Development Fund Proposal
Author: Hashclaw.AI
Created: 2026-04-11
SIG alignment: Application developers — App Building and Developer Experience (documentation, examples, training), per Development Fund Proposal Review Process.
Label: daml-tooling
Champion: Nguyan Vinh — @v9n
Abstract
CC Privacy Club (HashClawAI/canton-edu) is an independent, open-source hub for Canton Network learning material (EN/ZH today), with explicit non-official positioning and links to official documentation.
This proposal requests Canton Coin (CC) to implement Agentic Education for Canton: software agents (e.g. Cursor Skills, scripted bots, or CI companions) that assist maintainers—proposing aggregation refreshes, drafting bilingual units (news / digest / chronicle, research reader guides, video cards, glossary & explainers—§1.4), preflighting i18n key coverage, and summarizing public on-chain/indexer series—while every publish path stays human-reviewed, PR-based, and source-anchored. Concretely, (1) agent-orchestrated aggregation (allow-listed sources, dedup, daily rebuild/deploy), (2) agent-assisted education generation (draft → checklist → maintainer merge, primary-source URLs), and (3) agent-informed on-chain / explorer analysis (reproducible snapshots, explicit methodology, not official statistics). Milestones also cover ja/ko i18n, production launch + IA, agent playbooks + evaluation logs, and 12 months maintenance with quarterly updates. Developer-community event series are out of scope (§1.3). Deliverables are objectively verifiable (manifests §1.4, docs/agentic-education/). Funding request: 200,000 CC (§6.2).
1. Objective and scope
1.1 Problem and ecosystem value
| Dimension | Problem | Ecosystem value |
|---|
| Signal vs noise | Official and community Canton signals are scattered (CIPs, forums, blogs, APIs); learners lack a single neutral index that still points upstream. | Aggregation with transparent sources reduces duplicate hunting and improves CIP / doc literacy. |
| Education throughput | Turning raw updates into structured explainers (EN/ZH, later JA/KO) is labor-intensive; quality drops without checklists. | A repeatable generation + review pipeline produces consistent educational snippets without posing as authoritative specs. |
| Trust | Unofficial summaries can read “official.” | Disclaimers, verify-at-source links, and Development Fund credit. |
| Sustainability | Aggregators rot when feeds break. | 12-month maintenance SLA, daily automated rebuild where configured, monitoring, and documented runbooks. |
| On-chain / indexer literacy | Learners see headlines but not how to relate claims to public explorer or API data. | Reproducible on-chain or indexer-backed figures and methodology improve critical reading—always labelled as third-party index data, not Canton Foundation truth. |
| Agentic maintainability | Ad-hoc LLM chats lack traceability; teams burn time re-explaining the same merge policy. | Documented agent playbooks (prompts, tools, refusal rules) turn recurring education updates into repeatable human–agent work items without autonomous publishing. |
1.2 In scope
- M1 — Site launch & IA for aggregation: Production readiness (§4): disclaimers, performance/accessibility smoke, deploy reliability, information architecture notes for which surfaces are aggregated vs hand-authored, launch announcement + baseline metrics.
- M2 — Multilingual: i18n helper (docs + scripts) with distinct
ja and ko checks (e.g. i18n:check:ja / i18n:check:ko), docs/i18n.md + docs/i18n/ja.md + docs/i18n/ko.md, two parity reports; ship ja + ko UI on home + nav + Learn entry (minimum) or documented gaps-ja.md / gaps-ko.md.
- M3 — Agentic aggregation, daily refresh, assisted generation & on-chain analysis: Agentic education layer: shipped under
docs/agentic-education/ — README.md (scope), playbooks.md (maintainer-facing agent workflows: aggregation PR prep, EN/ZH draft, i18n gap triage, on-chain digest), guardrails.md (must-cite-official, must-not-claim-endorsement, stop conditions), and eval-log.md with ≥10 recorded trial scenarios (input → agent output → human verdict). Aggregation + daily refresh: docs/aggregation/sources.md, docs/aggregation/runbook.md, docs/operations/daily-update.md tying ≥1×/calendar-day GitHub Actions full build + deploy to refresh allow-listed feeds and on-chain snapshot jobs (~20:00 CST cadence per repo). ≥8 aggregation integration PRs in docs/grants/m3-aggregation-prs.md. Generation: docs/content-generation/workflow.md (human-in-the-loop agentic pipeline: agent proposes diff → maintainer edits → merge), checklist, docs/grants/m3-generation-units.md with ≥18 units (§1.4); ≥6 rows flagged agentic: true with playbook_id referencing playbooks.md; ≥4 rows onchain:true with docs/onchain/methodology.md + docs/onchain/sources.md. No narrative ship without maintainer approval per §1.4.
- M4 — Maintenance: 12 months from M1: ≥10 maintenance PRs (or batched equivalents), ≥4 quarterly public update posts,
docs/grants/maintenance-log.md including weekly rollup of daily build success/failure (link to Actions), and docs/grants/final-report.md covering agentic workflow adoption (time saved, failure modes), aggregation, daily refresh, generation quality, and on-chain analysis outcomes.
1.3 Out of scope (explicit)
- Not protocol, node, or synchronizer engineering. Not custody, trading, or token sales.
- Not implying Canton Foundation / Digital Asset / Canton Network endorsement.
- Not in this proposal: Developer-community event programs (meetups, salons, conferences, travel budgets, RSVP campaigns, or Canton Foundation–coordinated physical/hybrid event series).
- Not fully automated “write production DAML” systems; generation stays educational, source-linked, and human-gated.
- Not unsupervised agent merges to
main: agents propose; humans approve (see docs/agentic-education/guardrails.md).
- Not operating a validator, not using non-public chain or participant data; on-chain work uses public explorer/API endpoints only, per
docs/onchain/sources.md.
1.4 Definitions & counting rules
Education content unit — counts toward ≥18 only if the row is in docs/grants/m3-generation-units.md after human merge. One row = one shipped learning artifact (site or agreed docs/education/ surface), any of the shapes below; same mechanical rules for all.
| Shape | One unit |
|---|
| News / digest / chronicle | Dated item or digest block: what changed, why it matters to builders, ≥1 primary URL (official post, CIP, repo release). |
| Research / long reads | Short reader guide (not a full reprint): thesis in plain language, link to PDF / repo / canonical page. |
| Video | Curated card: title, ≤2 sentence context, canonical watch URL (official or allow-listed channel); transcript pull-quote optional if permitted by source. |
| Glossary / Learn blurb / explainer | As today: coherent paragraph(s) with verify-at-source links. |
Mechanical thresholds (every unit row): EN ≥90 characters (excluding URLs) or substantive ZH equivalent in the same row; ≥1 canonical primary-source URL per shipped language (ZH deferral allowed if noted in the row). ≥6 / 18 rows: agentic: true + playbook_id (anchor in playbooks.md) + merge SHA (agents never self-merge). ≥4 / 18 rows: onchain:true — includes a chart, table, or numeric call-out from docs/onchain/sources.md, interpreted per docs/onchain/methodology.md (lag, reorgs, non-official). onchain:true may overlap agentic:true.
Aggregation integration PR — counts toward ≥8 only in docs/grants/m3-aggregation-prs.md: merged PR whose primary diff stays inside allow-listed paths in docs/aggregation/sources.md (e.g. scripts/**, .github/workflows/**, docs/aggregation/**, agreed public/generated/** or src/data/**). ≤2 PRs may add trivial translations.ts labels only, flagged aggregation+labels; they count for 8, not for 18, unless a separate generation-unit row exists.
GitHub routine maintenance — does not count toward 8 or 18; it satisfies M4 when logged in docs/grants/maintenance-log.md. Examples: dependency / security updates within policy, Actions or Pages pipeline fixes, issue / label triage, README or runbook nips, branch-protection or token hygiene (GitHub-only secrets). Net-new learner copy belongs in generation units or aggregation manifests per rules above, not “maintenance” by label alone.
Double-count: one merge may not bump 8 and 18 unless the aggregation+labels exception applies and a distinct generation-unit row is added.
Reviewer sign-off: MAINTAINERS.md — ≥2 merge-capable reviewers for generation PRs; Champion @v9n — ≥4 spot-checks on merged generation PRs (including agent-opened or agent-assisted drafts); notes in milestone thread or docs/grants/champion-spotchecks.md.
2. Technical approach
| Workstream | Approach | Verifiable outputs |
|---|
| Site | Astro static site, src/i18n/translations.ts, GitHub Actions → Pages; open LICENSE. | Tag + CI green + live URL in milestone note. |
| Launch (M1) | Checklist + link sweep + IA doc docs/grants/m1-aggregation-ia.md listing which pages are aggregated feeds vs static narrative; stub docs/operations/daily-update.md pointing to Actions schedule; stub docs/agentic-education/README.md (1-page vision). | Artifacts under docs/grants/m1-launch/ + IA + stubs on main. |
| i18n (M2) | Per-locale ja / ko checks; reports i18n-report-ja.*, i18n-report-ko.*. | Helper + routes + reports merged. |
| Aggregation + daily refresh (M3a) | Allow-listed HTTP/RSS/mail-public sources; rate limits; daily workflow refresh in CI; normalize to typed entries; fail-soft with uploaded CI logs on failure. | docs/aggregation/runbook.md + docs/operations/daily-update.md + ≥8 PRs listed in m3-aggregation-prs.md. |
| Agentic education (M3d) | Maintainer agents (Cursor Skills under .cursor/skills/ and/or documented CLI flows) run bounded tasks: open PR with prefilled template, never push to main alone. | docs/agentic-education/ complete set + eval-log.md (≥10 scenarios); .cursor/skills/ updates as needed. |
| Education generation (M3b) | Agent proposes diff from playbook → human edits → PR checklist → merge; manifest m3-generation-units.md. | docs/content-generation/workflow.md + ≥18 manifest rows (≥6 agentic: true). |
| On-chain analysis (M3c) | Agent may draft chart captions; human validates methodology; read-only public APIs. | docs/onchain/methodology.md + ≥4 onchain:true rows + frozen artifact paths. |
| Maintenance (M4) | MAINTAINERS.md, docs/sustainability.md, docs/grants/maintenance-log.md with weekly digest of daily build health. | Log + quarterly links + final report. |
i18n helper — ja vs ko
- Japanese (
ja): Missing-key / orphan-key diffs; line-length hints as needed.
- Korean (
ko): Separate entrypoint and report from ja; formal-register UI baseline notes.
- Shared: Fail-closed or documented manual gate under agreed coverage thresholds.
Backward compatibility: No Canton protocol, node, or SDK changes.
3. Architectural alignment
4. Milestones and deliverables
Milestone 1 — Site launch, IA & production readiness (Week 8)
- Deliverables:
docs/grants/m1-launch/checklist.md; Lighthouse or equivalent export; m1-links.md; docs/grants/m1-aggregation-ia.md; docs/operations/daily-update.md (stub, daily rebuild Actions); docs/agentic-education/README.md (Agentic Education scope); public launch post URL; docs/grants/m1-traffic.md baseline.
- Verification: Tag
milestone-m1 + paths on main.
Milestone 2 — Multilingual support & i18n helper (ja + ko) (Week 18)
- Deliverables:
docs/i18n.md, docs/i18n/ja.md, docs/i18n/ko.md; helper + two reports; shipped ja/ko routes (or gap files + plan); EN/ZH CI clean.
- Verification: CI logs + URLs in milestone note.
Milestone 3 — Agentic aggregation, daily refresh, assisted generation & on-chain analysis (Week 32)
- Agentic education package:
docs/agentic-education/playbooks.md, guardrails.md, eval-log.md (≥10 scenarios), cross-links from docs/content-generation/workflow.md; .cursor/skills/ updates on main if applicable.
- Aggregation + daily refresh:
docs/aggregation/sources.md, docs/aggregation/runbook.md, completed docs/operations/daily-update.md (cron, secrets policy none or GitHub-only, failure alerting path); ≥8 PRs in docs/grants/m3-aggregation-prs.md; ≥30 consecutive calendar days ≥95% daily workflow success (documented; platform-wide outages excluded).
- On-chain analysis:
docs/onchain/sources.md + docs/onchain/methodology.md; snapshot scripts + ≥4 onchain:true manifest rows with artifacts.
- Generation:
docs/content-generation/workflow.md + checklist; m3-generation-units.md ≥18 rows; ≥6 with agentic: true + valid playbook_id.
- Verification: Spot-check 3 random
agentic:true rows against eval-log.md patterns; all 4 onchain:true rows; Champion ≥4 spot-checks on agent-opened or agent-assisted PRs.
Milestone 4 — Ongoing maintenance & annual close-out (Month 12 from M1)
- Deliverables:
docs/grants/maintenance-log.md (≥10 editorial/engineering cycles plus ≥48 weekly one-line summaries of daily build health—~1 per week over the grant), ≥4 quarterly update links, docs/grants/final-report.md (agentic adoption metrics, daily refresh reliability, aggregation error budgets, generation throughput, on-chain lessons).
- Verification: Merged files + links.
5. Acceptance criteria
- M1: All §4 M1 artifacts on
main (including docs/operations/daily-update.md stub + docs/agentic-education/README.md); disclaimers accurate; IA doc lists aggregation boundaries.
- M2: Separate
ja/ko validation; shipped routes or documented gaps + plan.
- M3:
docs/agentic-education/ contains playbooks.md, guardrails.md, eval-log.md (≥10 scenarios); m3-aggregation-prs.md ≥8; m3-generation-units.md ≥18 per §1.4 with ≥6 agentic:true + playbook_id and ≥4 onchain:true; daily workflow ≥30 days ≥95% green; allow lists respected; Champion ≥4 spot-checks; no agent-only merges to main.
- M4: Maintenance log (including weekly daily-build rollup) + quarterly posts + final report.
- No false authority throughout.
- Public good: repo public; new prose defaults CC-BY 4.0 where not otherwise constrained by third-party quotes.
6. Funding request and milestone breakdown
6.1 Total request
200,000 CC = 30,000 (M1) + 20,000 (M2) + 30,000 (M3) + 120,000 (M4). M1: launch and aggregation IA. M2/M3: i18n and agentic pipelines. M4: 12 × 10,000 CC for twelve months of maintenance (ops, quarterly posts, final report). Final amounts follow Canton Development Fund Committee process.
6.2 Requested CC (200,000 CC)
| Milestone | CC | Rationale |
|---|
| M1 — Launch + aggregation IA | 30,000 | Production hardening, IA, analytics baseline, first public cut of daily + agentic stubs. |
M2 — Multilingual + i18n (ja + ko) | 20,000 | Helper + separate ja/ko QA + shipped routes or documented gaps. |
| M3 — Agentic aggregation + assisted generation + on-chain | 30,000 | Pipelines, playbooks/eval, editorial, agent API if used (§6.3). |
| M4 — 12-month maintenance (12 × 10,000 CC) | 120,000 | Ops; CC at maintenance month 1 and month 6 (below). |
| Total | 200,000 | |
M4 CC release: 60,000 CC after maintenance month 1 acceptance (maintenance-log.md started, weekly daily-build rollup, §5 M4 on track); 60,000 CC after maintenance month 6 acceptance (mid-grant review). 120,000 CC total M4 = 12 × 10,000 CC planning basis. Invoicing per Fund / Committee rules.
6.3 M3 budget — example mapping (30,000 CC)
| Sub-item | CC (band) | Notes |
|---|
| Aggregation + daily CI engineering (scripts, workflows, monitoring) | 7,000–11,000 | Rate limits, retries, per-day job reliability |
| Agentic education (playbooks, guardrails, eval-log, Skill wiring) | 3,000–5,000 | Maintainer-facing; no unsupervised prod bots |
| Human editorial + fact-check for generated units | 6,000–9,000 | ≥18 units × review depth (includes agentic rows) |
| On-chain / explorer API usage & snapshot storage | 2,000–5,000 | Public indexers only; ToS-respecting quotas |
| Optional agent LLM/API (draft assist only) | 0–5,000 | Omit if no third-party LLM budget |
| Source hygiene / legal spot-check (quotes, logos) | 2,000–4,000 | Brand-kit compliance for on-site graphics |
| Contingency | remainder | Feed/API or upstream schema changes |
Payment: M1–M3: CC on milestone acceptance (§5). M4: §6.2 (months 1 and 6). >6-month grants: 6-month re-evaluation if required by program rules.
Adoption and distribution plan
| Channel | Action |
|---|
| Site | RSS/sitemap where applicable; “Last refreshed” on aggregated and on-chain sections; README badge or link to Actions daily workflow. |
| Repository | Runbooks, docs/agentic-education/, and Skills reusable by forks (swap disclaimers + sources). |
| Public venues | ≥4 quarterly posts (forum / X / Discord—follow channel rules) pointing first to official docs for normative claims; ≥1 post may highlight a reproducible on-chain chart with methodology link. |
| Quantitative habit | Maintenance log captures weekly daily build pass rate; final report states median days-to-recover from aggregation failure (if any). |
| Foundation | Co-marketing when offered; no event-series obligations in this grant. |
Evidence of technical capability
- Shipped stack: HashClawAI/canton-edu — Astro, bilingual
translations.ts, GitHub Actions.
- Live site URL at submission.
- Scheduled refresh: documented daily ~20:00 CST build/deploy and chart jobs (
translations.ts home.buildSyncIntro / buildSyncItems, npm run chart:build, CI)—extended here to allow-listed aggregation and on-chain snapshots under the same daily gate.
- Maintainer agents:
.cursor/skills/ in repo; this work formalizes docs/agentic-education/ playbooks and eval-log.md.
- EN/ZH: prior
translations.ts PRs at scale.
Open-source and reusable outputs
| Output | License |
|---|
| Site + scripts | Repo LICENSE |
| Runbooks, agentic-education docs, workflow templates, on-chain methodology | CC-BY 4.0 (default) |
| i18n helper | Same as repo |
Co-Marketing
Milestone notes for Foundation channels where appropriate; Canton Development Fund credit in contribute / about per brand kit.
Rationale (summary)
Agentic Education for maintainers (agents propose, humans merge): aggregation, daily refresh, on-chain explainers, assisted generation, ja/ko, 12-month maintenance—verified via manifests, eval-log.md, and repo artifacts. CC funds playbooks, evaluation, and guardrails for auditable public-good delivery.
Mapping to proposals/_template.md
_template.md section | Location in this proposal |
|---|
| Abstract | Abstract |
| Specification — Objective | §1 Objective and scope |
| Specification — Implementation | §2 Technical approach |
| Specification — Architectural alignment | §3 Architectural alignment |
| Specification — Backward compatibility | §2 (footer) |
| Milestones and deliverables | §4 |
| Acceptance criteria | §5 |
| Funding | §6 |
| Co-Marketing | Co-Marketing |
| Motivation | Abstract; §1.1 |
| Rationale | Rationale (summary) |
References
