench/tools-manifest.json · docs/workbench-model-inspection.md · scripts/build_workbench_models.py · scripts/test_workbench_models.py · validation/evidence-workbench/2026-09-24/public-release/site/observation.json · validation/evidence-workbench/2026-09-24/public-release/browser/receipt.json · validation/evidence-workbench/2026-09-24/public-release/files.json · docs/evidence-workbench-handover.md · docs/workbench-sidebar-demo.md · docs/remaining-work.md · validation/client-connection/2026-09-25/sidebar/observation.json · validation/client-connection/2026-09-25/sidebar/README.md

DWP-BL-026: Occurrence-specific DMG and ADM reading help

Priority P0; aggregate status in_progress.

Work package Kind Status Executor Next action or retained result
DWP-BL-026.bounded-implementation implementation recorded_complete agent Retain the deterministic two-paragraph overlay, exact-span checks and all 20 local body marker links; original proposals and their hashes remain unchanged.
DWP-BL-026.public-verification live_observation recorded_complete agent The paired Chapter 60 v1 release was publicly verified on 26 September 2026. Preserve the byte-identical release receipts and fallback tag. This closes only the two-paragraph display check; no new corpus release, legal acceptance or ChatGPT sidebar acceptance is implied.
DWP-BL-026.held-out-coverage research recorded_complete agent Deterministic processing and source controls are retained: 12 initial cases passed; the first held-out gate retained 23 passing cases and one failed expectation; all 24 remain frozen. A separately reviewed fresh 24-case set passed, as did unchanged-expectation regression after the cache repair. No automated content-processing model calls. Review unsupported mixed-case forms, meaning scope and remaining extraction defects separately.
DWP-BL-026.specialist-review independent_review waiting human Review explanations, disputed source associations and legal-reference meanings without promoting machine extraction or model prose to official authority.
DWP-BL-026.corpus-public-verification live_observation recorded_complete agent Exact consumer b633038c and immutable corpus/workbench data passed all 16 public checks on 27 September. All 40 case sidecars were hash-bound; staff-016 reading-help navigation was browser-checked. Preserve the v1 fallback. Two capital composite records remain unmatched. The final release assets separately bind documentation deployment and the paired tag; no legal or specialist acceptance is implied.

Evidence: docs/reading-help-ch60.md · domain-profile/reading-help/ch60.json · domain-profile/reading-help/README.md · reading-help-ch60.json · scripts/build_reading_help.py · scripts/test_reading_help.py · docs/learning-path.md · validation/reading-help/2026-09-26-v1/reading-help-public-browser-receipt.json · validation/reading-help/2026-09-26-v1/reading-help-ci-deployment-receipt.json · validation/reading-help/2026-09-26-v1/provenance.json · docs/reading-help-rollout-work-log.md · evaluation/reading-help-rollout/replays/reading-help-20260927-01/report.json · reading-help-corpus/manifest.json · docs/reading-help-corpus-contract.md · evaluation/reading-help-rollout/initial-gate-02.json · evaluation/reading-help-rollout/heldout-gate-01.json · evaluation/reading-help-rollout/heldout-03-controls.json · evaluation/reading-help-rollout/heldout-03-gate-01.json · evaluation/reading-help-rollout/heldout-03-producer-replay-02.json · evaluation/reading-help-rollout/workbench-link-report.json · evaluation/evidence-workbench/reading-help-manifest.json · docs/reading-help-demo-2026-09-30.md · evaluation/reading-help-rollout/heldout-03-producer-replay-03.json · docs/reading-help-release-2026-09-27.md · evaluation/reading-help-rollout/release-20260927/manifest.json · evaluation/reading-help-rollout/release-20260927/public-attempt-03/receipt.json