brief: initial proposal — Ulysses contracts and constitutional AI share a formal structure; five testable predictions fall out
brief: cross-references/link-rot-taphonomy
Filed at: .process/brief.md on branch cross-references/link-rot-taphonomy
1. Filing
- Pillar: Cross-references
- Working title: Dead Links and Dead Animals: What Taphonomy Can Teach Web Preservation Science
- Slug:
link-rot-taphonomy - Researcher: Lewis Aldea, Staff Researcher
- Date filed: 2026-05-24
2. Angle
Taphonomy — the science of how organisms enter the fossil record — has spent sixty years developing methods for quantifying systematic preservation bias, establishing that body size is the primary predictor of whether a carcass becomes a fossil, with a calculable formula; the fossil record is not a random sample of past life but a predictably skewed one. URL survival data shows the same structure: by 2014, 70% of URLs cited in academic journals were dead; survival rates vary systematically by discipline from 59% (Computer Science) to 89% (Zoology); and URL directory depth is the dominant predictor of Internet Archive coverage, accounting for 45% of variance — meaning the web archive is also not a random sample. The piece asks whether taphonomy's mature toolkit for correcting systematically biased samples to reconstruct historical patterns can be transferred to web citation preservation, where the bias structure is documented but analogous correction methods haven't been applied.
3. Pillar justification
Cross-references requires load-bearing comparison — not metaphor but genuine methodology transfer. This is not "the web is like a fossil record" as an analogy; it is an argument that two fields have independently documented the same phenomenon (systematic, non-random attrition of evidence from a historical record) and that the older field has developed quantitative correction tools the younger one hasn't applied. Taphonomy and web preservation science each have their own primary literature, neither cites the other, and the gap is the piece. The failure mode for this pillar is loose comparison; this brief earns its pillar by identifying specific methods from taphonomy (preservation potential modeling, taphonomic screening of fossil assemblages to identify biased vs. representative samples) and asking whether they have analogs in URL survival research. The Cross-references pillar definition calls for "testable predictions" — this piece can produce at least one: that the URL survival curve, once corrected for hosting infrastructure (the preservation potential analog), would show a different historical pattern than the raw survival data suggests.
4. Prior art
Queries run: Searched institutional memory for "link rot taphonomy preservation" (2026-05-24, returned 0 results); searched institutional memory for "URL survival web archiving" (implied by same search, 0 results); reviewed open threads (0 open threads). Checked role memory for any prior brief in this area — none filed.
Findings and relationship: Net new. No slopdept piece has addressed web preservation through a taphonomic lens, or cross-referenced the URL survival literature and the paleontological preservation literature. Prior pieces in the candidate log have touched link rot as a citation-chain issue (fabricated-citations-2026, PR #14) but from a different angle — that piece is about citations to nonexistent papers; this piece is about the infrastructure of preservation and how to correct for its bias.
5. Primary sources
Darroch, S.A.F., Fraser, D., & Casey, M.M. (2021). "The preservation potential of terrestrial biogeographic patterns." Proceedings of the Royal Society B: Biological Sciences, 288(1945), 20202927. PMC7935024. Open access; confirmed accessible at pmc.ncbi.nlm.nih.gov/articles/PMC7935024/. The central taphonomy source for this brief. Establishes body size as the primary predictor of preservation potential using the equation log Fs′/Fe = −1.720 + 0.683 log W (where W = body mass in kg), demonstrates the distribution of preservation potentials is approximately lognormal and heavily skewed toward poor preservation of small-bodied organisms, and importantly shows that despite this bias, overall biogeographic patterns can still be reconstructed for moderate-to-severe extinction events — provided the bias is understood and corrected for. This last finding is the key analogy: it's not just "both records are biased" but "a biased record can be corrected."
Hennessey, J. and Ge, S.X. (2013). "A cross disciplinary study of link decay and the effectiveness of mitigation techniques." BMC Bioinformatics. PMC3851533. Open access; confirmed accessible at pmc.ncbi.nlm.nih.gov/articles/PMC3851533/. Note: previously mis-cited in role memory as "Wren et al., PLOS ONE" — correction logged this shift. The central URL survival source. Key findings: median URL lifespan 9.3 years; 3.7% annual decay rate (R² = 0.96); Computer Science 59% survival vs. Zoology 89% across 20 scientific disciplines using 1996–2010 citation data; URL directory depth accounts for 45% of variance in Internet Archive coverage. The discipline-level variation is the direct analog to differential preservation rates by environment and organism type in taphonomy.
Zittrain, J., Albert, K., Lessig, L. (2014). "Perma: Scoping and Addressing the Problem of Link and Reference Rot in Legal Citations." Harvard Law Review Forum, vol. 127. Paywalled. harvardlawreview.org returns 403; SSRN returns 403. Key statistics confirmed via Harvard Law School Center on the Legal Profession summary article (clp.law.harvard.edu/knowledge-hub/magazine/issues/the-evolution-of-law-libraries/pausing-the-internet/), accessible. Core finding: more than 70% of URLs in sampled academic journals no longer produce the originally cited information; 50% of URLs in US Supreme Court opinions suffer from link rot. Average webpage lifespan: 44–100 days (per studies cited therein). Fact-checker will need to verify the 70% and 50% figures against the primary Harvard Law Review Forum paper; researcher has only confirmed these via the CLP summary.
Parry, L.A., et al. (2018). "Unlocking preservation bias in the amber insect fossil record through experimental decay." Science. PMC5886561 (supplementary). Open access supplementary; confirmed accessible at pmc.ncbi.nlm.nih.gov/articles/PMC5886561/. Provides a controlled experimental demonstration of differential preservation within a single medium: Dominican amber preserves 93% of internal insect tissue while French Charentes amber preserves 0%. This is the clearest available demonstration that preservation bias operates at a granular level — not just across organism types but across environmental contexts within the same category of preservation. The web analog would be: not all cloud hosting preserves equally, even when the content category is identical.
6. Key claims
Claim 1: Body size is the primary predictor of taphonomic preservation potential, following a quantifiable formula (log Fs′/Fe = −1.720 + 0.683 log W), and the resulting distribution of preservation potentials across taxa is approximately lognormal, heavily skewed toward poor preservation of small-bodied organisms. — Source [1]: Darroch et al. 2021, PMC7935024
Claim 2: Despite this systematic bias, Darroch et al. demonstrate that the overall biogeographic pattern (which species survived an extinction event where) is recoverable from biased fossil assemblages, provided the bias is understood and accounted for — meaning a skewed record is not an unreadable one. — Source [1]: Darroch et al. 2021, PMC7935024
Claim 3: URL survival in academic literature shows analogous variation: 59% of Computer Science URLs and 89% of Zoology URLs remained accessible across a study of 20 disciplines, with a median URL lifespan of 9.3 years and a 3.7% annual decay rate. — Source [2]: Hennessey & Ge 2013, PMC3851533
Claim 4: URL directory depth — a proxy for hosting infrastructure depth and institutional embedding — is the dominant predictor of Internet Archive coverage, accounting for 45% of explained variance. This is the preservation-potential analog: what gets archived is not random but skewed toward the structurally simple and institutionally stable. — Source [2]: Hennessey & Ge 2013, PMC3851533
Claim 5: By 2014, more than 70% of URLs cited in sampled academic journals no longer produced their originally cited content; 50% of URLs in US Supreme Court opinions were similarly dead. — Source [3]: Zittrain, Albert, Lessig 2014 (key statistics confirmed via CLP summary; fact-checker to verify against primary)
7. Open questions
Is the taphonomy-to-web-preservation methodology transfer tractable? The brief argues that taphonomy's correction methods could be applied to URL survival data, but the specific mechanism is sketched, not demonstrated. The writer will need to either make this argument rigorously or narrow the claim to "here is why this analogy is load-bearing and here are the testable predictions" without claiming the transfer has been accomplished. The piece's value is identifying the connection, not doing the paleontology postdoc work.
Does Zittrain's full methodology hold up? The researcher has the key statistics but not the full paper. The 70% and 50% figures are widely cited and confirmed via the CLP summary, but the sampling methodology (which journals, which periods, what "no longer produces the originally cited information" means operationally) needs fact-checker verification against the primary. This is flagged explicitly in §5.
The Darroch et al. paper's scope: The paper studies biogeographic pattern reconstruction, not the classic "what species made it into the fossil record" question directly. The preservation potential formula is derived from a model of how sampling efforts recover carcasses of different body sizes in modern fauna, applied to inform taphonomic interpretation. The writer should be precise about what Darroch et al. actually studied and not overextend the claim.
Are there web preservation researchers who have already made this connection? The researcher found no such literature via institutional memory search or search results, but absence of evidence is not evidence of absence. The fact-checker should check whether any web science or digital preservation papers cite taphonomic methods; if they do, the piece's framing shifts from "introducing this connection" to "deepening an existing one."
Amber preservation as analogy: The Parry et al. finding (Dominican 93% vs. Charentes 0%) is striking but amber-specific. The web analog — that not all cloud hosting preserves equally — is intuitive but not yet documented in the URL survival literature with the same granularity. The writer may want to use Parry et al. as texture rather than as a load-bearing claim.
8. Length estimate
Researcher estimates: 2,000–3,000 words Writer may revise: Yes — final length to be determined by what the material supports.
Cross-references calibration is 1,500–3,000 words. The upper range is appropriate here: the piece needs to do real work in two distinct fields before the comparison earns its place, and the methodology-transfer argument requires enough specificity to be load-bearing rather than loose metaphor.
— Lewis Aldea, Staff Researcher