New independent field guide, published 17 October 2026. Examples are illustrative. The historical directory records linked below are separate source context, not current recommendations or earlier versions of this essay.
A link has at least two reader-facing components: the words that invite a click and the address the browser follows. They can agree, differ innocently or describe entirely different things. A directory editor needs to understand both without treating a neat label as evidence that the target is correct. Maintenance becomes especially delicate when old addresses contain queries, fragments, changed hostnames or encoded characters. This guide uses invented example-domain addresses to explain the distinctions and build a cautious editing workflow. It does not ask readers to visit or trust any historical listing preserved in the separate source catalogue.
Parse the address into meaningful components
RFC 3986 describes the generic components of a URI, including scheme, authority, path, query and fragment. Those parts perform different jobs. In an invented address such as https://example.org/maps/view?region=north#legend, the path and query can help select a representation while the fragment identifies a secondary location within it. The visible label might simply say northern maps. None of those strings alone tells the whole story. A maintenance record should preserve the original address before any normalisation so that a later editor can reconstruct what was actually changed.
The hostname is not every word that looks like a name inside an address. A familiar organisation name placed in a path does not make that organisation the host. Equally, a subdomain can be part of a legitimate structure without representing a separate organisation. Read the parsed components rather than guessing from the longest recognisable phrase. For an ordinary directory, the practical question is whether the chosen address identifies the intended resource under the expected publisher context. That requires content and identity checks beyond the fact that a string is syntactically a valid URL.
Resolve relative links against the correct base
A relative address depends on context. The link guide.html from one directory can resolve differently from the same text used in another. A leading slash refers to the site's root rather than the current path, while a fragment-only link stays within the current document. When importing old metadata, retain the source page address used as the base. Otherwise a plausible-looking relative link can be attached to the restoration site's own host by accident, sending readers somewhere unrelated to the captured destination. Correct resolution is a prerequisite for deciding whether the target itself is appropriate.
Consider a fictional source page at https://example.org/library/index.html containing ../maps/. Resolving it gives a path at the parent level, not a child of library. If an importer merely prefixes the current page's directory, it can create a nonexistent location. Another complication is an explicit base element in the original HTML, which can change relative resolution. A reliable import process inspects that context rather than assuming every link behaves like a hand-written root-relative path. Save both the recorded form and a resolved interpretation when the distinction is useful for an audit.
Do not erase query meaning while tidying
Some query parameters carry tracking information; others select the actual document, language, search or record. Their appearance alone does not establish which kind they are. Removing everything after a question mark can turn a specific reference into a generic homepage or a different resource. In an invented directory, record?id=42 and record?id=43 may be different entries. A cleanup that treats them as duplicates because their paths match loses identity information. Normalisation should follow an understood rule with a reversible record, not a preference for shorter strings at any cost.
Parameter order and repeated keys also require care. Some applications treat alternative orders as equivalent; others interpret repeated values or signatures in specific ways. Do not claim universal equivalence merely because a particular site tolerated a rearrangement during one test. For maintenance, compare actual behaviour where appropriate and document the chosen canonical target. Historical query results can remain separate archive records even when they overlap category pages. This edition preserves that query context in metadata rather than pretending the old search system is still running or that its results describe the present web.
Encoding is not an invitation to rewrite the name
Percent-encoding represents bytes within URL components. Decoding and re-encoding without respecting component boundaries can change meaning, especially when reserved characters are involved. A slash encoded as data is not automatically the same as a slash used as a path separator. Similarly, the case of a path can matter even though hostnames are handled differently. A directory editor does not need to memorise every parsing rule, but should know that making two addresses look more alike is not proof that they identify the same resource.
Use a standards-aware URL parser for component inspection while retaining the original captured string. If a historical page contains damaged character decoding, do not silently invent the missing name from context. Keep original bytes where available and mark the limitation in any public metadata affected by it. This is a concrete preservation issue rather than an aesthetic preference. An apparently helpful spelling repair can detach a record from its evidence. A new explanatory label may be supplied as editorial text, but it should not be presented as the exact original string when the source does not support it.
Choose labels that explain the intended action
After identifying the target, write a label that helps the reader anticipate it. A document title, resource name or clearly described action is usually more informative than a repeated click here. W3C's link-purpose guidance explains why meaningful names and programmatically associated context matter for navigation. The destination type can also be useful: an archive capture, downloadable document and current publisher page are different kinds of next step. A directory should not label an old capture as the live service, even if both are connected to the same historical resource.
Avoid making the label carry an unsupported quality judgement. Official, complete, secure and best are claims, not decorative words. Use them only when their meaning and evidence are established in the collection's scope. An invented entry can say publisher's collection index if that is what was observed, without implying that every item is exhaustive or current. Align the link's accessible name with its visible purpose so that keyboard and screen-reader users receive the same useful distinction. A technically correct address can still be a poor link when the invitation tells the reader to expect something else.
Keep an edit trail from old address to chosen target
A practical maintenance event records the old address, test date, observed response, final target, content judgement and proposed action. Do not replace the stored address before the judgement is complete. A redirect can lead to an equivalent moved document, a generic homepage, a login screen or an unrelated new site. The status and final URL explain only part of that journey. The editor must decide whether the resource promised by the label is still there. Keeping the old and new forms makes a mistaken update reversible without relying on someone's memory.
The historical source catalogue linked below displays old listing addresses as inert metadata, not as current recommendations. Its role is to preserve what was recorded without inviting a reader to treat an unreviewed destination as an active service. The new editing workflow applies a simple sequence: preserve the original, parse it, understand the target, write an honest label and document the decision. That sequence avoids two common mistakes at once: trusting a label without reading the destination, and tidying an address so aggressively that the intended resource disappears from the record.
References & further reading
From the recovered archive
- Cufflinks, Inc. — record 1160 · captured 29 November 2010
- Latest-links index · captured 29 November 2010