Archive · Tranche 2

The evidence curation protocol

The rulebook underneath every register in this archive — the five claim statuses, how a claim entered the record, what it took to close a research question, and the explicit list of things desk research could not settle.

Original title
Evidence Curation Protocol
Original date
22 May 2026
Project phase
Stage 1 — after the first community meeting
Purpose at the time
Define how the evidence base was structured and maintained, so that any agent or volunteer touching the research did it the same way.
Status at the time
Live operating protocol from 22 May until the campaign concluded.
Source provenance
Derived from internal/research/evidence_curation_protocol.md (private working corpus; unpublished, unchanged).
Publication treatment
Substantially intact
Derived / prepared by
Claude (Fable 5) with Adrian Wedd, 18 August 2026
Prepared
2026-08-18
Human review
Adrian Wedd — publication review completed 19 August 2026
Published
2026-08-19
What was changed for publication
  • Published substantially intact. This is the constitution of the project's epistemics and it reads better whole than summarised.
  • The file index in section 1 refers to private working files, several of which are withheld. The paths are left in place because they show the shape of the evidence base; see the corpus map for what each one was and why it is or is not published.
  • Section 5 — the list of things that could not be resolved by agent research and needed a professional — is the most useful part and is unaltered. Every item on it was still unresolved when the campaign ended.

Document begins

Evidence Curation Protocol

Bottom Pub Co-op — Internal | Stage 1 Last updated: 2026-05-22

This document defines how the project’s evidence base is structured, maintained, and handed off. Read this before adding, updating, or referencing any research file.


1. Evidence index

All active research files, their purpose, and current status.

Internal research (internal/research/)

FilePurposeStatus
claim_register.csvWorking record of specific claims needing review before public useLocal only — gitignored by policy (internal/research/*.csv). Not in tracked source. Treat as a working tool, not a durable audit trail. Key claim decisions are recorded in this protocol and in the research questions register.
research_questions_register.mdOpen questions indexed by ID; tracks progress toward answersActive — update status when evidence changes
comprehensive_gap_analysis.mdCross-stream gap analysis: what research is missing, what needs professional adviceReference — update when new gaps identified
outstanding_legal_due_diligence.mdRegister of unresolved legal questions across licensing, structure, and property streamsReference — feeds lawyer brief (#19)
sanity_check_report.mdEvaluation of case study accuracy and internal claim riskReference — static snapshot; superseded for case studies by case_study_cleaned.md
case_study_cleaned.mdVerified case studies with source notes and analogy caveatsActive — source of truth for case study claims
case_study_validation_report.mdDetailed validation against deep researchReference — read alongside case_study_cleaned.md
case_study_verification.mdTriage checklist for case study useReference — see triage bucket table for current status
commercial_hotel_history.mdHistorical record of the Commercial Hotel, CygnetReference — update if new historical material found

Public research (docs/research/)

FilePurposeStatus
official_sources.mdAuthoritative sources agents should verify and expandActive — add new primary sources here, not inline
grant_landscape.mdRegister of potential grant/social finance sourcesReference — verify against current guidelines before any application
eoi_stage_leakage_audit.mdStage discipline audit of the public EOI draftReference — static; re-audit if EOI draft is materially revised

Strategy research (internal/strategy/)

The briefing documents in internal/strategy/ (legal_and_licensing_briefing.md, finance_assumptions_book.md, etc.) are stream-level research, not the claim register. They inform claims but do not supersede it. When a briefing produces a new factual claim or disproves an existing one, update claim_register.csv and research_questions_register.md accordingly.


2. Claim status taxonomy

Every claim in claim_register.csv carries one of five statuses. Use these and no others.

StatusMeaningPublic use
FACTVerified from a named, retrievable sourceState directly. Cite if asked.
ASSUMPTIONPlausible but not verified from sourceInternal use only. Do not state publicly without caveat.
OPENNot yet answerable — needs data, verification, or adviceDo not state. Note as an open question if it arises.
PROFESSIONAL-ADVICE-NEEDEDRequires legal, financial, or technical expertiseDo not state in any form until advice is received and approved.
DEBUNKEDShown to be false or unverifiableRemove from all documents. Do not reintroduce.

These are the same five categories used in the message_house.md proof ladder. Keep them aligned.


3. How a new claim enters the record

A claim is any assertion that could appear in a public document, media pitch, or working-group communication.

Note: claim_register.csv is a local working tool (gitignored). For claims that matter to the project record, the status and recommended action should also be captured in this protocol (§5) or in the research questions register, both of which are tracked in git.

Step 1 — Identify. Is the claim already in claim_register.csv? If yes, update the existing row. Do not add duplicates.

Step 2 — Assign status. Apply the taxonomy above. When in doubt, use OPEN or PROFESSIONAL-ADVICE-NEEDED — never escalate a claim’s confidence without a source.

Step 3 — Record the source. Every FACT claim must name a retrievable source in the source_doc field. If no source exists, the claim is ASSUMPTION or lower.

Step 4 — State the recommended action. What needs to happen before this claim can be used publicly?

Step 5 — Cross-reference. If the claim relates to an open research question (research_questions_register.md), note the Q-ID. If it blocks a workplan task, note the issue number. If the claim is critical (PROFESSIONAL-ADVICE-NEEDED or DEBUNKED), record the decision in this protocol §5 so it survives in git regardless of the local CSV.


4. How a research question gets closed

A question in research_questions_register.md moves from Open → Resolved only when:

  • A named authoritative source directly answers it (FACT standard), or
  • A professional adviser (solicitor, accountant, licensing authority) has provided a formal answer.

Moving from Open → Researched is appropriate when desk research has produced a working answer that still requires professional confirmation.

Questions that turn out to be decisions (not research questions) should be moved to the decisions log (internal/decisions/decision-log.md), not left as perpetually open research questions. See Q-012 for an example.


5. What still needs professional advice

These items cannot be resolved by agent research alone. They are blockers for downstream work.

ItemBlocksRoute
Legal structure: distributing vs non-distributing co-opQ-002, Q-004, C-007, C-008; finance model; governance designSolicitor with co-op experience. See lawyer brief (#19).
Active membership rule draftingQ-003Same solicitor, after Q-002 is resolved.
Fixed returns / securities regulationC-001 to C-003, Q-005No public statement until Disclosure Statement drafted and approved. Solicitor + Registrar.
Liquor licence: natural person pathwayQ-006Researched (natural person required). Formal legal confirmation of mechanism still needed. Same solicitor.
Licence transferabilityQ-007Requires current licence details (CBOS Tasmania check) plus legal advice.
Building works scopeQ-008Heritage architect + building inspector. Cannot assess without physical access (#16, #18).
Revenue and wage assumptionsQ-009, C-004 to C-006Hospitality accountant with Tasmanian comparables. Stage 2.

6. Maintenance rules

Who maintains this. The evidence and research curation portfolio (see #35 and the private/agents/portfolios/evidence-research-curation.md agent spec). In the absence of a named volunteer, agents maintain it between sessions and flag drift to the working group.

Frequency. Review claim_register.csv and research_questions_register.md after any research session, any new external source, or any external communication that makes a claim traceable to this register.

What triggers an update.

  • A new external source confirms or contradicts a claim → update status and source
  • A professional provides an answer to a research question → move to Resolved and add source
  • A claim appears in a public document that is not in the register → add it immediately
  • A DEBUNKED claim reappears in any document → remove and note in the claim row

What does not trigger an update. Internal brainstorm notes, draft documents, or agent session outputs do not update the register until a human has reviewed and approved the underlying claim. Agent work populates drafts; humans approve the move to FACT.


7. Handoff to the working group

The evidence base is ready to hand off when:

  1. claim_register.csv has no claims in OPEN status — each is either FACT, ASSUMPTION, PROFESSIONAL-ADVICE-NEEDED, or DEBUNKED.
  2. research_questions_register.md has no question with status Open that an agent could answer from desk research — those should be Researched.
  3. The lawyer brief (#19) is sent, covering all items in the PROFESSIONAL-ADVICE-NEEDED column above.
  4. This protocol document is included in the admin vault and accessible to the working group.

At handoff, the working group takes ownership of decisions derived from the evidence (structure choice, active membership rule, etc.). The research curation portfolio continues to maintain the record as new evidence arrives.

End of document ← Back to the archive