Business plan — the register

Started 6 Nov 2026. The plan as a register: one section per thing a credible plan needs, every statement tagged, every statement sourced. Built section by section in the order in §0. This file is the source of truth; slides and review copies are generated from it, never edited separately. Lives in the private founders repository. Channel partners may be named; prospects are not named until they are clients.

Tags. [agreed] the founders have said yes · [proposed] one founder has put it forward, or it was argued in the pack without a decision · [open] undecided, with the question stated. Evidence. receipted lessons from real work exist · part some receipts · hypothesis designed, not run. The plan must not overclaim: only sourcing (document build, evaluation) and part of elicitation (ARIA/GIS interviewer, BIA tool) are receipted; every strategy-stage asset is hypothesis; WSV is a reference architecture and a laptop prototype, not a running gateway; the harness (builder/QA agents, repo template, leak check) has not yet run a client build. Sources. S# founders' pack slide · WN working note (16–20 Oct) · T planning session 5 Nov 2026 (sessions/sunrise.docx) · SN:name Sunrise note · WSV# WSV deck slide · D:date decision in chat.

0. Order of work and decisions log

Order agreed 6 Nov: offering → customers and segments → sector sequencing → pricing and entry offer → go-to-market → operating model and founders' roles → firm vs company, funding, runway → competition and the Copilot answer → IP and legal → market size → financial model → risks and milestones (running register). [agreed] D:2026-11-06

1. Purpose, mission, vision

• Purpose — To turn the friction in organisations into their advantage. [agreed] S13
• Mission — We turn friction into micro-assets: provable pieces of work that run themselves and get better with use, and we make organisations able to keep building their own. [agreed] S13
• Vision — Every organisation compounds: the way it works is an asset it owns, not a cost it carries. [agreed] S13
• Positioning headline: The friction in your business is the opportunity in your business. Shorthand: We turn friction into assets. [agreed] WN
• These are the frame, not the first client conversation. Too abstract for a CIO; the pitch is in §2.1 and §6. [agreed] T

2. The offering — what we sell

One offering, four layers. Sold top-down (the client buys an engagement), built bottom-up (the factory and the architecture make it possible). [agreed] D:2026-11-06

2.1 The first pitch: taking the friction out of IT management

• First market: IT management in mid-size regulated operators and the companies that bid into regulated sectors. The founders' home turf: strategy, demand, architecture review, PMO, sourcing, service delivery. [agreed] T, WN (20 Oct), SN:knowledge_work_archetypes
• The pitch is concrete: the demand cycle is broken (idea on Monday, approved Tuesday, SOW Wednesday); the ARB becomes exception-only; the strategy stops ageing in a PDF. Not "friction", not "AI". [agreed] T
• The strategy is the entry point and the baseline every later decision reads from (is this aligned, funded, architecturally conformant). Whether it is priced nominally or as an asset is §5's first question. [proposed] T; conflict recorded in Appendix B.1
• What we leave behind is the spine, the eleven registers that every branch reads and writes, and a trained process-asset owner. The spine is what makes the offer extensible to compliance, risk, regulatory reporting and operations later; that extension is roadmap, not pitch. [agreed] T, S8, S11

2.2 What recurs, decided enough to plan on

• The factory licence (method and updates) and the strategy agent (standing assessment on triggers) are the recurring elements. [proposed] S5, SN:tech_strategy §4
• Owning a gateway platform is not part of the offer at launch; WSV is delivered as "buy, then close the gap" (deterministic label-at-source in front of a market gateway; local model and SOCI/CIRMP rule mapping added) where a client needs it. [proposed] WSV12, T. The open "platform or method-only" question (S13) is reframed: which parts of the factory to productise first. [open]

2.3 Framings considered and not chosen (6 Nov)

• Three products sold as a portfolio (WSV, factory, engagements): three sales motions for two to four people; splits the proof story. Not chosen.
• "Provable AI for knowledge work" as the lead: true and kept as the positioning sentence, but abstract to a CIO and the buyer role does not exist yet. Kept as positioning, not pitch.

2.4 How this differs from the labs' deployment arms (11 Nov)

Ode with Anthropic and the OpenAI Deployment Company (§9.5) share our delivery model word for word: embed, build inside the process, leave something behind. They do not sell micro-assets; they sell the thing micro-assets replace: embedded engineers who build bespoke systems around a frontier model and stay to run them. The divergence is each property on S3: [proposed] D:2026-11-11

The line to test: their engineers stay because the system needs them; ours leave because it doesn't. The risk (S6) is that from outside the two look the same until a client asks three questions: who can rebuild it, what proves it, and who owns the loop. Those three questions are the pitch when a CIO says the labs will do it for them (§6.2). [proposed]

3. Customers and segments

3.1 Who the customer is

• The organisation: a mid-size operator in a regulated sector, or a company bidding into one, whose IT governance is done by hand: no EA tooling, a small IT leadership team, a PMO that is a spreadsheet or absent, an ARB that is a meeting. Where being wrong costs something (regulator, concession, contract). [proposed] SN:HANDOVER mid-size ruling 3 Nov, T, S5
• The buyer: the CIO or head of IT, and the people who feel the friction under them — head of architecture, PMO lead, commercial or procurement manager. We talk to them about their strategy, their demand queue, their ARB, their sourcing — never about friction or AI. [agreed] T
• The role we create: the process-asset owner (the same person as the AI-governance owner). The buyer role does not exist before the first engagement; it is installed by it. [agreed] S5, S6
• Size bracket in staff or IT spend: not yet defined. [open]

3.2 Three routes to the customer

The boutique route is the part of thread 8 we keep: partner with firms that hold relationships we do not, rather than enabling any consultancy that asks. The Mobility Futures arrangement needs settling before the first engagement runs under it, and the terms it sets become the template for any second boutique: who owns the client relationship, who holds the IP in the method, the assets and the client's spine, how revenue is shared. [open] → §10

3.3 Not our customer, for now

• Large government agencies: laggards; they buy after the market has. [proposed] T
• Large regulated accounts the labs' deployment companies will staff. [proposed] S6
• Front office and customer-facing work: real customers meet the AI; back office first. [proposed] T
• Generic "AI enablement" of professional-services firms, agriculture, property: parked; different value proposition (speed, not proof) and a leakage risk. [open] T

3.4 Where the spine goes after IT (depth before breadth)

Within an existing client, the next circles on the capability map are compliance, risk and regulatory reporting (the spreadsheet-out, spreadsheet-back cycle), then asset and operations management. These are extensions sold to a client who already owns the spine, not new segments. [proposed] T, WN 20 Oct

4. Sector sequencing

Rule: no third sector until receipts exist in two. [agreed] D:2026-11-06, from S6 ("receipts in two more domains, fast")

5. Pricing and revenue model

5.1 Principles

• Price the asset and the outcome, not the days. The Big 4 price by time; our elapsed time is a fifth of theirs, so day rates would price us at a fifth of the value. [agreed] S10, SN:tech_strategy §5
• Every asset is baselined and measured, so the number we quote is what it did, not a promised percentage. [agreed] S3
• Something must recur. Enablement paid once is a consulting firm; the factory licence and the living strategy are the recurring lines. [agreed] S5
• A fraction of a Big 4 engagement that proves more. Mid-size clients have lower willingness to pay; the comparison point in the room was A$200–300k for a strategy that ends on a shelf. [proposed] SN:HANDOVER, T. The fraction is not set. [open]
• Cost assumptions are metered or local, never subscription-subsidised. [agreed] WSV8, T (Appendix B.5)

5.2 Revenue lines

5.3 The entry offer — Appendix B.1 resolved as a proposal

The problem: nobody has to buy a strategy (T), yet the strategy is the baseline every later asset reads from, and the pack says sell it as provenance, not discount (S10).

Decision: C, for the first three clients, then B. [agreed] D:2026-11-06. A and B stay on the table as alternatives to return to if C proves hard to explain or the credit defers too much cash. The free option is kept only for work that is explicitly our own dogfooding (as the BIA tool was), not for client engagements. Dollar figures for the strategy price, the credit and the licence are set in §12 once §7 and §8 give the cost base. [open]

5.4 Open questions for §12

1. What fraction of a conventional strategy price the entry engagement carries, and how it varies with client size. [open]
2. Licence and living-strategy fees for a mid-size operator; whether they are one line or two. [open]
3. Who carries model and hosting cost: bundled, passed through, or the client's own tenancy. [open] T
4. How many entry engagements convert to licence, and in what time; the assumption behind any first-year number. [open]
5. The channel split with Mobility Futures. [open] → §10

6. Go-to-market

6.1 The motion: sold by demonstration, from inside

• The first sale at each client is made from inside a real piece of work, to a role that does not yet exist. A founder (or the boutique's person) holds a role with authority over the process; the first asset is shown working, not proposed. [agreed] S5
• Relationships decide the first clients; without them it is door-knocking. The three routes in §3.2 are the relationship map. [agreed] T
• We sell a business problem, never AI. The words in the room are the client's: your demand queue, your ARB, your roadmap, your bid. "Friction", "micro-assets" and "the factory" are for the plan, the pack and the partner, not the first meeting. [agreed] T, S6

6.2 The pitch to a CIO (the IT-management version of the story)

1. Your demand cycle is broken: ideas go into a form and nothing happens for six months, so the business buys around you. [proposed] T
2. Everything IT checks before saying yes is mechanical: strategy alignment, budget, architectural conformance. Done against a current strategy and a standards register, it takes seconds; the humans look only at exceptions. [proposed] T, S8
3. That needs a strategy that is current, not a PDF. We draft yours in days from your own documents and one interview, you correct it, and it stays alive, re-assessed when things change. [proposed] S10, SN:tech_strategy. Caveat carried: hypothesis until the first trial.
4. Every piece of it carries its source and its status, so you can show the board, the auditor and the regulator where each figure came from. [agreed] S4 (receipted for sourcing; the claim is made only for what is receipted)
5. You own the result; we keep the method. [agreed] S5
6. When the CIO says "Anthropic or OpenAI will do that for us": ask who can rebuild it without them, what proves it, and who owns the loop; and whether the bill scales with how much their agents run (§2.4, §9.7). [proposed] D:2026-11-11

6.3 The sequence inside a client

Strategy (entry, §5.3) → the demand → ARB → PMO chain → sourcing (receipted) → service delivery model → compliance, risk and reporting (§3.4). Entry can be anywhere the pain is; the spine grows outward. [proposed] S8, SN:knowledge_work_archetypes §4

6.4 The channel motion

With a boutique (Mobility Futures first): the boutique opens the door and holds the domain; we run the method and the factory; the client sees one team. Who fronts the engagement, who owns the client and how it is paid are §10 questions, to be settled before the first engagement runs this way. [proposed] / [open] D:2026-11-06

6.5 What we need in hand to sell

6.6 Who sells, and when

Who carries sales among three founders, and how much time it gets alongside current commitments through Christmas, is §7. The near-term calendar: meet Chen on direction; pitch deck alongside the build; first strategy prospect via Mobility Futures as the first trial. [agreed] as actions, T

7. Delivery and operating model; founders' roles — placeholder

To be worked out together by the three founders, not drafted here. [agreed] D:2026-11-06

What the section will need to settle, carried from the sources: three founders [agreed]; names, backgrounds, time available before and after Christmas, and the relationships each holds; who embeds first (S13) and who carries sales (§6.6); capacity as the ceiling (S6); how a channel engagement is staffed (§6.4); the builder/QA role split and the review gate as the way work is run (S11). All [open].

8. Firm vs company; funding and runway — framed, not decided

The question from S13: an advisory firm that scales with people, or a method company that scales with IP. Funding, runway and hiring follow from it. [open] S13, WN

What is already said that bears on it:

• Six months of unpaid or nominally paid work to earn reference customers and assets, fundraising off them. [proposed] T. Implies a company. Runway for those months is not quantified. [open]
• Free work is for our own dogfooding only; client work is priced and credited (§5.3). [agreed]
• The entry offer defers cash by design (the credit), which lengthens the runway needed. [proposed] §5.3

Decision is the founders'. The plan needs from it: the legal form, who is paid what and when, the runway in months and dollars, and whether a raise is planned at all. [open] → §12

8.1 Exit as an option — sell to a lab's deployment arm or a Big 4 (raised 11 Nov)

A possible target, not a plan: build something one of three buyer classes would acquire. [proposed] D:2026-11-11

What the option changes:

1. §8 tilts to company: an exit needs a company that owns its IP, not a partnership that pays salaries. [proposed]
2. Build what is acquirable: the gate and the SOCI dictionary as a product; the method embodied in the harness, not a document; receipts and named references in two regulated sectors. This is S6's defensibility list; the work is the same either way. [proposed]
3. Vendor neutrality becomes a dated decision: neutral in deliverables always; the stack we build on makes us legible to one buyer class or another. Decide knowingly, later, with the exit in view. [open]
4. Clean IP matters more: the Mobility Futures terms (§10.3) and the receipts clause (§10.2) must leave the core unencumbered. [proposed]
5. Caveat: build to last and be acquirable as a consequence. See also §9.7: the conflict-of-interest positioning cuts against a lab exit; one of the two must lead. Neither lab's arm has Australian operations yet; a plan aimed at a named buyer inside a window that is shorter than the pack assumed is fragile. [proposed]

9. Competition and the written Copilot answer

9.1 The competitor map (from the critique of 16 Oct; positions as at then)

9.2 The Copilot answer (verified against public sources, 6 Nov 2026; US list prices; re-verify before each client conversation)

What the room said on 5 Nov was right in shape and wrong in three details. Corrected:

Three things Copilot does not do, which are the offer: hold a spine of keyed registers with status and source that documents are generated from and verified against; make the joins deterministic checks with a decision register a regulator can read; carry a method with a QA role and a learning loop. Copilot is where the client's people will read and ask; the spine lives in their tenancy or repo; the factory is ours. [proposed] T, D:2026-11-06

Sovereignty, corrected. The transcript assumed third-party models in Copilot run offshore only if chosen. Verified: since January 2026 Anthropic's Claude models are enabled by default in commercial Microsoft 365 Copilot tenants outside the EU and UK; when used, prompts and grounding data leave Microsoft's processing boundary to Anthropic as a sub-processor, on infrastructure primarily in the United States, outside Microsoft's data-residency commitments and audit controls. Admins can disable it in the Microsoft 365 admin centre. Microsoft's own OpenAI models process within the Microsoft boundary; in-country processing of Copilot interactions for Australia was announced for end-2025, and Microsoft's April 2026 update puts local inferencing across Copilot, Copilot Studio and Power Platform in Australia by end-2026. Status of the Australian rollout to be confirmed before it is stated to a client. Residency is not jurisdiction (WSV4, WSV13) either way. [proposed] D:2026-11-06

The answer to a CIO, in three lines. Keep Copilot; it is where your people will read the strategy and ask it questions. What you are buying from us is the thing behind it: a spine of registers your documents are generated from, with every figure's source and status, and a method that proves the work. And a setting to check: whether Claude is on by default in your tenant, because if it is, Copilot prompts are already leaving Microsoft's boundary. [proposed]

9.3 Deployment: where the factory runs at a client

Options raised: the client's own cloud tenancy (Azure/Foundry or an onshore region); a local box at the client (an Apple-silicon or GPU machine, the WSV prototype shape); a box rented from us; our laptop for the trial. Gateway software routes but does not tier; the deterministic gate is ours to add in front (WSV12 "buy, then close the gap"). The choice is a deployment profile per client and a cost line in §5; the plan assumes metered or local cost, never consumer subscriptions. [open] T, WSV12

9.4 "Won't Microsoft and the labs do this for everyone?"

The written answer the pack asked for. They will ship agent governance (Agent 365 exists), audit logs and probably per-agent audit trails; the labs will keep driving build cost to zero. They will not ship: a deterministic tier gate mapped to SOCI and CIRMP definitions, because it depends on the client's labelling discipline and asset register; a spine of registers that the work itself keeps true; a method with a QA role and receipted rules; or the process-asset owner. Those are specific to the organisation, its regulator and its consequences, which is why the labs are structurally uninterested in them (WN). The honest risk stays: the gap narrows from above, and the window is two to three years (S6). [proposed] S6, WN

Sources checked 6 Nov 2026: Microsoft Learn (Copilot Studio billing rates; Anthropic models in Copilot apps; connect to AI models), Microsoft 365 blog (in-country processing, 4 Nov 2025, updated 3 Apr 2026), Directions on Microsoft (Jan and Jun 2026), third-party licensing guides dated Jun–Oct 2026.

9.5 The market as at November 2026 — what has changed since the October critique

The critique of 16 Oct (S6) was right in direction and understated in degree. Verified 11 Nov 2026:

The labs now sell implementation, not models. OpenAI launched the OpenAI Deployment Company on 11 May 2026 with more than US$4B, acquiring Tomoro (150 forward-deployed engineers); its backers include McKinsey, Bain and Capgemini, and it sits on top of the February 2026 "Frontier Alliances" with McKinsey, BCG, Accenture and Capgemini. Anthropic, Blackstone and Hellman & Friedman launched Ode with Anthropic on 15 July 2026 at US$1.5B, built on Fractional AI (100 engineers), Claude-first, and explicitly aimed at mid-size organisations across financial services, healthcare, retail, manufacturing and software, with private-equity portfolios as the pipeline. Ode has since acquired Casper Studios; international expansion is signalled but not detailed. [proposed]

Correction to the pack. S6 says "aim below the labs at mid-sized regulated operators they won't staff." Ode targets mid-size by design. What still holds: their pipeline is PE portfolios, their geography is the US, their posture is Claude-first, and their unit is an engineer, not a method. Our ground is narrower than "mid-size": mid-size regulated operators and bidders in Australia, where proof to a regulator is the purchase, reached through relationships and boutiques rather than a PE sponsor. Appendix B.6. [proposed]

Both labs are in Australia. OpenAI launched locally in December 2025. Anthropic has registered an Australian subsidiary, opened a Sydney office under a general manager for Australia and New Zealand, signed an MoU with the federal government, and is approved to supply non-corporate federal agencies for official information. Neither deployment arm has announced Australian operations yet; expect them within the window. [proposed]

The Big 4 and MBB have repriced around AI and are moving down-market. Accenture reported about US$2.7B of generative and agentic AI revenue in FY2025 and about 110,000 AI and data professionals at the end of FY2026. BCG attributes about a quarter of 2025 revenue to AI. McKinsey says it runs about 25,000 agents alongside about 40,000 people and has cut headcount. Deloitte abolished the analyst-to-manager title ladder from June 2026. PwC Australia launched agentic-powered managed services for private and mid-market clients on 7 Aug 2026, starting with finance and expanding into cyber, procurement and digital trust. The pyramid that priced the twelve-week strategy is being dismantled by the firms that own it; the four-week draft-first engagement is no longer a novelty on its own (S10's caveat, now a fact). [proposed]

Australian AI boutiques are many, consolidating, and sell on the same words. Local lists name Mantel Group, Team400, Datacom, RUBIX, Sandlabs and a dozen others, mostly "agents in production for the mid-market," priced by the hour (indicatively A$200–350 for specialists, A$250–500+ at the Big 4). None found positions on per-asset proof, a deterministic boundary or a method as the product. [proposed] (list sources are vendor-written; treat as a map, not an assessment)

The gateway category has consolidated into security vendors. Portkey was acquired by Palo Alto Networks in May 2026 and is now the Prisma AIRS AI Gateway; Prompt Security went to SentinelOne, Aim Security to Cato, Lakera to Check Point. Kong added an Agent Gateway (April 2026) for LLM, MCP and agent-to-agent traffic. LiteLLM remains the most-adopted open-source self-hosted gateway and had two CVEs in 2026. Skyflow's LLM Privacy Vault is the reference product for reversible tokenisation with policy-governed re-identification, delivered as SaaS with a Kong plugin and PII/PHI/PCI entity classes. These products sell to CISOs, classify with models, and know nothing of SOCI entity classes. [proposed]

Australian sovereign hosting now exists as a product category. Macquarie Cloud Services' Launch AI (API access to models in an Australian sovereign private cloud, managed, predictable pricing); SCX.ai (NSW and SA nodes, regulated-workload positioning); ResetData's AI-F1 (Melbourne, H200); Sovereign Australia AI (NextDC, B200s, building Australian models); Blue Crystal Solutions (private LLM deployment on-prem or private cloud, from about A$500 a month); and Sovereign AI Infrastructure Inc. (Melbourne, air-gap-capable nodes and "edge appliances for critical-infrastructure sites", launched June 2026). The Vault tier's hardware question has answers we do not have to build. [proposed]

What this does to the window. Narrower and shorter than the pack says: the labs are in the country, the Big 4 are in the mid-market with agentic managed services, and the gateway plumbing is owned by security vendors with sales forces. What is still not sold by anyone found: a deterministic, regulator-readable tier decision mapped to SOCI and CIRMP; per-asset evidence for knowledge work; a method that installs a client-side owner and learning loop. Two to three years is now the optimistic end. [proposed]

9.6 WSV: partner, resell or build?

Answer: assemble, don't build; own the gate rules and the proof; partner for hosting; resell nothing. WSV decomposes into five parts, and the answer differs by part. [proposed] D:2026-11-11

Consequences:

• "Buy, then close the gap" (WSV12) is confirmed as the offer, with a sharper statement of the gap: the gate rules, the SOCI entity dictionary, the deployment profile and the evidence pack. The gateway and the hosting are the client's or a partner's.
• Partnerships to pursue, in order: one sovereign host for the "rented box" option (Macquarie Launch AI is the obvious first conversation; SCX for critical-infrastructure positioning); one tokenisation engine (Skyflow for a reference integration, an open-source engine for on-prem); one open-source gateway as the default build target. Integration partner, referral at most; no resale margin, no exclusivity. [open]
• Anthropic's arrival is an opening as well as a threat: an Australian partner programme for regulated sectors would put us where Ode is not yet. Vendor neutrality holds in deliverables; it does not stop a partnership. [open]
• Where this leaves "platform or method-only" (S13): productise the gate and the dictionary as a small, installable component of the factory, not a platform. The platform worth owning was always the factory; the gateway is somebody else's. [proposed]
• Dogfood the Vault. The gate rules, the entity dictionary and the CIRMP mapping are Vault-grade IP by our own definition; build them on a local model under our own gate, not in a frontier tool. It is the first receipt for the tier and the line a CIO will believe: we built our own crown jewels the way we tell you to. [proposed]
• Risk: the security vendors add a "deterministic policy" tier and a SOCI template once an Australian customer asks. Mitigation is speed and receipts: the first CIRMP evidence pack generated from a live gate is the receipt that matters. [proposed]

9.7 The conflict-of-interest argument — insulating clients from the meter (raised 11 Nov)

The labs' deployment arms are paid in engineer-hours and measured, upstream, in model consumption. Ode holds Anthropic equity and is Claude-first by charter; the Deployment Company is majority-owned by OpenAI. Their best outcome is a client whose systems need their engineers and their model. We are the party without that conflict: the asset's spec is the source of truth, the routine bulk runs on a local or sovereign model at fixed cost, the frontier is rented for the residual, and the model is a line-up choice per tier that the client can change. This is WSV13's "own the bulk, rent the residual" turned from a value line into a competitive one. [proposed] D:2026-11-11

Stated carefully, so it survives a sceptical CIO:

• Not "the price will go up": per-token list prices have mostly fallen. The claim is that consumption scales with agent autonomy (WSV8: 97% of one month's tokens were agents re-reading their own context), the flat plans hide a ~28× gap to metered cost, heavy users are already rate-limited, and an embedded model-first team has every incentive to let agents run. Dependence, not price. [proposed] WSV8
• The pitch question: can you rebuild it without them, and run it without their meter? The deployment profile (§9.3, §9.6) is the answer; the gate is the proof of what left. [proposed]
• Keep it vendor-neutral in form: the same argument applies to a Microsoft-first or AWS-first integrator; the labs' arms are only the sharpest case.

Tension to hold. This positioning sits badly beside the exit-to-a-lab option (§8.1) and the Anthropic partner-programme idea (§9.6). A lab might still buy a governed-consumption story for regulated sectors it cannot otherwise reach, but we cannot lead with both. The Big 4 and security-vendor exits are unaffected: the gate, the dictionary and the own-the-bulk profile are what a Big 4 assurance practice would want. Decide which story leads before the first partner conversation with a lab. [open]

Sources checked 11 Nov 2026: OpenAI announcement and coverage of the Deployment Company (May–July 2026); Ode launch coverage (Forbes, Dealroom, PYMNTS, July–Oct 2026); Accenture FY2026 results coverage; PwC Australia media release 7 Aug 2026; IT Brief and Modelwire on Anthropic Australia (Jan–Oct 2026); AI gateway comparisons (Tetrate, Lyzr, WSO2, MarkTechPost, Sept–Oct 2026); Skyflow documentation; CRN Australia and vendor pages for sovereign hosts. Third-party lists and vendor pages are maps, not assessments.

10. IP and legal — the questions, and the principles we start from

Not legal advice; the answers come from the lawyer (the contact via Mike, T). What follows is what the plan already commits to and what it needs decided.

10.1 What we own, what the client owns, what returns

10.2 The receipts clause (new, from §5.3)

What it must say: that we may record lessons from the engagement in anonymised form and use them to improve the method; that the client's name may be used as a reference, with approval of wording; that nothing client-identifying, commercially sensitive or Safe/Vault-tier enters the core; that the credit against later work is the consideration. What it must not do: give us rights over their data, their spine or their documents. [proposed]

10.3 Partnering with Mobility Futures

Open and tricky (D:2026-11-06). The questions, in the order they need answering:

1. What each side contributes. Mobility Futures: transport relationships, domain knowledge, the first prospect, possibly delivery people. Us: the method, the factory, the architecture, the build.
2. Who owns what results. The generic core stays ours regardless of whose engagement produced the lesson; the client's spine is the client's. The gap: transport-specific templates (a bus-operator capability model, a transport regulatory overlay) built on their engagements. Proposed: a domain profile co-owned or licensed back to them for their use, with the generic core untouched. [proposed]
3. Whether they may use the method without us. If yes, under what licence and fee; if no, how that is enforced when the method is a document and a repo. Honest view from the pack: not protectable by IP; defensibility is speed, receipts and the harness (S6). The protection is a licence plus the fact that the harness and the learning loop are hard to run without us.
4. Who fronts the client and holds the contract, and therefore who carries liability and payment.
5. Revenue share for engagements they bring, for licences that recur after they have stepped back, and for the strategy agent.
6. Whether Mobility Futures' material is a tier question: anything they hand us about the prospect is Safe by default.

The terms set here become the template for any second boutique (§3.2). To be settled before the first engagement runs under it. [open]

10.4 Other legal matters carried

• Licensing of outputs versus the way they are generated: clients get the outputs and the spine, not the factory's internals. The lawyer has done this shape before for an agriculture SaaS client. [proposed] T
• Recording and capture consent in workshops and interviews; statements about individuals excluded; attribution internal-only unless agreed. [agreed] SN:tech_strategy §5
• Privacy: the automated-decision-making transparency requirement due December 2026 applies to clients who run our assets on personal information; our registers are the evidence. [proposed] WSV6 (indicative, not legal advice)
• Legal form of the venture: §8.

11. Market size — later (needed for a raise, not for the first clients)

12. Financial model — later (a runway sheet first; the full model once §8 is decided)

13. Risks and milestones — running

Risks carried from S6: "micro-assets" one paragraph from "an agent for every workflow"; capacity-bound delivery; breadth unproven; latent demand and a buyer role that does not exist; the labs may ship audit trails. Added 5 Nov: cost assumptions must be metered or local, never subscription-subsidised (WSV8); trade-secret leakage if we enable other consultancies. Added 6 Nov: a triggered strategy agent built inside Copilot Studio is metered, so the recurring licence must carry its own run cost or run outside Studio; Claude is on by default in Australian Copilot tenants, which is a finding for the client and a tier question for us. Added 11 Nov: Ode targets mid-size and both labs are in Australia; PwC Australia sells agentic managed services to the mid-market; the gateway category is owned by security vendors who could add a deterministic tier on request. The window is shorter than the pack says. Milestones: the first strategy prospect's day-three draft; Chen meeting; pitch deck; first upstream receipt. Added 11 Nov: the first CIRMP evidence pack generated from a live gate (the receipt that answers the security vendors, §9.6); our own gate rules and SOCI dictionary built under our own gate on a local model, as the first receipt for the Vault tier and the proof we practise what we sell. [proposed]

Appendix A — Threads from the 5 Nov 2026 planning session

Appendix B — Contradictions to resolve

1. Entry-offer pricing. T: strategy free, ~$10k or a credit. S10 and SN:tech_strategy §5: asset-priced, sold as provenance, not discount. Resolved 6 Nov (§5.3): priced, credited against the first build or licence, balance paid in receipts (option C); A and B kept as alternatives.
2. Government in the beachhead. S5/S13 name government; T says agencies are laggards. Proposed resolution: regulated operators and bidders, not agencies (§3).
3. Founder count. Pack: four. Resolved 6 Nov: three (§7). Pack and one-pager to be corrected when next regenerated.
4. ARB/PMO: branch or front door. S8/S11 treat EA/ARB and PMO as branches off strategy; T makes the demand chain the lead. Proposed resolution: the demand chain is the pitch, the strategy is the baseline it reads from; both stand (§2.1).
5. Cost assumptions. T mentions businesses stacking consumer subscriptions; WSV8 says never build on subsidised pricing. The plan assumes metered or local cost.
6. "Aim below the labs" (S6). Ode with Anthropic targets mid-size organisations by design. Resolved 11 Nov (§9.5): our ground is mid-size regulated operators and bidders in Australia, reached through relationships and boutiques, where proof to a regulator is the purchase. Pack slide 6 to be corrected.

Appendix C — Receipts ledger (what is actually proven)