docs(mar-886): funding dossier (stage 8) — prepared for review, not submitted

[Thebeatkicks]

MAR-886 stage 8 (no Linear id — workspace free-issue limit; intent lives in docs/foundation/README.md).

Nothing was sent, posted, submitted or delivered. No application filed, no community contacted, no message left this machine except HTTP GET requests to public pages. Every outreach text in the dossier is an unsent draft for Henrik's later use.

Documentation only. No code changed, no test was run, no application was launched.

What this adds

Eleven new files under docs/funding/, plus docs/foundation/stage-8-handoff.md.

File	Contents
docs/funding/README.md	Programme facts with source and check date, status and next step per track, readiness table per track with the gap named on every row. Three columns — official requirement / our assumption / open question — wherever facts appear
walrus-brief.md, starknet-brief.md	Application drafts: one-liner, user problem, why the ecosystem, bounded solution, reusable deliverables, existing base, demo, team facts, risks, next 90 days
milestones-and-budget.md	Work packages in three funding classes so nothing is funded twice; effort ranges; cost assumptions with basis or "unknown"; funding gap as a formula; token-valuation risk with no rate assumed
evidence-index.md	Claim → evidence, with a row per stage 1–7 acceptance item marked planned, and every external source with its fetch result
demo-script.md	Two recordable demos, each ending on a failure case; says plainly that no video exists
ecosystem-and-competition.md	Primary-source review with reuse / integrate / build calls
adoption-and-business.md	Falsifiable user hypotheses, interview guide, pilot with explicit failure criteria, revenue hypotheses, measurement plan, unsent outreach drafts
licensing-and-release.md	Own code, direct dependency licences, proposed public/commercial boundary, the UNLICENSED/private gap, release checklist
security-and-operations.md	Threat model, limitations, incident and revocation path, what needs external review. Tests are not an audit
applicant-checklist.md	What the applicant must provide — no data in Git

Programme facts, fetched 2026-09-07

• Starknet Seed Grants (https://www.starknet.io/grants/seed-grants/, read): "up to $25,000 in STRK", non-dilutive; MVP or PoC required; projects already live on Starknet with users are not eligible; rolling, reviewed in rounds, ~4 weeks; KYC/KYB mandatory; three-month plan with a check-in; active Starknet community involvement or prior hackathon/builder programme participation required — we have none.
• Walrus grants/RFPs (https://www.walrus.xyz/rfp/, read): no amounts disclosed; open RFPs sit behind an Airtable link and are not listed on the page; review begins ~2 weeks after an RFP opens; criteria include ecosystem engagement — we have none.
• Fetch failure recorded, not worked around: docs.wal.app returned HTTP 403 (root and two sub-pages) and docs.walrus.site did not resolve. No blob-id, epoch, pricing or encryption fact is quoted as verified anywhere in the dossier.

The two findings that reshaped the pitches

1. Walrus already ships Walrus Memory — a portable, "independently verifiable" memory layer for AI agents (https://www.walrus.xyz/, read 2026-09-07). The brief is re-scoped to a passive, third-party-verifiable deliverable for a recipient who runs no agent, and states that if the open RFP list wants agent memory, the application should not be filed.
2. starkclaw already writes our Starknet sentence — MIT, in keep-starknet-strange, "Don't give an AI your wallet. Give it a session key with hard limits, and enforce those limits on-chain" (read 2026-09-07). The brief is re-scoped to the join — transaction ↔ receipt ↔ approval ↔ run ↔ span, plus a published enforcement table with a direct-bypass negative test — and says contributing upstream would be a good outcome.

Readiness

Neither track is ready to apply. Walrus: no verified matching RFP, no integration, docs unfetchable. Starknet: no PoC, and the community requirement is unmet. Every readiness row names its gap.

Evidence class

Nothing runtime. Repository reads at 620aa93, dependency licences read read-only from the main checkout's node_modules, and public web pages fetched 2026-09-07 with URLs and failures recorded. No fixture test run by this lane, no harness, no application launch, no network transaction. All stage 1–7 acceptance rows are planned.

Handoff: docs/foundation/stage-8-handoff.md, including the "Needs orchestrator" list.

🤖 Generated with Claude Code