Research & Knowledge / 2027–2030 Programme
Twenty-workstream implementation and evidence register
This is a durable status and implementation record consolidating work already undertaken, current deliverables and tasks that require independent testing, customer participation or additional technical infrastructure. It avoids presenting the author's established investigations as not begun, while not presenting draft designs as verified deployments.
Evidence standard: committed static pages, completed diagrams, prototype behaviour and approved external validations are different accomplishments. Each record below identifies what is actually documented and what remains to be proved.
| # | Workstream | Current evidence-based state | Link | Documented deliverable | Verification / next gate |
|---|---|---|---|---|---|
| 1 | Complete comprehensive AI Knowledge Atlas | Initial 7-domain/35-topic taxonomy created; exhaustive scientific coverage remains under review. | Existing evidence or workspace | Breadth-first subject map, cited depth links and revision controls | Expand missing disciplines using independent taxonomy audits |
| 2 | Technology Opportunity Window Index | Six-factor model and 28 venture candidates recorded. | Existing evidence or workspace | Opportunity dimensions, explicit score semantics, backtesting requirements | Replace judgemental scores with calibrated outcome evidence |
| 3 | Validate 28 venture opportunities | All ideas documented with assumptions and validation tasks; no paid market evidence verified. | Existing evidence or workspace | Interview and pilot criteria, source checks and comparison framework | Secure permissioned buyer interviews; no invented demand |
| 4 | Build comparative systems registry | 20-sector atlas and record schema created; source-level cases require verification. | Existing evidence or workspace | Maps, tools, edges, evidence status, comparison protocol | Collect real architecture cases and cite primary sources |
| 5 | Prototype research and investigation agents | Deterministic local demonstration and 12-stage architecture available; autonomous tools not deployed. | Existing evidence or workspace | Runnable browser comparison, JSON export, agent process and security gates | Develop read-only retrieval connector, independent verification and labelled test set |
| 6 | AI stack consolidation advisory pilot | Service blueprint documented; no external client pilot carried out. | Existing evidence or workspace | Scope, permitted evidence, overlap matrix, safeguards, deliverables | Validate willingness to pay with real consenting organisation |
| 7 | Complete reusable evidence-verification framework | Reference architecture and claim-evidence rules defined; production API not deployed. | Existing evidence or workspace | Evidence provenance, contradiction and uncertainty fields | Test extraction and false citation detection |
| 8 | Create 2027–2030 commercial roadmap | Prioritised venture register and entry route documented; milestone dates provisional. | Existing evidence or workspace | Stages from research to qualified pilot to offering | Cost and staffing gates after buyer validation |
| 9 | Audit website omissions and duplication | Existing research, venture and government materials cross-referenced; full site audit unverified. | Existing evidence or workspace | No-overwrite audit specification and linkage rules | Run site-wide crawl, compare pages and inspect broken links |
| 10 | Build technical demonstration portal | Working local architecture-comparison demo published at lab page. | Existing evidence or workspace | Interactive tool, synthetic scenarios, JSON export and staged expansion plan | Add backend only after auth, source, safety and privacy controls |
| 11 | Develop legacy-to-AI integration assessments | Opportunity mapped; implementation remains prospective. | Existing evidence or workspace | Readiness scorecard, trust boundaries and API mapping | Permissioned read-only system sample |
| 12 | Create compliance and ethics assessment engine | Control and evidence taxonomy created; no legal verdict engine certified. | Existing evidence or workspace | Role/jurisdiction/obligation and separate ethical-impact records | Test against legal specialists and jurisdiction versions |
| 13 | Interactive visual architecture viewer | Relationship list and comparison demo available; full graph editor not yet built. | Existing evidence or workspace | Component/edge schema and baseline-vs-comparison logic | Implement accessible graph views and layout alternatives |
| 14 | Audit Agency.ai service catalogue | Existing portal identified; full 50-service mapping not established. | Existing evidence or workspace | Crosswalk schema: service, venture, dependency, differentiation | Extract catalogue with source IDs and de-duplicate |
| 15 | Research-gap and novelty assessment protocol | Gap-first, falsification-led criteria documented; no exhaustive overlap claim. | Existing evidence or workspace | Claim, strongest prior art, unresolved limitation, measurable contribution | Systematic literature retrieval and independent validation |
| 16 | Research and technical-report templates | Core document and provenance structure described; publication-specific templates need QA. | Existing evidence or workspace | Sections for abstract, prior art, method, evidence and limitations | Adapt to actual journal/conference submission instructions |
| 17 | Technology and market monitoring dashboard | Monitoring dimensions and source anchors defined; automated feeds not enabled. | Existing evidence or workspace | Signals and quarterly reassessment framework | Set authorised recurring monitoring with sources and clear thresholds |
| 18 | Establish AI QA benchmarks | Illustrative comparison issues are visible; benchmark dataset not validated. | Existing evidence or workspace | Accuracy, source recall, false positive, robustness and cost metrics | Develop human-labelled control cases |
| 19 | Historical opportunity-window case studies | 2017 robotics case and substitution hypothesis linked; broader evidence needed. | Existing evidence or workspace | Case narrative, chronology, contemporaneous sources and alternative explanations | Test market decline claim against actual sales and education data |
| 20 | Automate research and productivity reporting | Dated source-backed entries exist; autonomous reporting is not authorised or operating. | Existing evidence or workspace | Status labels, commit evidence, audit trail and review before publishing | Build signed-off report generation and deduplicated log updates |
Detailed implementation of workstream 5 — Research agents
Start with reproducible, read-only analysis. Distinguish diagram-reading, retrieving independent sources, making claims and approving conclusions as separate authorisation steps. A working comparison demonstrator is linked below; full autonomous data acquisition is not yet deployed.
| Sequence | Engineering action | Method | Exit criterion |
|---|---|---|---|
| 1 | Discovery contract | Define precise question, jurisdiction, classes of systems, scope and prohibited actions. | Signed-off scope; approved sources; reject inaccessible or restricted data |
| 2 | Read-only research intake | Obtain diagrams, specs and public source pages with original URLs and timestamps. | Record source licenses, versions, retrieval method and hashes |
| 3 | Visual and text extraction | Convert each source into entity, relationship and claim candidates. | Use model outputs as provisional; assign evidence spans and uncertainty |
| 4 | Independent validation | Check extracted relations against primary documents, references and contradictions. | Mark unverified or unsupported; avoid inferred implementations |
| 5 | Normalise and compare | Unify names with controlled vocabulary, compare nodes and edges and alternative views. | Metric definition, structural differences, comparator and no false equivalence |
| 6 | Tool and compliance review | Map vendor dependencies, safety controls and applicable duties by region/use case. | Risk questions, role and jurisdiction; not a legal certificate |
| 7 | Benchmarks and adversarial tests | Measure citation precision/recall, extraction error, false positive and missing data. | Human-labelled test cases and explicit pass/fail thresholds |
| 8 | Human sign-off and release | Review findings and permission before any public output. | Corrections, redactions, versioned review record and reversible release |
Detailed implementation of workstream 10 — Technical demonstration portal
Stage public interaction without claiming operational agent capabilities before they exist. The current browser demonstrator processes only contained sample data and exports its comparison report.
| Sequence | Engineering action | Method | Exit criterion |
|---|---|---|---|
| 1 | Define product boundaries | Offer read-only public demonstrator with synthetic fixtures; separate from operational research agents. | No login needed; no personal data, credentials or hidden vendor claims |
| 2 | Design stable object model | Systems, nodes, edges, sources, controls, review flags and versioned findings. | JSON schema and documented assumptions |
| 3 | Build accessible interface | Keyboard-controlled selections, labelled table, readable output and meaningful focus. | Responsive design, mobile review and contrast test |
| 4 | Implement deterministic comparison | Component set comparison, directed edge comparison and evidence gap prompts. | Demonstrable calculation, no made-up compliance scores |
| 5 | Export reproducible outputs | Use JSON report download with scenario IDs and method label. | Portable structured output and consistent reruns |
| 6 | Run functional tests | Check same-system comparison, zero overlap, edge differences and missing controls. | Human-inspectable results and regression cases |
| 7 | Protect against untrusted content | Render remote text as text; no direct execution or unreviewed HTML. | Cross-site scripting and prompt-injection tests before remote intake |
| 8 | Add real document import after review | Only permitted public or uploaded sources with provenance metadata. | File type/size limits and retention policy |
| 9 | Integrate constrained agent services | Server-side secrets, scoped tools, allowlists, rate limits and approvals. | No credentials in browser; no autonomous external writes |
| 10 | Evaluate before wider release | Document performance, errors, costs and usability; maintain rollbacks. | Independent acceptance criteria and incident procedure |
Open the interactive research-agent prototype · AI Knowledge Atlas · Comparative Systems Atlas · 2027 Venture Register
Evidence and policy anchors
UK AI Adoption Survey 2026 · NIST AI RMF · OWASP Agentic Applications · W3C Provenance
This register makes no claim to having interviewed customers, obtained legal sign-off, scanned every site file, trained models or deployed production agents. Versioned evidence and field tests are necessary before those claims can be made.