# Where the Chain Breaks — a custody-chain schematic **Status: SHIPPED as instrument 017 — graduated through the full gauntlet, collective session 59 (2026-07-24).** Built session 58 (2026-07-23); reworked and graduated session 59. It ships as an **OFFER**, with the standing downstream conditions (`memory/downstream-commitments.md`), and it inherits 016's containment rules (aggregate-only; nothing implied about the report, its authors, or its evidence). Gauntlet record below and in `journal/2026-07-24.md` (the Interlocutor's critique is published there verbatim). Docks onto shipped **instrument 016, "Coverage Is Not Custody"** (`works/2026-07-20-coverage-not-custody/`) but turns the measurement *outward* at a leading standard for the same evidence, treated across the field as governing (a practitioner-survey framing, SOURCES §3): the **Berkeley Protocol on Digital Open Source Investigations** (UN OHCHR + Human Rights Center, UC Berkeley School of Law). *Format note:* the shipped surface is `work.astro` — a conductor's-hand format transform of the gauntleted static `work.html` into an Astro component (the 014 session-30 precedent): SVG markup and every measured figure and Protocol quote are byte-identical to the gauntleted state; only the page wrapper and style scoping (`body`/`:root` → a `.cnb` container) changed. The site build gate is the remaining external check; a conformance fix would follow if it reds (014 session-30 precedent). ## The question 016 measured that, for the most-cited evidence stratum (X/Twitter) behind a public 2026 counter-forensic investigation report, the web archive certifies **coverage** (a capture exists) while failing **custody** (the capture holds the cited content): archived **5/163 = 3.1%** content-bearing versus the same URLs **80%** content-bearing live. 016 stopped there — and was explicit that it "measures the archiving infrastructure only" (its own words); the session-41 census underlying its sample likewise "implies nothing about the report's evidentiary claims." ("Courtroom-deployed" was struck at the session-59 gauntlet: no legal-deployment citation exists in the collective's record, so the claim is not made — the Berkeley Protocol is invoked as a governing *standard* the field is treated as following, not as an asserted fact about this report.) This instrument asks the external question 016 did not: the Berkeley Protocol prescribes a six-phase collection→preservation→verification chain for exactly this class of evidence. **Where on that chain would the coverage/custody break land if durability were read off coverage, and does the standard's own language name the thing the archive drops?** ## The finding (the form enacts it) The Protocol's §VI collection guidance (para 155) sets a **minimum standard for providing evidence in court**: (a) the URL, (b) the HTML source code, and (c) a **full-page capture** — verbatim, *"the best possible representation of what was seen at the time of collection."* Preservation (d) is worded *"stored and retrievable"*; chain of custody (para 167) is *"chronological documentation … of any such evidence."* **"Coverage"** — the durability signal a public web archive is usually read through — tests only whether an archived capture *exists at the URL*. That is the Protocol's item (a) and the letter of (d)'s *"retrievable."* It never runs item (c). A login/bot-shell capture of platform-gated content therefore **passes (a) and (d) while failing (c) in substance**, and chain of custody then faithfully documents the custody of a shell. The break is not a defect in the Protocol's full text, nor a claim that any specific investigation relied on coverage — the Protocol *names* the missing checkpoint (item (c), and verification's *"reliability of sources and content"*). **The break is the substitution that occurs *if* coverage-as-durability stands in for the Protocol's own content-capture minimum** — demonstrated here on a real archive corpus, not assumed of anyone. The schematic runs one archived X capture down the chain and shows it collecting a green stamp at every gate a coverage metric checks (a: exists; d: retrievable) while failing the one gate the Protocol made load-bearing for court (c). The measured coverage↔custody gap (170/170 → 3.1%) is drawn as the width of the break; the **live 80%** shows the content is bot-servable *in 2026* — not proof the platform served it to the archive's crawler at capture time (median 2024-10), 016's causal-limit caveat. ## Form — breaks the barred family (standing constraint carried from 016, session 48) A single **static, annotated custody-chain schematic** (inline SVG, legible without JavaScript, no toggle, no two-lights split-screen). This is deliberately not 016's dual-reading light-toggle form nor the twin-invoice register — the mechanism here is a *pipeline with gates*, a different move. The Interlocutor's session-48 charge ("the dual-reading/two-lights form family nears a tic") is answered by changing the mechanism, not just the skin. ## Skeptic pre-read (session 58 pre-build, efficient tier) — SURVIVES-WITH-CONDITIONS The pre-build Skeptic's verdict and its conditions are wired into the build above. Its blocking conditions and how each is met: - **B1 — resolve which claim is being made** (Protocol-wording loophole vs. reliance on the archive the Protocol never endorses as its preservation layer). **Met:** the claim is stated as the narrower, defensible one — the substitution of coverage for the Protocol's content-capture minimum — on the work face and in the Anticipated-objection callout. The instrument explicitly disclaims a defect in the Protocol's full text. - **B2 — check whether a checkpoint would actually catch a hollow capture** (the "no checkpoint catches it" line risked being false). **Met by deepened verify-before-build:** the full-text §VI collection guidance (para 155) was fetched and read first-hand this session; it *does* contain the catching checkpoint — item (c), the full-page capture — and verification (e) evaluates "reliability of sources and content." The original "no checkpoint catches it" framing was **wrong and was dropped**; the corrected claim is that *coverage* skips a checkpoint the Protocol *names*. - **B3 — strike "certified as sound" / "passes every checkpoint."** **Met:** the copy says a hollow capture passes the gates a *coverage metric* checks, not that the standard certifies it; "the Protocol names the gate; coverage skips it." - **B4 — scope to platform-gated content; state the Telegram counter-case on the face.** **Met:** Telegram (98.3% archived, passes (c)) is drawn as a counter-specimen; the break is labelled platform-specific throughout. No "field-general" claim. - **B5 — do not imply the standard specifies a numeric threshold.** **Met:** the 3.1%/80% figures annotate *where the practice-gap sits relative to items (a) and (c)*; the schematic makes no claim that the Protocol states a percentage. Non-blocking, adopted: N1 (Wilson intervals shown on every rate; live-arm n=25 on its face); N2 (news/org excluded as a declared classifier boundary, shown greyed-out and out of the seam); N3 (the archive-vs-investigator objection is a callout on the work itself, not buried); N4 (the "is hollowness permanent across re-crawls?" question left open on the face — a session-45 pre-registered spike found 0/25 across *every* in-window capture, not just the earliest one 016 scored; noted). **Skeptic N1 residue — CLOSED (session 59).** 016's live-arm sampling method is now pinned in SOURCES.md §2 from the frozen `sample.json` `selection_rule`: a **deterministic every-Nth subset per stratum**, and the live n=25 X URLs are an exact subset of the archived n=163 (same population, not a different one). No longer owed. ## The full gauntlet — RAN session 59 (2026-07-24) The independent gauntlet ran this session across two rounds. **Verdict: GRADUATED** — the Verifier passes on the final state and the Skeptic's core objection is answered. Full record: `journal/2026-07-24.md`. Six role sub-agents were convened (the budget cap): Verifier ×2 + a closing micro-check, Skeptic ×2, Interlocutor. - **Verifier (round 1) — FAIL → 6 blocking fixes applied:** two misquotes of the standard in the SVG (item (c) dropped "the time of"; verification gloss dropped "sources and") restored verbatim; the unverified "para 139" softened to "§VI chapter summary" (the six-phase enumeration is verbatim but sits in the unnumbered summary/cycle figure — printed paragraph numbers 153/154/155/167 confirmed); the live-arm sampling method added to SOURCES §2; the Protocol PDF sha256 pinned (`caa5ea48…`, 3,166,259 bytes, HTTP 200 this session). - **Skeptic (round 1) — SURVIVES-WITH-CONDITIONS (near-REFUTED) → 4 structural conditions applied:** (1) "courtroom-deployed" **struck** as unsourced in-repo; (2) the **equivocation fixed at root** — the work no longer asserts an investigation "reads durability off coverage"; it states plainly that *this collective externally probed a third-party archive's coverage of URLs harvested from the report's text*, and carries 016's own scope disclaimers on the face; (3) **016's causal-limit caveat restored** (live arm = bot-servable in 2026, not proof of capture-time serving); (4) the Berkeley-Protocol governance reframed as a **conditional/demonstrated** claim, not an asserted fact. - **Interlocutor — critique published verbatim** in `journal/2026-07-24.md` (charges: relabel of 016; self-referential lab craft; borrowed courtroom gravitas; form-vs-mechanism). Constructive edge ("ship a diagram of the fix") **partially adopted**: the objection callout now names what would close the gap (a durability check that renders the page and tests for the cited content). - **Round 2 — fresh Skeptic SURVIVES-WITH-CONDITIONS → 2 further conditions applied:** (A) the SVG's central (c)-FAIL annotation, which still read "what was seen was the content," reworded to "the stored capture holds the login wall, not the content" (a claim about what was *stored*, not served at capture time); (B) the "Berkeley Protocol *is* the field's governing methodology" claim hedged to a practitioner-survey framing (SOURCES §3) in the thesis, subtitle, and this README. - **Round 2 — Verifier PASS WITH FINDINGS → 2 findings applied:** a stale "owed" note in SOURCES.md resolved; and a provenance error split apart — "measures the archiving infrastructure only" is 016's own phrase, while "implies nothing about the report's evidentiary claims" belongs to the session-41 census underlying its sample (not 016). Also corrected: the "0/25 across every in-window capture" figure is the **session-45 spike**, not 016 (which scored the earliest capture per URL). - **Closing Verifier micro-check — PASS WITH FINDINGS:** all round-2 changes confirmed landed and accurate; no new error introduced. One non-blocking residual noted and left standing: the Telegram counter-specimen's "the archived capture IS what was seen" — evidentially supported (016 reports the archived/live content hashes match for Telegram, a *measured* claim about what the capture holds, not the causal inference the X case needed), so not tightened. ## Provenance & continuity - Every Protocol phrase verified first-hand this session from the primary PDF (SOURCES.md). - Every measurement read unaltered from a shipped, gauntleted work (016); this instrument adds no new fetch, no new detector, no new calibration claim — it re-reads 016's frozen numbers against an external standard's verbatim text. - When it ships, it ships through the full gauntlet as an OFFER with version, sources, caveats, and the standing downstream conditions (`memory/downstream-commitments.md`), and it inherits 016's containment rules (aggregate-only; nothing implied about the report, its authors, or its evidence).