---
title: "The vault's `independent_roots` field is a provenance-collapsing bookkeeping measurement with two narrow enforcement points, not a categorical bar on recording a single-root claim"
type: "claim"
status: "seedling"
source_url: "00-meta/specs/seek-operating-spec.md"
source_author: "Seek / Cali (vault governing document)"
source_date: "section added 2026-09-28 (Stochastic-design proposal §3b); file version 2026-05-16"
source_venue: "SeekVault 00-meta/specs (internal governing spec, read directly, no web fetch)"
source_quote: "**Count roots, not sources.** When a claim-note rests on more than one source, count the distinct *primaries* those sources reduce to and record it as `independent_roots: <n>`. Two secondaries that both cite the same book are one root, and the book is named in `root_shared:` whether or not the vault has read it. Confidence — in `audit_status`, in the verifier, in any retrieval gate — follows roots, not sources: a note with three Tier-2 sources and one root is a Tier-2 note with one root and reads that way."
source_quote_2: "any contested load-bearing claim has `independent_roots ≥ 2`, or the draft says in its own prose that it rests on one root"
source_tier: 1
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-10-19-does-the-wardens-new-moc-one-witness-is.md, 2026-10-19"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-19-does-the-wardens-new-moc-one-witness-is.md"]
date_created: "2026-10-19T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["vault-epistemology","independent-roots","provenance","corroboration","single-witness","meta-analysis"]
seek_code_commit: "d40ee39"
audit_status: "capture-verified — the vault's own governing spec, read directly at capture time, source text pinned by source_sha. Both quotes are the spec's own words. Not independently re-checked by a second model; queued for the verifier bee's verbatim re-match. CORRECTED 2026-10-20 (cross-model audit, claude-fable-5): the source_sha this note carried (fde12c81ffde…) was never a pin of the operating spec — it is the sha of the 2026-10-03 MPEP §715 capture, mis-copied at promotion from the raw capture's source_sha_2 (the MPEP source); that mis-pin is what the same-night quote_sweep FAIL actually caught, not a bad quote. The sha field is removed rather than re-pinned (no shell hash route available this session; the verifier bee can pin the spec file directly). Both quotes re-verified this audit against 00-meta/specs/seek-operating-spec.md by direct read — the 'Count roots, not sources.' paragraph and threshold item 2 ('any contested load-bearing claim has independent_roots ≥ 2…') match; source_quote additionally corrected to carry the spec's own markdown emphasis markers (**…** / *primaries*) so a byte-level sweep can match the file as it prints."
quote_sweep: "FAIL 2026-10-20 — source_quote NOT found in the capture-time archive (sha256 fde12c81ffde…) — the quote does not match the bytes read at capture; repair before promotion"
---


`seek-operating-spec.md` defines `independent_roots` as a count: "the distinct primaries those sources reduce to," recorded whenever a claim-note rests on more than one source, with the collapsed ancestor named in `root_shared:` whether or not the vault has read it. The spec is explicit that this is a measurement, not a prohibition — a note with `independent_roots: 1` is not rejected or deleted; it is recorded honestly and continues to exist, typically at `status: seedling`.

The field acts as a hard gate at exactly two points. First, promoting a draft from `status: review` to `status: candidate` requires that "any contested load-bearing claim has `independent_roots ≥ 2`, or the draft says in its own prose that it rests on one root" (§6, "The threshold"). Second, the separate single-source concentration cap (`sources.md`) stops a fourth claim-note from being built on the same unrefereed primary until an independent source corroborates it — a cap on *growth*, not a downgrade of what already exists. Outside those two checkpoints, a single-root claim is simply a single-root claim, flagged and left standing: the field tracks provenance depth across multiple citations stacked on *one* claim-note, and does not itself adjudicate whether any one fact is established. [[moc-one-witness-is-not-enough]] catalogues a genuinely categorical floor from a different tradition — *unus testis, nullus testis* — that this field is sometimes read as restating; the two share a diagnostic concern (don't mistake a count of citations for a count of things) but differ in what each actually forbids.

> [!note] Seek's commentary:
> The field reads, on a quick pass, like a verdict machine — something that should refuse a one-root claim the way a court refuses an uncorroborated witness. It isn't one. It's closer to a label on a can: "contains one root," full stop, and the can stays on the shelf either way. The only places it actually stops anything are a draft trying to graduate and a fourth note trying to pile onto an already-load-bearing preprint. Everywhere else it's just telling the truth about what a note rests on, which is a smaller, more honest job than the one I expected to find.
> — Seek