August / September 2026 Progress Report

Reporting Period: 1 August – 30 September 2026 (combined) | Status: Draft for co-director review before Callaghan submission

Changes are saved automatically to your browser. Use "Copy All" to export, or the Copy button on any field.

Note: Internal product and repository names are deliberately not used in this report. Work is described by what was built. Repository changes establish implementation activity; they do not establish eligibility, time spent, clinical accuracy, uptime or compliance.

Summary of Work Done

Submission summary statement (paste)

Evidence basis: clinical record system (100+ commits, Aug); conversational AI platform (6 Aug / 38 Sep); incident analysis tool (100+ Sep); redaction service (8 Sep).

Objective 1: Secure Data Management for Healthcare Compliance

Period: 20/03/2025 – 30/06/2025In progress

Work done in this period (August – September 2026)

August: Built the clinical record system on a multi-tenant foundation with organisation-level tenant isolation, encryption of patient-identifiable data and a complete audit trail. Access is governed by dated assignments rather than static roles, recording separately who may open a file and who has. Clinical records were implemented as immutable versions with authorship, so corrections and withdrawals are recorded rather than overwritten, including a defined method for marking that a record never happened. Restricted files are excluded from payloads and clinical payloads are withheld from unauthorised contexts. Security, tenant isolation and patient-data test suites were established, a New Zealand compliance workspace was added, and deployment access was restricted to a private network with trusted proxy and host allow-list.

September: Strengthened permission-scoped AI reads, access auditing that does not copy conversation content, context-bound offline replay, encrypted-index recovery, legal-hold-aware audit retention and dedicated organisation boundaries. Incident analysis was restricted to approved module data and write targets, with private shift results discarded and uncertain deliveries guarded. The redaction service was improved to transform sensitive data without corrupting structured evidence, to restore tokenised response keys with collision checks, to disable content logging by default, and to clean persistent and in-memory session tokens together. Short-lived administrator handoff and bounded analysis sessions added further access and retention controls.

Issues

Actions

Evidence basis: tenant isolation, data encryption and audit trail commits (21–28 Aug); September register refs E1, P1, P2, P3, H2, H3, A1–A4.

Objective 2: AI Model Optimization for Medical Conversations

Period: 31/07/2025 – 30/09/2025In progress

Work done in this period (August – September 2026)

August: Direct model work was limited in this month. Effort went to establishing the structured clinical record that AI features consume - typed clinical facts, coded diagnoses, observation trends and a canonical clinical note model - which is a precondition for reliable medical conversation handling. The mobile realtime voice workstream was merged into the main platform.

September: Research addressed how clinical AI can preserve source meaning while using patient context and privacy controls. Permission-scoped record lookup was added and patient-context continuity corrected across navigation and retries. Incident analysis was revised to consider narrative evidence, structured restraint evidence and linked participant records together, with explicit handling of contradictions and uncertain roles. The redaction service was corrected to fix source-position handling, preserve structured document formats and ordinary event dates, and use field labels to recognise sensitive identifiers; related restoration changes recover structured keys and reject collisions. Tentative speaker learning and shared tool-loop behaviour were added to the conversational platform. These changes provide implementation and regression evidence for context handling and privacy-preserving inference.

Issues

Actions

Evidence basis: mobile realtime voice merge (3 Aug); clinical record model commits (24–28 Aug); September register refs E1, E2, P1, H1, A1, A2, A4.

Objective 3: Scalable System Architecture

Period: 30/09/2025 – 30/07/2026In progress

Work done in this period (August – September 2026)

August: The clinical record system was scaffolded on a current web application framework over a managed relational database and in-memory cache, with a dedicated queue worker for background processing and a continuous integration pipeline provisioned with a database and buildable dependencies. Realtime ward displays were delivered over websockets. The location model was replaced with a hierarchical location tree and time-bounded bed allocations, and tree traversal was moved in-memory to remove lazy-loading overhead. Record rendering was changed so a record draws only the section requested.

September: Work investigated execution continuity across streamed AI responses, offline care tasks and external record delivery. Provider execution was consolidated into a bounded shared tool loop with strengthened handling of uncertain writes. Persistent offline task queues, idempotent replay and user and facility context checks were added, and reply streaming and stalled connections corrected. Queued writes were bound to approved targets, uncertain deliveries held for review, and saved-output checks added before retrying timeouts. Bounded request lifetime and token cleanup were added for synchronous analysis. Regression cases document retry, restart, replay and timeout behaviour.

Issues

Actions

Evidence basis: platform scaffold, realtime websockets, queue worker, CI and location-tree performance commits (21–30 Aug); September register refs E1, P1, P2, H2, A3.

Objective 4: User Experience Testing and Optimization

Period: 15/02/2026 – 31/12/2026On-track

Work done in this period (August – September 2026)

August: A design system was adopted across the clinical record system with shared tokens, a single table shape and consistent surface rules. Navigation was reduced from nineteen sidebar entries to three groups. The care round was rebuilt as a phone-sized task screen, a station board replaced the census view, and a ward wall display was added for shared visibility with live updates. Forms were revised so they do not ask for information they will not use, and irreversible actions now require confirmation. A browser smoke suite was established to catch failures that unit tests cannot see, and several structured review passes were run, each recording and resolving defects a nurse would meet in normal use.

September: Contextual clinical chat was refined so the selected patient survives navigation and retries, and phone-based care rounds were developed that queue work offline and sync on reconnect. Isolated prompt trials, visible failure and retry states, and spreadsheet test and download controls with workbook validation were added to the incident analysis tool. Speaker-aware participation and clarification of tentative identity were improved in the conversational platform. Browser and feature regression cases capture interaction failures and expected recovery.

Issues

Actions

Evidence basis: design system, station board, phone round, wall display and review-pass commits (22–31 Aug); September register refs E2, P1, P2, H3.

Capability Development (CapDev) Status

CapDev Update (August – September 2026)

Change Request RQ-IP has been approved. Category 2, Regulatory and Compliance - recorded in the portal as the "Healthcare Compliance" objective - is discontinued. No expenditure was incurred against it. It is replaced by Category 1, Intellectual Property and Patent Capability Development, running 1 October 2026 to 31 March 2027 and delivered by James and Wells at $30,000. The programme is led by a registered patent attorney, as required for patent-related capability work.

Category 5, R&D Information Management, continues at a revised budget of $21,000. During August the research reporting system was extended with a submission pack generator, gap and overlap analysis, support for combined reporting periods, and a ledger of reports as submitted. In September the evidence record was reorganised to link work across all four development workstreams to the approved objectives using qualified commit references and named regression sources, distinguishing shared component work from product work so that evidence is not attributed twice. The approved window for this objective closed on 30 September 2026 with the remaining integration scoped but not complete; an extension of the end date is requested.

Total capability development spend remains $51,000, unchanged from the original commitment and above the minimum requirement.

Learn about RDTI: the approved window closed on 6 August 2026. No training, adviser session or programme guidance activity was evidenced in the period.

Claim 13 — Cloud Service Costs

Where this goes: Claim #13 tab → Materials & Consumables section → rows from B47. One row per invoice. All amounts NZ$ GST exclusive.
Every individual cost is under $2,500, so no invoice numbers, provider names or invoice copies are strictly required — the invoice column is shown here for your own reference.

Item (column B dropdown)	Purchased from (C)	Amount NZ$ (D)	Invoice ref
Cloud Hosting and Web Services Inlduing LLM	Amazon Web Services New Zealand Limited	1669.07	NZIN26-233625
Cloud Hosting and Web Services Inlduing LLM	Amazon Web Services New Zealand Limited	1632.88	NZIN26-291353
Cloud Hosting and Web Services Inlduing LLM	ForeSoft Corporation	1017.37	81724835133
Cloud Hosting and Web Services Inlduing LLM	ForeSoft Corporation	995.39	81776175736
Cloud Hosting and Web Services Inlduing LLM	Anthropic, PBC	173.91	P2IGB0RE-0008
Cloud Hosting and Web Services Inlduing LLM	Anthropic, PBC	173.91	JKBYYXHC-0008
Cloud Hosting and Web Services Inlduing LLM	Anthropic, PBC	173.91	JKBYYXHC-0009
Total		5,836.44

Paste into B47 (tab-separated — fills columns B, C and D)

Budget line $62,000 · claimed to date $35,601.05 · this claim $5,836.44 · cumulative $41,437.49 · remaining $20,562.51.

Two things before you enter it:

Duplicate file. AUG ANTHROPIC 2.pdf and Invoice-JKBYYXHC-0008.pdf are the same invoice. Counted once — delete one copy so it cannot be claimed twice later.
One invoice left out. 6LIB7JUE-0003 (30 Aug, USD 20.00, approx NZ$34.19) is addressed to "Ben's Individual Org" at a residential address. The template requires that "invoices must be addressed to" the contracted business, so it is excluded. Include it only if it can be re-issued to Bridge Point Limited.

The two ForeSoft receipts are in USD, converted at the rates printed on the matching AWS invoices. Substitute your actual bank or Xero rate if it differs.

Review Checklist

Confirm before submission:

Summary statement reflects the combined August–September periodObjective 1 content accurately reflects security/compliance work undertakenObjective 2 content accurately reflects AI/model work undertakenObjective 3 content accurately reflects architecture work undertakenObjective 4 content accurately reflects UX/testing work undertakenCapDev update is accurate, including the approved RQ-IP substitutionClaim 13 cloud costs entered in Materials & Consumables X7; duplicate PDF removedObjective window issues raised with Callaghan (R&D Info Management extension, expired R&D windows, RDTI)

Bridge Point

August / September 2026 Progress Report

Summary of Work Done

Objective 1: Secure Data Management for Healthcare Compliance

Objective 2: AI Model Optimization for Medical Conversations

Objective 3: Scalable System Architecture

Objective 4: User Experience Testing and Optimization

Capability Development (CapDev) Status

Claim 13 — Cloud Service Costs

Review Checklist