Workshop: when do link labels supply missing premises?
Question and commission
When does descriptive link text supply a premise needed to reconstruct a note's claim, rather than repeat its body or help readers navigate elsewhere?
Posed: 2026-11-08, by the operator during the unresolved-work reconsideration. The operator commissioned a new workshop to ask the right question, design the experiment carefully, retain its evidence, and only then write supported claims. The question above is the initial formulation; sharpen it before choosing cases. The purpose is to learn what link text contributes, not to rescue an existing note or justify a predetermined review gate.
Starting point, not established conclusions
The deleted note, “Descriptive link labels may supply the self-sufficiency a
reconstruction gate would check,” is retained as an unverified starting
record from revision
7c3fa9b8d140812a5d56b8af99452169cc797a21. It reports five initial reconstruction
PASS verdicts and a later comparison of three mature notes with and without descriptive labels; all six later verdicts
were PASS. These are reported results, not yet independently recovered evidence
for this workshop.
The note mixes a label-based explanation, an experimental account, and advice against adding a Commonplace review gate. Its test does not by itself establish that the body-writing convention caused the result, that reconstruction gates are redundant, or that mature notes generally do not need labels. A reader that fails to detect a missing premise could also produce an all-PASS result.
The writing-as-thinking workshop previously carried the thin-draft test. This workshop now owns this investigation; its other process-transfer work remains separate. The frontier scan records how the candidate was selected. The subsequent reassessment found that it needs experimental design rather than immediate execution of that shortlist recommendation.
Evidence publication policy
Raw captures in evidence/ are excluded from validation and site publication by
.commonplace-validation-ignore. Stored links preserve their source spelling,
not live navigation. This framing and any live conclusions remain published;
the relative evidence link above supports checkout use.
Evaluation boundary
Keep three questions separate:
- What information do bodies and descriptive link text each supply?
- Can the chosen reader and assessment method detect a missing necessary premise, rather than accept an unsupported reconstruction?
- Would a particular review check add enough useful detection to justify its operating cost? This is a later operational question, not a conclusion from label ablation alone.
Distinguish reconstruction from truth and inference validity. A reader can recover an argument whose premises are false. A fluent reconstruction can also silently import a premise absent from the supplied text. The protocol must say what the reader is allowed to assume and how assessors identify that import.
Do not treat the old all-PASS results as calibration. Before relying on a new comparison, demonstrate that its assessment can distinguish a missing-premise case from an adequately supplied case. Keep answers and expected labels out of the reader's context. Resolve ambiguous cases explicitly rather than forcing a binary verdict.
Separate label changes from body enrichment. Distinguish label text at inline links from footer glosses if the intervention changes both. An informative label may add information, while removing a label may also remove ordinary sentence content; record exactly what the comparison changes. Constructed cases can test a mechanism but cannot establish how often the problem occurs in real drafts.
Next work and evidence retention
First recover the earlier prompts, input versions and outputs where available; record missing material without recreating it as historical evidence. Then write a bounded protocol that states the claim at risk, contrasts, case-selection basis, controls, assessment method, isolation requirements, run budget and what each possible result would license. Choose the actual cases, method and sample size from that question; this framing does not fix them.
Freeze the protocol and input identities before confirmatory execution. Keep exploratory changes and their reasons distinguishable from that test. If controls fail or a consequential case is unresolved, repair the instrument or narrow the question before interpreting the target comparison.
Retain exact supplied texts, prompts, model/runtime settings, reader access boundaries, outputs, assessments and deviations. Record negative and inconclusive results. Specify the durable evidence destination before writing library claims, so the claims can cite inspectable evidence rather than an account in a workshop that will disappear. Follow the destination collection and type contracts.
No new experiment has run. The operator approved deleting the old library note on 2026-11-08 without a public redirect. Its text remains an unverified starting record above. Any replacement claim must follow the evidence; keeping a standalone theory note is not required. No gate adoption or review-system change is part of this investigation by default.
Closure
Close when the question has a bounded answer or an explicit inconclusive result, the evidence and its limits have a durable home, any warranted claims have been written without local operating detail, and the old note's approved deletion has been completed. Update the predecessor workshop's pointer, extract the results, and close under the workshop collection contract. A negative result or no new theory note is a valid outcome.