What this method is for
Personal archives rarely arrive as orderly collections. The difficulty is not simply storage. It is preserving the relationships among materials without forcing them into a false single story.
They accumulate as notebooks, loose pages, photographs, cloud libraries, chats, academic files, drafts, source packages, audio, public posts, repositories, and formats their creator no longer remembers making.
The Fieldlight Archive Method is a repeatable human–AI workflow for bringing a heterogeneous personal corpus into usable form while keeping source and representation, evidence and interpretation, private custody and public release, creator and collaborator, and uncertainty and conclusion distinct.
The method is designed for archives that are still alive: still growing, still being remembered, and still changing the creator's understanding of the work.
The central unit is the source, not necessarily a file.
A bound notebook may have hundreds of photographs. A tarot reading may have several views. A fiction work may survive as a Scrivener project, Word file, PDF, public page, and audio recording. A public-post archive may contain thousands of fragments. Those files are representations of sources or works; they do not automatically become thousands of equally weighted analytical objects.
This distinction prevents two common failures: mistaking file count for the size or significance of the archive, and analyzing every representation independently until the work or event they represent disappears.
Begin with the largest coherent source unit that can be responsibly described. Promote an individual representation when it is independently cited, materially different, differently restricted, or necessary to establish a version or evidence claim.
No single preservation surface should be presumed complete.
A preservation surface is the mechanism through which a record survives: a platform export, local application cache, app library, filesystem copy, repository, screenshot, public page, or physical object. One source may leave different—and differently incomplete—representations across several surfaces. One meaningful relationship may also require evidence from more than one domain.
Completeness must therefore be stated against a declared set of inspected surfaces and dated source states. The central unit remains the source; a claim about historical completeness belongs to the reconciled surface set.
The ten-stage path
The stages are ordered, but not perfectly linear. Later evidence may require a corrected source boundary, a split session, a revised date basis, or a retracted interpretation. The trail records those changes instead of hiding them.
Scope
Define one acquisition boundary before touching the material. State what is included, what is excluded, which preservation surfaces and dated states were inspected, and what privacy default applies.
Completion test
Another person can state exactly what the accession covers—and what it does not.
Preserve
Copy or export originals without editing them. Retain original filenames, directory structure, metadata, and the surface and state from which each representation was acquired.
Completion test
Counts reconcile within each acquired state, preserved files verify, and no state has been silently merged into another.
Inventory
Assign stable identifiers and record source identity, custody, rights, privacy, preservation location, surface and state, representation count, and processing status.
Completion test
Every representation points to a source or unresolved assignment, and surface-specific absence is not treated as nonexistence.
Derive
Create working copies for OCR, thumbnails, contact sheets, format conversion, transcription, annotation, or redaction. Document each transformation and preserve its relationship to the master.
Completion test
The master remains untouched and the derivative's origin is traceable.
Orient
Create a navigation record before deep interpretation. Declare coverage as sampled, near-complete, or complete; map structure, dates, material classes, privacy risks, named works, standout regions, gaps, and uncertainties.
Completion test
A later reader can navigate the source and knows which portions were not reviewed.
Select
Define a bounded evidence packet around one question, event, motif, work, version lineage, or contradiction. A dense source does not stand for the corpus; a striking artifact does not authorize a whole-period conclusion.
Completion test
The packet is small enough to review closely and every included record has a stated reason.
Transcribe or extract
Create faithful text, structured description, or data from selected representations. Preserve unreadable passages, uncertain identities, missing pages, ambiguity, and version distinctions.
Completion test
Another reviewer can locate the evidence and distinguish what was seen from what was normalized.
Connect
Record substantive relationships as claims with source, target, relationship type, evidentiary basis, confidence, privacy, note, and next verification action. Keep creator knowledge, observation, documentation, external sourcing, and inference separate.
Completion test
Distinct evidence bases have not collapsed into one voice.
Synthesize
Write a dated analysis using traceable claims. State the question, corpus boundary, strongest evidence, counterevidence, contradictions, missing evidence, and confidence. Meaning is allowed; its route must remain visible.
Completion test
The conclusion can be audited backward to sources and its failure conditions are identifiable.
Review and publish
Create a public derivative only after reviewing privacy, rights, authorship, third-party exposure, locations, credentials, metadata, quotation, context, and the difference between source truth and release copy.
Completion test
The release is intentional, traceable, appropriately redacted, and does not imply that public access transfers ownership.
Human and AI responsibilities
This is a human-governed method that uses AI as a high-throughput archival and analytical collaborator.
The collaboration works only if correction is treated as evidence. A rejected grouping, withdrawn claim, or creator correction documents where automation or interpretation exceeded the record.
What Version 1.0 has been tested on
The method was developed through active use rather than designed in the abstract. Its first private implementation has been tested across:
- Nine storage-unit notebooks, associated loose material, and 526 documentation photographs.
- A photo-based divination corpus in which 266 representations were resolved into 258 sessions after eleven over-grouped pairs were split.
- Mixed academic source trees containing creator work, collaboration, instruction, research material, datasets, and third-party sources.
- Born-digital writing packages, including project containers and multiple export formats.
- Public writing, web pages, audio, social-post series, and public software repositories.
- Cross-domain version and lineage questions requiring negative evidence and abandoned branches to remain visible.
These applications establish that the workflow is operational across multiple media. They do not prove that its categories are universal or that AI-assisted interpretation is inherently reliable.
Failures that changed the method
Version 1.0 includes rules learned through error.
Search results are not an accession.
A changing visual-search population while cloud sync was unsettled led to a new rule: freeze a dated cutoff, preserve it, document instability, and treat later arrivals as an incremental accession.
Photographs are not sessions.
Full-resolution review showed that eleven initially grouped image pairs were distinct arrangements. The source register was split without recycling existing identifiers.
A dense source is not the whole corpus.
The notebook with the largest capture run became a useful pilot, but could not stand in for the other eight. Breadth-first orientation now precedes deeper batch analysis.
Folder location is not authorship.
Academic and project folders mixed creator, collaborator, instructor, classmate, and third-party work. Role-specific authorship and rights fields are now required.
A narrow accession cannot support a whole-body verdict.
An early synthesis over-weighted historical notebooks and writing packages while missing current technical and institutional work. Whole-corpus claims now require a declared cross-domain boundary and active search for disconfirming evidence.
Every available export is not every available preservation surface.
A conversation created April 23, 2025 at 19:08:39 UTC survived in two local cache states dated April 24 at 04:01:08 and 05:30:01 UTC, yet its stable identity and exact title were absent from all three official exports. The cache is not presumed complete; the finding requires surface-state reconciliation before a completeness claim.
Compelling symbolism is not validation.
Symbolic and narrative relations may guide selection and generate questions. They do not establish ontology, causality, prophecy, or external fact; later reasoning must be allowed to validate, falsify, constrain, or operationalize them.
Minimum viable archive
A small archive does not need a complex database. It needs a durable trail:
- originals;
- an integrity or transfer note;
- a source register;
- a preservation-surface statement naming inspected states and known gaps;
- working derivatives;
- one orientation map per source or coherent series;
- bounded evidence packets;
- transcriptions or structured descriptions;
- connection claims with basis and confidence;
- dated analysis memos;
- separate, reviewed public derivatives.
Implementation package
Version 1.0 includes the working materials needed to adopt or adapt the method.
Public derivative, not private master. The companion workbook publishes reusable structure, aggregate field-test evidence, and a public-work register without exposing the private archive catalog. The method license does not extend to the private archive or to referenced personal works.
What the method protects
The Fieldlight Archive Method is not a promise that enough data will reveal one final explanation for a life or body of work.
It protects something more useful: the ability to investigate without destroying the conditions of investigation.
It preserves the original before interpretation. It makes uncertainty inspectable. It permits AI to help at scale without granting it silent authority. It allows private material to contribute to understanding without making private material public. And it leaves a trail by which later readers—including the creator—can disagree, correct, and begin again.