Start Record
Give the new session the project boundary before substantive work begins.
Project:
Workstream:
Objective for this chat:
Prior record or handoff:
Source files being used:
Free method · documented openly
Keep the evidence. Separate the authority. Resume the work.
The Record Method is a practical system for carrying serious AI work across chats, models, files, and time without letting a model’s summary become the project’s truth.
The archive may be large. Working memory should remain bounded. Preserve the full record, decide what currently governs, retain what was superseded, and give the next AI only the state it needs.
Minimum protocolStart Record + End Record
AuthorityHuman disposition only
ContinuationCurrent state + source references
The original continuity protocol
Begin with a boundary. End with a record.
The method began with a simple Storage-chat idea: every serious session should state what it is working from and leave behind a compact, searchable continuity note. The larger system makes that discipline durable.
End Record
Leave the topics, changes, and next step in a form that can be found and resumed.
Topics:
Summary:
Search terms:
Decisions or changes:
Open items:
Source record IDs:
Next step:
Preserve
Keep original conversations and files.
Catalog
Assign projects and workstreams.
Review
Separate proposals from evidence.
Decide
Approve, reject, revise, defer, or supersede.
Continue
Generate the next-chat handoff.
Controller law: the system may be mechanically privileged—able to verify, catalog, compare, and prepare proposals—but it remains decisionally unprivileged. It cannot ratify itself.
Authority is visible
A useful summary is not the source it summarizes.
The Record Method keeps evidence, readable transforms, model proposals, human actions, and approved state in distinct classes. Agreement among models may raise confidence; it does not create authority.
Source evidenceOriginal conversation, file, or verified proof objectPreserved; not overwritten by analysis
Derived readableMarkdown, extraction, summary, index, or mining appendixUseful but subordinate to source
CandidateProposed fact, decision, instruction, conflict, or questionRequires human review
Human dispositionAccept, reject, revise, defer, or supersedeAuthorizes selected output
Current stateBounded project state and open workMay be recorded after explicit approval
Structured disagreement
Three models advise. A fourth audits. The human decides.
Triple Stack Review Packet v1 is a review specification, not a packaged desktop tool. It preserves independent model responses, a blinded comparison, disagreement types, evidence quality, options, and human disposition. It is an advisory process connected to a human-owned record—not a majority-vote truth machine.
AIndependent review
BIndependent review
CIndependent review
04Blinded audit
Human disposition: ACCEPT · REJECT · REVISE · DEFER · SUPERSEDE
Continuity
Which proposal preserves the approved state, contradicts it, or depends on an obsolete decision?
Disagreement
Is the difference factual, interpretive, temporal, scope-based, or a difference in risk and priorities?
Options
What does each path preserve, change, risk, require, and leave for the human to decide?
How the public system fits together
The method governs. Keep catalogs. ChatVaultAI operates the local workflow.
ChatVaultAI
The working Windows product for source custody, local search, Keep, candidate review, confirmed context, backups, and handoffs.
RecordMethodKeep
The catalog layer for unassigned intake, projects, workstreams, evidence classes, source indexes, and continuity outputs.
The Record
The smaller human-approved governing state: decisions, definitions, constraints, supersession, open items, and evidence references.
Current integrated build boundary
Desktop 0.1.16 includes ACCEPT / REJECT / REVISE / DEFER / SUPERSEDE actions, governed current state, deferred boundaries, supersession history, and governed handoffs. The human remains the authority. Version 0.1.16 is a signed release candidate; installer access currently requires GitHub repository access. See ChatVaultAI for current distribution details.
Continue with any model
Use the method manually with any AI tool. Before sending a handoff, select the sources and recipient, remove material outside the task, and approve the exact context. Treat the next model’s answer as a proposal until a human accepts, revises, rejects, defers, or supersedes it.
Current publication status
The method is free to read and use from this site. The Start / End Record template is downloadable now. A separately licensed public repository, validators, and a packaged toolkit are not yet published. The method does not require a ChatVaultAI installation or a particular model.
Use it today
Download the Start / End Record template, operate the method manually, or use the integrated workflow in the current ChatVaultAI build.