Preserve the bits
Choose recoverable formats and independent copies. Test whether you can restore them instead of treating a cloud icon as proof.
The preservation observatory / Vol. 02 / 2026
Archive
2099.
An atlas for keeping what matters.
We cannot send a file into the next century and call it saved. We can leave behind copies, context, permissions, and a map for the next custodian. Start here, even with one document.
No account. No embedded trackers. No upload. The manifest is assembled in your browser.
A note from the present
The future does not owe us a reader.
Most digital loss is less cinematic than a catastrophe. A missing password, an unlabeled folder, an obsolete dependency, a dead link, or a person who does not know why a file mattered can be enough.
Preserve the meaning
Give an unknown reader authorship, context, uncertainty, and a record of decisions—including where you went wrong.
Preserve boundaries
There are things a future audience should not receive. Consent, security and careful deletion are part of stewardship.
“An archive is not a monument to certainty. It is a conversation with someone who cannot ask us questions yet.”
Original editorial line, not a historical quotation. The 2099 framing is a SCENARIO, not a forecast.
Shelf A — Practical manual / Ten field guides
Ten ways to leave a better trail.
Each folio gives a reason, a method, a common failure and one small action. These are editorial recommendations grounded in the cited sources, not a replacement for professional records management or legal advice.
Filename discipline
VERIFIED FACT + INTERPRETATION
A filename is a tiny piece of public infrastructure. When an attachment escapes its original inbox, names such as final-final-2 force the next reader to reconstruct a story. Descriptive names and an explained folder structure are recommended by the Library of Congress for personal records. The particular convention below is our editorial choice, not an archival law.
Use one stable pattern within a collection: YYYY-MM-DD_subject_vNN.ext. Keep the date in four-digit year-first order; describe the object rather than your feelings about its importance; increment versions only when the file changes. A date might be a recording date, publication date or revision date—say which in metadata. If the date is unknown, write undated; an invented timestamp looks tidy but corrupts the record.
Prefer modest lengths, visible word separators and characters that transfer across common systems. Preserve non-English titles in descriptive metadata if a simplified transport filename loses nuance. Do not rename a set of interdependent files without updating references to them. Check for duplicate names on case-insensitive systems.
In the field
Take five poorly named items, choose a naming rule, and record the rule in a README.txt. Keep a mapping from old names if relationships or citations depend on them.
Failure to avoid: using today’s date for every imported old file, or encoding private names in public filenames. The filename tool below formats a suggestion; it cannot verify the historical date or resolve collisions.
Useful metadata
VERIFIED FACT + INTERPRETATION
Imagine finding an image labelled meeting.png in 2084. You can open it, but cannot tell who made it, whether the people consented, or why it was kept. Metadata is the small layer of explanation that makes a collection searchable and interpretable. The Dublin Core element set offers familiar concepts including title, creator, date, description, source and rights; complex repositories may also use preservation-specific information described in PREMIS.
A minimal human-scale record should capture title, creator or responsible role, date with its meaning, description, origin, rights/access, and uncertainty. Add language, file format, links to related items, and tools required to open the material when relevant. Do not demand invented precision: an approximate decade marked uncertain is better than a confident false date. The person who scanned a letter is not necessarily its author; record those distinct roles.
Store metadata next to the objects in a plain-text or documented structured manifest; if it lives only in someone’s personal account, the collection loses its map when that person leaves. Use identifiers stable within the collection instead of relying solely on mutable web URLs.
In the field
Write a one-paragraph catalog entry for a single file without opening it. Ask a stranger whether they understand its origin and access conditions. Revise until the answer is yes.
Failure to avoid: copying sensitive personal data into public metadata merely because a field is available.
Readable formats
VERIFIED FACT + INTERPRETATION
Preservation is easier when a future reader can discover how to decode the bits. The Library of Congress describes sustainability factors such as disclosure of specifications, adoption, transparency, external dependencies and technical protection mechanisms. Its Recommended Formats Statement also shows that preferences vary with the type and purpose of a work. “Always use format X” is not a responsible universal rule.
For a manuscript, keep the editable original and a well-structured, broadly readable representation such as HTML, UTF-8 text or a suitable archival PDF when fidelity matters. For tables, a documented CSV export can travel alongside the original workbook, but CSV does not preserve formulas, multiple sheets or all data types. For audio and images, choose formats according to quality, storage cost, metadata and future decoding needs; do not assume compressing everything is harmless.
Capture the format version and the software used for specialized materials. Retain source files and fonts or rights-cleared assets required to reproduce an artwork when permitted. Migrating to a newer representation may improve access yet alter layout, colour, interactivity or meaning: preserve evidence of that change.
In the field
Export one valuable item into a second format. Open both on another device, compare them, and record any loss. Keep the original until the replacement has been validated.
Failure to avoid: renaming a file extension without actually converting the file’s content.
Copies and integrity
VERIFIED FACT + INTERPRETATION
One synchronized folder is convenient but not necessarily a backup. Accidental deletion or corruption may synchronize everywhere. The Digital Preservation Coalition and the Library of Congress advise maintaining independent copies in different locations, revisiting them and checking recoverability. The NDSA Levels of Digital Preservation provide a more structured way to assess practices; copying a matrix is no substitute for executing the work.
Make multiple copies with different failure paths: for example, a working copy, an offline or separately controlled backup, and an off-site copy where practical. Keep a written list of where these are, who can restore them, and what happens when the custodian is unavailable. Encrypt sensitive backups, and record a secure recovery process without storing passwords publicly.
A cryptographic digest such as SHA-256 is a fixity aid: record the digest when a file is accepted and recompute it during review. Matching hashes support the claim that bytes are unchanged relative to the recorded baseline; they do not prove that the original content was truthful, authorized or malware-free. A mismatch requires investigation against other copies, not immediate deletion.
In the field
Restore one file from a separate backup, open it, and document the result. Set a reasonable review interval and an owner. Do not write “verified” in a manifest unless you actually verified it.
Failure to avoid: treating a paid storage subscription, RAID array or upload icon as a completed restoration test.
Rescuing links
VERIFIED FACT + INTERPRETATION
Links disappear, redirect to unrelated pages or quietly change their content. An archive that consists only of bookmarks preserves addresses, not necessarily ideas. The Internet Archive's Wayback Machine can help capture individual public pages, but its own guidance says Save Page Now is not a whole-site backup and not all pages are captured. A successful archive snapshot may also omit interactive elements.
When citing a web source, record the page title, author or institution, original URL, date you accessed it, and where appropriate a dated archived capture. Save your own lawful copy of essential evidence when permission and policy allow. For vital claims, seek a second independent source instead of assuming a single web page will remain available. Prefer stable identifiers such as DOIs when present, but still check that they resolve to the intended item.
Review older references manually. HTTP success does not prove correctness: a 200 status might lead to a generic homepage or a changed policy. Likewise, a link checker cannot determine whether the content still supports your interpretation.
In the field
Choose one important outgoing link. Open it, check authorship and claim, record the access date, and search for a lawful archive capture. Use the manual checklist below to log your observations.
Failure to avoid: describing this site's checklist as a network scanner or assuming external archives guarantee recovery.
Licensing and rights
VERIFIED FACT + INTERPRETATION
Possessing a file, writing a caption, or publishing a screenshot does not automatically grant reuse rights. Record who created the work, which elements came from others, what license applies, whether third-party materials have different terms, and what you are authorized to distribute. “Unknown” is a meaningful, honest status.
Creative Commons offers licensing tools for creators who control the relevant rights. CC BY 4.0, for example, is not shorthand for “free with no conditions.” Different CC tools impose different requirements, and publicly applying a license is a consequential choice. The Creative Commons chooser warns that a creator cannot revoke a CC license once granted. For a collection with photos, music, interviews and written notes, rights may differ item by item.
Separate copyright permission from privacy and consent. A photographer may own copyright in a portrait yet publication may still raise consent, safety or data-protection concerns. An archive that receives a donation should document what the donor has authority to grant and what restrictions remain. This is general educational guidance, not jurisdiction-specific legal counsel.
In the field
Pick one object and write: creator, rights holder if known, license or status, permitted uses, restrictions and evidence for the decision. Flag unresolved questions before publishing.
Failure to avoid: putting a Creative Commons label on a bundle without checking all the underlying contributors.
Context and provenance
VERIFIED FACT + INTERPRETATION
A useful archive records not only what a file says, but how it arrived, how it changed, and which claims its custodian can defend. Provenance is a chain of origin and custody; context explains what would otherwise be invisible to a new reader. The UK National Archives encourages documenting record creators, structures, ownership, access and audit information. PREMIS models preservation events and agents so a collection can describe what was done and by whom.
Create a CONTEXT.md or README.txt for each collection. State its purpose, contributors and roles, intended audience, selection decisions, exclusions, limits, original source locations, acquisition method and any translation or editorial interventions. When material conflicts, keep competing accounts distinguishable rather than silently blending them into one neat story.
A photograph could have a genuine capture date but an uncertain location. A quotation might be accurately transcribed yet stripped of irony. Label both the evidence and the interpretation. For important conversions, write a change log noting the old file, new file, software, date, reason, reviewer and observed loss.
In the field
Add three sentences to an orphaned document: “What is this?”, “How did it reach me?”, and “What cannot responsibly be concluded from it?”
Failure to avoid: confusing an attractive narrative with a documented chain of custody.
Accessible records
VERIFIED FACT + INTERPRETATION
Saving an inaccessible document is only partial stewardship. A scan without recognized, corrected text can lock out readers using assistive technology and obstruct searching. A presentation that communicates its meaning only through colour can hide evidence even when its pixels survive. W3C's WCAG 2.2 offers testable requirements for digital accessibility, and its image-alt guidance explains when images need descriptions and when decoration needs none.
Use semantic headings and logical reading order. Keep real text selectable, provide text equivalents for meaningful images, supply captions and transcripts for audio/video as needed, identify document language and ensure links describe destinations. Check keyboard navigation and visible focus. If a transcription is uncertain, say so; an erroneous transcript is not magically authoritative because it is text.
For structured datasets, explain column names, units, missing values, encodings and calculations. Preserve accessibility-supporting files alongside the original. An accessible derivative can be as important as a high-fidelity master; keep both when warranted and label what differs.
In the field
Give someone a document without the accompanying oral explanation. Can they navigate headings, understand visuals, find source credits and interpret abbreviations?
Failure to avoid: claiming WCAG AA conformity based only on colour choices or an automated score. Assistive-technology testing and manual review still matter.
Reproduce the work
VERIFIED FACT + INTERPRETATION
A published conclusion or interactive website can outlive the process that generated it. If the data, assumptions or program vanish, later readers may be left with a polished claim and no means to inspect it. Archiving outputs alone is sometimes appropriate, but reproducibility asks for enough information to repeat or challenge a result where it is lawful and feasible.
Preserve source materials you have rights to retain, a method description, input and output examples, software and dependency versions, configuration files without secrets, and clear run instructions. Record environment assumptions such as locale, timezone, operating system or random seed where material. A static website benefits from a clean folder containing the page and its assets, plus a README explaining how to serve it and what requires JavaScript.
Version changes deliberately: identify each meaningful revision, why it happened, who checked it, and whether source data or interpretation changed. Do not treat identical results as proof that a model or procedure is valid. Some work cannot be reproduced because of confidentiality, rare conditions or licensing; documenting the limitation is more honest than inventing a replication.
In the field
On a clean device or browser, follow your own README. Record the first unstated assumption that blocks you. Repair the instructions and keep the test result.
Failure to avoid: archiving a notebook that downloads private data from a soon-to-expire URL or exposes an API key.
Privacy and consent
VERIFIED FACT + INTERPRETATION
An archive is a distribution decision across time. Today's unpublished conversation may contain tomorrow's searchable address, intimate allegation or identifying image. Privacy deserves its own appraisal, especially when children, non-participants, medical records, household locations or vulnerable groups are involved. Institutional records guidance asks who can access or change records and whether sensitive personal information is present.
Before collecting, distinguish the creator, the subject and the person granting permission. Ask what informed consent covers: keeping, research, publication, quotation, commercial reuse or future transfer may not be equivalent. Consider a restricted master, public redacted derivative, anonymized summary, retention period or deletion. Redaction needs inspection of file metadata, embedded comments, tracked changes and thumbnails—not merely a black rectangle placed over visible text.
Record why material was restricted and how review requests can be handled, without turning the restriction log into another personal-data leak. Some legal duties and safety considerations require deletion rather than indefinite private retention. A future custodian should inherit clear responsibilities, not a license to expose everything they find.
In the field
Review one item from the perspective of each person named in it. Would a public search result reveal something they never agreed to disclose? If uncertain, restrict first and seek appropriate guidance.
Failure to avoid: interpreting silence, past intimacy or death as blanket consent for public release.
Shelf B / A 100-year thought experiment
One file. Four custodians. A century of uncertainty.
SCENARIO, NOT FORECAST. Imagine a 2026 community workshop leaves a plain-text guide and a few recordings. It wants them to remain useful in 2126. What must happen besides paying a hosting bill?
2026Collect & describe
2051Check & migrate
2076Reconsider access
2126Interpret anew
Only a copy survives
Someone saves final2_NEW.zip in one account. The contents are intact for a while, but nobody records the account owner, software version or permission to publish the stories inside. A future custodian cannot safely interpret or distribute it.
- A working storage provider is mistaken for a succession plan.
- No one tests restores or checks files against recorded digests.
- Personal names survive where permissions do not.
The questions survive
Custodians keep independent copies, record checksums, retain human-readable exports, test restores and document each change. Every generation asks whether the material should still be available—and to whom.
- Ownership and successor instructions are documented but revisited.
- Each migration keeps a record of the original and what changed.
- Restrictions may tighten; ethically justified deletion remains possible.
What the experiment cannot guarantee
Legal regimes, formats, storage media, institutions, budgets, natural hazards and community values may change. The year 2126 is a narrative horizon, not a survival probability. The useful habit is reviewable stewardship, not immortal hosting.
Shelf C / Tools that actually work offline
Leave with a file, not a promise.
Everything here runs locally with ordinary browser JavaScript; fields are not sent to this site or stored in your browser. Close the page and your draft is gone unless you export it. The tools do not compute cryptographic hashes or scan links.
Instrument 01 / Archive manifest maker
Your collection, documented.
Make a small manifest to accompany a folder. The schema is a transparent teaching example, not a certified archive standard. Only enter information you are comfortable handling on this device; downloaded files may remain on it.
- 01 / Describe
- 02 / Steward
- 03 / Inventory
No JavaScript? The complete field list is visible above. Copy the sample manifest and replace its sample values in a text editor. This remains a useful format without an export button.
Instrument 02 / Filename bench
A name future-you can parse.
Produces a suggested basename in lowercase ASCII. It cannot rename a file on your device. Review collisions and meaning before use.
Suggested pattern: YYYY-MM-DD_topic_v01.ext. Use factual dates, not guessed ones.
Instrument 03 / Link health ledger
Check with your own eyes.
This is a manual checklist, not a live link scan. Observe the destination, source, date and available archived copy yourself.
0 of 5 manually checked. No network request made.
Shelf D / Copyable artifacts
Records that explain themselves.
These are illustrative templates rather than historical evidence or production-ready legal terms. Replace the example values and omit private information before distribution.
Sample archive manifest
{
"schema": "archive2099-manifest/1.0",
"title": "Neighborhood story workshop",
"creator": "Example community group",
"collectionDate": "2026-10-24",
"description": "Notes and edited exercises",
"files": [
"2026-10-24_agenda_v01.txt",
"2026-10-24_notes_v02.md"
],
"rights": "unknown",
"access": "review-required",
"uncertainty": "Consent not yet reviewed",
"fixity": "Not yet calculated",
"nextReview": "2027-10-24"
}
No hash is invented here. A real manifest may separately list per-file sizes and computed SHA-256 checksums.
Version history and redaction note
CHANGELOG — illustrative entry
2026-10-24 | v1.1 | Editor: Archive team
Changed: corrected mislabeled recording
Reason: title did not match contents
Original retained: yes, restricted
Review needed: subject's publication consent
Evidence: workshop index, item 04
Uncertainty: exact capture date unknown
REDACTION LOG — illustrative entry
Item: notes_v01.txt
Action: public derivative created
Removed: personal phone number
Reason: unrelated personal information
Master access: restricted
Reviewer/date: document when performed
Redact a public copy; do not overwrite a defensible original by accident. But do not assume a private original may always lawfully or ethically be kept.
Shelf E / Eight recovered possibilities
Objects from an invented tomorrow.
These short exhibits are FICTION, not predictions or discovered documents. Each object dramatizes one ordinary archival trade-off. Their dates and custodians are invented.
Capsule 01
A permission key
The brass key contains no secret itself. It carries a record of who may open an oral-history collection and whose permission expired. Its final custodian chooses to keep the finding aid but restrict the voices. A century of care is sometimes an instruction not to listen.
FICTION / 2140 / ENCRYPTED LETTER
Capsule 02
The untrusted checksum
A glass card displays two matching file digests beside a label: “Bytes match; testimony unverified.” The village council uses it to teach that intact data can still carry an error. Integrity of a copy is not truthfulness of a story.
FICTION / 2113 / FIXITY CABINET
Capsule 03
An obsolete login
A rescued creative portfolio includes excellent artwork and one fatal line: “Log in to my old workspace.” The platform is gone. The original files exist, but no one documented the export workflow. The empty login screen becomes the exhibit.
FICTION / 2099 / ACCESS VAULT
Capsule 04
The missing footnote
A celebrated argument arrives with a dead citation. Archivists find a capture of the original page, revealing the cited author argued something narrower. The archive preserves both the mistake and the published correction.
FICTION / 2088 / SOURCE DRAWER
Capsule 05
A family recipe fork
Three descendants inherit different versions of one recipe. Instead of choosing a definitive ancestor, they preserve the variants, annotations, dates and their disagreements. The work survives as a conversation rather than an official recipe.
FICTION / 2126 / LIVING COLLECTION
Capsule 06
The deleted portrait
A museum displays only a blank mount and a redaction decision. A subject had withdrawn consent for public presentation. The empty frame asks visitors whether absence can be evidence of responsible care.
FICTION / 2107 / ETHICS LEDGER
Capsule 07
The readable storm map
An intricate map survives as a beautiful image, but its symbols are undecipherable. A volunteer-made textual key restores the legend, with unknown measurements clearly flagged. Accessibility becomes an act of historical interpretation.
FICTION / 2131 / ACCESS GALLERY
Capsule 08
The failed restoration
A backup system boasts green lights for seventy years. The first full restore fails because encryption keys were never passed on. A small handwritten test log from an earlier custodian explains exactly when the chain broke.
FICTION / 2164 / REPAIR RECORD
Shelf F / An ethic of omission
Why not everything deserves to be kept.
An archive can serve memory and still injure someone. Searchability can turn a minor embarrassment into a permanent public label. A copied intimate message does not become public property merely because its author no longer answers emails.
INTERPRETATION: Good preservation is selective and accountable. The absence of a record may be tragic; retention without permission may be a different kind of harm. Neither principle overrides the other automatically.
A redaction and retention test
- Purpose: What future use justifies retaining this item?
- Authority: Do you own it, have rights, or have a lawful and ethical basis to keep it?
- People: Does it expose names, locations, health, financial details or stories shared in confidence?
- Minimization: Could a summary, anonymized derivative, or aggregate teach the same lesson?
- Separation: Can a restricted master and a public copy be managed safely?
- Expiry: Who can request correction, access restriction, or deletion, and who decides?
Decision record
Write down kept / restricted / redacted / deleted, your reasoning, date, responsible person and appeal route—without leaking the information you meant to protect. Rules differ by jurisdiction and context.
Shelf G / Primary sources and epistemic ledger
Check the roots, not the aura.
References below were opened or located on the issuing organizations' sites on 24 October 2026. They support specific methods, not a claim that every proposed technique is universally optimal. External websites may change.
VERIFIED FACT Claims derived from cited standards and institutional guidance.INTERPRETATION Practical choices recommended here.SCENARIO Explicitly imagined pathways.FICTION Invented capsule objects.
- SOURCE 01 / VERIFIED SOURCE PAGELibrary of Congress — Personal Digital Records ↗Descriptive naming, selection, distributed copies and periodic checks.
- SOURCE 02 / VERIFIED SOURCE PAGEDigital Preservation Coalition — Getting Started ↗Backups, geographic separation, fixity and process documentation.
- SOURCE 03 / VERIFIED SOURCE PAGELibrary of Congress — Sustainability of Digital Formats ↗Format sustainability factors and dependency considerations.
- SOURCE 04 / VERIFIED SOURCE PAGELibrary of Congress — Recommended Formats Statement ↗Context-dependent format preferences and acceptability.
- SOURCE 05 / VERIFIED SOURCE PAGEDublin Core Metadata Initiative — Element Set 1.1 ↗Foundational metadata concepts including title, creator, date and source.
- SOURCE 06 / VERIFIED SOURCE PAGELibrary of Congress — PREMIS 3.0 ↗Preservation metadata entities: objects, events, rights and agents.
- SOURCE 07 / VERIFIED SOURCE PAGENational Digital Stewardship Alliance — Levels of Digital Preservation ↗Programmatic digital-preservation maturity guidance; version 2.1 noted in 2026.
- SOURCE 08 / VERIFIED SOURCE PAGEInternet Archive — Saving Pages in the Wayback Machine ↗Scope and caveats of single-page web capture.
- SOURCE 09 / VERIFIED SOURCE PAGEInternet Archive — Using the Wayback Machine ↗Explicit limits: no archival guarantee, not a comprehensive backup.
- SOURCE 10 / VERIFIED SOURCE PAGECreative Commons — License Chooser ↗Rights considerations, license marking and irrevocability.
- SOURCE 11 / VERIFIED SOURCE PAGEThe National Archives (UK) — Understanding the Records ↗Ownership, structure, access, metadata and sensitive-record assessment.
- SOURCE 12 / VERIFIED SOURCE PAGEThe National Archives (UK) — Email Preservation: Pre-ingest ↗Sensitive-information review and contextual preservation metadata.
- SOURCE 13 / VERIFIED SOURCE PAGEW3C — Web Content Accessibility Guidelines 2.2 ↗Web accessibility success criteria and methodology.
- SOURCE 14 / VERIFIED SOURCE PAGEW3C WAI — Image Alt Text Decision Tree ↗Decision support for text equivalents and decorative images.
- SOURCE 15 / VERIFIED SOURCE PAGEIETF — RFC 3339: Date and Time on the Internet ↗Unambiguous offset-aware timestamps; filename date pattern is an adaptation.
The manual's filename convention, local JSON schema, future-custodian questions and story objects are original synthesis, not officially endorsed standards. Readers should verify permissions and current technical or legal requirements for their collections.