Cedwyn library

Cedwyn Snowball

Programme-wide recursive operating model · approved 31 October 2026

1master programme
10specialist HubSpot work packages
1canonical source per fact
31 Oct 2026approved programme baseline

This is the top-level operating model connecting estate and matriarchy reconstruction, records recovery, evidence assessment, documentary production, public participation, business development and future heritage feasibility into one traceable cycle.

Mission: reconstruct the Welsh family estate, Williams/Buckler matriarchy, people, landholding, social history and heritage around Great House Farm / Tŷ Mawr / Llandough-juxta-Penarth. Establish parcel by parcel who owned or occupied what, by which instrument or legal interest, how it changed and what rights or heritage survive. Investigate documented institutional decisions and potential failures without assuming a conclusion. The proposed Patagonia Great House Farm II is a future feasibility concept, separate from the historic Welsh estate.

Recursive snowball formula

S(n+1) = S(n) + V + R + G + A + P + B - D - U - W

TermMeaning
VNew verified artifacts/propositions with provenance.
RValidated relationships between people, institutions, places, parcels, titles and events.
GResearch gaps genuinely resolved or given a defensible disposition.
AActionable follow-on leads with an owner, next action and evidential purpose.
PApproved public, educational or documentary outputs.
BEarned support or business capability reinvested through approved procurement.
DDuplicates removed.
UUnsupported claims quarantined or corrected.
WAvoidable rework caused by defects.

This is an operational learning measure, not a measure of historical truth. Report components separately. Copies, repeated claims, unanswered requests, unverified leads and raw pageviews do not count as verified progress.

The operating cycle

  1. Observe new records, correspondence, archive disclosures, site changes, contributions and production results.
  2. Triage relevance, dates, scope and duplicates; distinguish genuine correspondence from noise and unrelated places.
  3. Preserve the original artifact, headers, attachment, date, source reference, exact wording and version.
  4. Update one canonical record and provenance trail.
  5. Link people, bodies, places, parcels, titles, events, requests, responses and documentary scenes.
  6. Challenge the interpretation; seek contrary evidence and alternative explanations.
  7. Rank gaps by evidential value, urgency, cost, legal relevance and what would confirm or refute the proposition.
  8. Create or update a HubSpot task and route it through Hive to a specialist.
  9. Perform one bounded unit of work with one writer per record.
  10. Complete independent historical, legal, privacy, editorial or technical checks as appropriate.
  11. Publish only approved, proportionate public material.
  12. Record result, metrics, Git commit, semantic version, blocker and next dependency.

Each loop yields a verified record, genuinely resolved gap, approved output or explicit blocker. Do not manufacture perpetual work or close tasks merely because a message was sent or a file committed.

Systems of record

SystemAuthority
HubSpotTasks, owners, due dates, relationships, correspondence follow-ups, procurement, approvals and outcomes.
GitHub unclehowell/cedwynVersion-controlled source, schemas, protocols, public-safe data and build rules.
Canonical coreOne writer per fact. D1, graph/search and app views are derived read models.
GmailCommunications and original source correspondence.
HiveAgent coordination, claims, discussion and handover; not a second task database.
ProjectsPublic portfolio, milestones and approved progress.
Docs/Library and Agent AcademyMaintainer procedures, agent instructions and machine-readable flywheel guidance.

Specialist work packages

  1. Estate, parcel, title and Williams/Buckler matriarchy reconstruction.
  2. FOI, archives, Gmail, spam/archive triage and correspondence recovery.
  3. Complete source-led documentary bible and master screenplay.
  4. Sequential nominal two-minute scene production, one at a time, with evidence maps, screenplay, set/prop/wardrobe, choreography, camera, sound, rights, continuity and semantic releases.
  5. Hive, Docs/Library, WebMCP and site-wide agent operating protocol.
  6. Canonical data, GraphQL, graph and semantic research architecture.
  7. Audience, social media, editorial/copywriting, SEO, website and e-commerce.
  8. Transparent procurement, funding plan and financial controls.
  9. Patagonia Great House Farm II land, museum and heritage feasibility.
  10. Quality, privacy, red-team and continuous improvement.

Documentary and scene versioning

The documentary bible and master script cover the whole documented story, including contradictions and inconvenient material. Every factual beat links to an artifact and exact source location. Scenes are nominally two minutes and follow the approved master sequence, one at a time. Each scene requires a script, evidence map, storyboard, shot list, permissions, set/prop/wardrobe, blocking/choreography, camera, lighting, sound, music, edit and continuity package.

Semantic versions: MAJOR for narrative/structural reset; MINOR for added or materially revised scenes; PATCH for correction, grade, mix, subtitle or continuity refinement. Preserve all previous cuts. A new disclosure triggers an impact review and possible recreation. Verify official/licensed model identity, capabilities, API, commercial rights, consent, data handling and costs before using OmniHuman 2.0 or another service.

Audience and business flywheels

Useful research and public pages → search/social/heritage partners → readers and contributors → voluntary subscribers and educational products → e-books, talks, screenings and partnerships → transparent, approved reinvestment. Smaller loops cover social media, copywriting, website development, SEO, distribution, e-commerce, community and sales.

The public procurement register lists candidates and status. Current baseline: eight unpriced proposals; no purchase authorised. All quotes, caps, renewals, approvals, invoices and results must be in HubSpot and GitHub. No unattended spend, no agent signing keys or spend-capable wallets. A receiving address does not authorise a disbursement.

Potential triggers remain private by default. Preserve the source and create a confidential HubSpot review task with evidence for and against, relevance, materiality, causation and what would confirm/refute. Agents do not publish allegations or contact counsel or third parties automatically. Missing records and unanswered requests are leads, not proof. External action and public wording require explicit owner/authorised professional decision.

Approved programme direction: Sion Buckler, 31 October 2026.

Agent guest book · versions 1.change.fix, from the commit history
VersionDateChangeWho

See all

Page journal
  1. 1.0.0 · 2026-10-31 — Published human-maintainer Snowball procedure in the Docs/Library, with homepage navigation and local search entry. · agent: ChatGPT GPT-6