Information Management: A Proposal

How to read the original

1. Read Losing Information at CERN and identify the organizational failure.
2. Read Linked information systems and The problem with trees, then model one cross-cutting relationship.
3. Read A solution: Hypertext and distinguish the proposal from later Web protocols.

The problem is losing relationships

Berners-Lee’s proposal addresses changing organizational knowledge: people, equipment, documents, and projects do not fit neatly into one permanent hierarchy. Links offer multiple routes through the same information without requiring a central redesign whenever relationships change. Read the proposal as an argument for incremental, distributed information management. Do not retroactively treat every later browser feature or HTTP convention as a mechanism already specified by the 1989 document.

Original reconstruction: a laboratory with two classifications

A laboratory maintains projects P and Q. Both use sensor S, which is calibrated by engineer E according to procedure D. In a strict project tree, placing S only under P makes Q’s dependency hard to discover. Copying S’s record into both branches creates a synchronization obligation: when calibration changes, which copy is authoritative?

Represent each entity once and add typed links: P uses S, Q uses S, E maintains S, and S follows D. A learner entering through P can reach S and then D. A technician entering through E can find S and both dependent projects, provided the system offers the corresponding reverse-link query or stored reverse relationships. Ordinary one-way hyperlinks do not automatically supply a complete backlink index.

Now replace engineer E with F. The organizational chart changes, but the sensor’s identity can remain stable. Update the maintenance relation and preserve an accountable record of the change. If the identifier instead embeds the employee’s current department and role, a routine reorganization can invalidate references. Naming policy becomes part of continuity, not merely an aesthetic URL choice.

Trace independent publication

Suppose the sensor record is maintained by one group and procedure D by another. The first group publishes a link to D without copying its full contents. When D is corrected, readers can follow the existing reference to the maintained resource. This reduces some duplication but introduces a dependency on the second group’s availability and naming policy.

If D disappears, the link does not repair itself. A useful interface should distinguish unavailable content from evidence that the procedure never existed. It can retain title and provenance metadata, record the last successful retrieval, or point to an archived version when policy permits. If reproducibility requires the exact procedure used for an old experiment, a link to the latest mutable document is insufficient. A version identifier or preserved snapshot is needed.

These are design extensions for the lab exercise, not claims that the proposal solved preservation completely. Decentralization makes participation and local ownership easier in some respects while distributing responsibility for longevity, access control, and quality.

Separate browsing, search, and authority

A link answers “what relationship did this author record?” It does not by itself answer “is this statement true?” or “is this user allowed to read the destination?” Search helps discover possible starting points, but cannot substitute for explicit relationships when a task depends on them. Likewise, one central search index can coexist with independently maintained documents; decentralizing publication does not imply every supporting service is decentralized.

Evaluate the prototype with concrete tasks: identify every project affected by a sensor replacement, find the procedure used for an earlier calibration, and locate the person currently responsible. Measure successful answers and stale references, not only link count. A larger graph can still make important facts difficult to find if labels and relationships are inconsistent.

The enduring lesson is to design identifiers, links, ownership, and editing around how knowledge changes. A document system should preserve useful paths through evolving work while making its gaps visible. The proposal opens that design space; a working system must still earn its claims about discovery, trust, and continuity.

Reconstruction exercise

Design a five-node graph for two courses sharing one prerequisite, its assessment, and its maintainer. Give explicit link meanings. Then trace what should happen when the maintainer changes and when the assessment is revised. Explain how a learner can reproduce the exact assessment completed last month without assuming that a link to the current version is enough.

Worked solution

Use nodes Course A, Course B, Prerequisite P, Assessment X, and Maintainer M. Add A requires P, B requires P, P assessed-by X, and M maintains P. Maintain a reverse lookup or explicit inverse relationships if users must find every course depending on P. When responsibility changes, replace the maintenance relationship while keeping P’s identity stable and retaining appropriate history. Revising the assessment should create a distinguishable version or immutable artifact if old attempts must remain interpretable. A completion record can reference X’s specific version along with its outcome and date. Linking only to the newest mutable X would silently change the meaning of the old record. The graph also needs an authorization policy, unavailable-resource behavior, and clear ownership of stale links. A stable identifier helps continuity but does not guarantee that content remains hosted or trustworthy.

Test the limits and transfer the idea

1. Graph algorithm and counterexample · allow 40 minutes

Follow impact through a changing information graph

The proposal contrasts a web of relationships with a single tree. Model an original laboratory using typed directed edges: P uses S; Q uses S; S depends-on D; D cites R; R cites D; E maintains S. You need to find projects affected if D changes. Only uses and depends-on edges express operational dependency. A citation by itself is not a dependency.

Trace a reverse search from D over dependency edges and give the affected projects. Show what goes wrong if every hyperlink is treated as operational dependency, including the citation cycle. Specify the additional index needed when documents publish only outward links. Then describe how a missing or stale source record limits the completeness of the result.

Get a starting hint

The graph traversal rule is part of the query. Link existence alone does not define impact.

Compare your reasoning with the worked solution

Reverse dependency search begins at D. The incoming operational edge S depends-on D reaches S. Incoming uses edges P→S and Q→S then reach P and Q. The affected project set is {P,Q}. E's maintenance relation can be separately queried to find a contact, but it does not make E a project dependency. The citation edges D↔R are excluded from this impact query.

If every link is treated as dependency, a traversal can incorrectly classify R as operationally affected and follow the D,R cycle forever unless it records visited nodes. A visited set prevents nontermination but does not repair the semantic error: the traversal still follows relationships that do not mean dependency. Both edge typing and cycle handling are needed.

Outward links alone tell a publisher what it references. Finding all incoming dependency links requires a maintained reverse index, explicit inverse edges, or a scan of the relevant published records. That index should retain source identity, link type, and version or retrieval time so a result can expose its provenance. When a source is unavailable or a link record is stale, the returned set is complete only relative to the indexed snapshot, not necessarily the current laboratory. Report that scope rather than claiming that no unlisted project is affected.

This is an original typed-graph exercise motivated by the proposal. It does not imply that ordinary Web hyperlinks enforce dependency semantics, automatically provide backlinks, or prevent broken links. The organizational problem requires naming and ownership conventions in addition to navigation.

Check your reasoning

• Finds P and Q with the correct reverse typed-edge trace.
• Separates semantic overreach from cycle termination.
• Specifies backlink indexing and bounds completeness by available evidence.

2. Data model and consistency · allow 45 minutes

Preserve an experiment while a document evolves

A laboratory report links to /procedure/current. On Monday that resource contains version V1, specifying a 5-minute calibration. Experiment X follows V1. On Friday the maintainer replaces the same resource with V2, specifying 8 minutes. A second department has copied V1 into its own wiki without provenance. Next month a reviewer asks which procedure X actually followed.

Design the minimum records and publication relationships needed to answer reliably without preventing the current procedure from evolving. Trace a correction to V1 that must be visible to future readers while preserving the exact original instructions. Explain the difference between a stable identifier, immutable content, availability, and authority. Include a failure case in which a content hash alone cannot answer the historical question.

Get a starting hint

One identifier can name an evolving work, while another identifies the specific version used. A correction relationship need not overwrite the old evidence.

Compare your reasoning with the worked solution

Give the evolving procedure a stable work identifier, and give V1 and V2 distinct version identifiers with preserved content, publication times, and responsible maintainers. The current-work relationship can point to V2 while experiment X's record references V1 specifically, along with its execution date and relevant deviations. The second department's copy should identify its source version and copy time or be marked as unverified when that provenance cannot be reconstructed.

If V1 contains an error, publish a correction record linked to V1 and, where appropriate, a corrected version V1c. Keep the original V1 available with a prominent correction notice and links to the new guidance. X's historical record still says what was used; a separate assessment records whether the correction affects X's validity. Quietly replacing V1 with amended text would falsify the reconstruction even if it improves current instructions.

A stable identifier preserves a name across changes, but it does not imply fixed bytes. An immutable version fixes content, but its host can still disappear. Availability requires preservation and retrieval arrangements. Authority concerns who is entitled or competent to issue the procedure, and is not established by either a URL or a hash. A hash can verify that retrieved bytes match recorded bytes, but if no copy survives it cannot reconstruct those bytes. Nor does an unexplained hash prove that X actually used that version rather than another. Link the artifact to an accountable experiment record. These are explicit modern preservation extensions to the proposal's changing-information problem, not mechanisms already fully specified in the 1989 text.

Check your reasoning

• Separates evolving work identity from immutable version evidence.
• Traces a visible correction without rewriting experiment history.
• Distinguishes identity, integrity, availability, and authority with a hash counterexample.

Check your understanding

Two project branches duplicate the same sensor record. What new obligation appears?

• Both projects become automatically isolated.
• All references become immutable.
• Updates must keep the copies consistent or designate an authority.
• No relationship can cross the branches.

Explanation

Duplication creates a consistency problem that a shared identity can reduce.

A record must reproduce the procedure used in an old experiment. What should it reference?

• A specific preserved procedure version.
• Only the current homepage.
• Only the current maintainer’s name.
• Whichever search result ranks first later.

Explanation

Historical reproducibility requires identifying the actual content used, not only a mutable current resource.

Original paper and sources

Tim Berners-Lee. Information Management: A Proposal. CERN proposal, March 1989; recirculated May 1990.

Read the original paper →

Read the primary paper: World Wide Web Consortium

• Information Management: A Proposal: World Wide Web Consortium