Batch 2
Run run-full-improvement-pass-on-note.md once per note, one note at a time, in the order listed. Read the workshop README for operating constraints before starting.
Do not start until batch 1 is complete, including any reframe follow-up it triggers. Every note here cites batch 1, so a batch-1 reframe moves these notes' premises before their own pass would read them.
Status: complete. The initial passes were invalidated at batch level because Batch 1 edited cited notes concurrently. After Batch 1 was committed, all four notes were rerun in binding order against that settled result. Every required reframe follow-up is complete.
Notes
| # | Path | Status |
|---|---|---|
| 5 | kb/notes/opposed-recompute-factors-do-not-decide-documentation-segmentation.md |
rerun complete — keep; prior reframe follow-up complete |
| 6 | kb/notes/artifacts-must-preserve-named-choice-scope.md |
rerun complete — keep (reframe); follow-up complete |
| 7 | kb/notes/addressability-grain-sets-a-matched-selective-read-floor.md |
rerun complete — keep |
| 8 | kb/notes/an-insufficient-summary-precedes-the-source-rather-than-replacing.md |
rerun complete — keep |
Order is binding. Two dependency edges fall inside this batch rather than across the boundary: 7 cites 5, and 8 cites 7. Note 8 is last because nothing cites it — it was written after the batches were first drawn and is a leaf.
Inbound dependencies on batch 1
- 5
opposed-recompute-factorscontrasts batch 1's note 1. - 6
artifacts-must-preserve-named-choice-scopecontrasts batch 1's note 2 and cites batch 1's note 3evidenced-by. - 7
addressability-graingrounds on batch 1's note 1, extends batch 1's note 4, and contrasts note 5 in this batch. - 8
an-insufficient-summary-precedes-the-sourceextends batch 1's note 1, contrasts batch 1's note 4, and contrasts note 7 in this batch.
Recheck each edge before its pass: if the target was reframed, the citing sentence here may state a claim its target no longer makes.
Per-note context
5. opposed-recompute-factors-do-not-decide-documentation-segmentation — retitled on 2026-08-23 from
"audience segmentation cannot be decided from one reader property", with a
redirect in properdocs.yml. Carries four consecutive "Consequence:" sections;
one of them, segmentation strips the drift detector from the low-traffic layer,
is a maintenance claim on an economics note and more general than its host.
Split candidate — deciding it before the pass would be legitimate, but deciding
it during is what the pass is for. Cites Commonplace as an existential
witness, so the collection-fit check is a live test of the choice-scoping
requirement here.
6. artifacts-must-preserve-named-choice-scope — two authors. The
initial write was revised by a second agent, which replaced a section and
dropped a ranking of the two repairs in favour of a selection rule. Its
empirical section restates sweep evidence rather than citing it, because the
notes contract forbids outbound links into kb/work/ and the source workshop
has since been deleted; that restatement is deliberate and should survive the
pass. This note also states the rule the pass's own collection-fit check
applies, so a finding against it deserves extra scrutiny either way.
7. addressability-grain — newest of the seven, written last. Its claim is
scoped to the recoverable-cache partition, which is what keeps the "only when"
universal rather than hedged; a pass that widens the scope would break that. Its
central evidence is a pair of instances running opposite ways under one rule, so
a compression finding that trims one instance would remove the discrimination
the note rests on. Cites Commonplace as an existential witness.
8. an-insufficient-summary-precedes-the-source-rather-than-replacing —
written 2026-08-23, after the batches were drawn, to close a dependency: the
workshop's disposition procedure asserted its claim without a note behind it.
Two features to protect. It separates a general ledger from its sharp case —
an insufficient summary can still narrow the source read, so the strict
universal is precedence, and the cost verdict follows conditionally where
nothing is narrowed. A compression finding that collapsed that distinction would
restore the overclaim it was written to avoid. And its ## Scope declares a
modality split: universal per (summary, source, question, required reliability)
tuple, statistical for the aggregate judgment about whether a layer is worth
maintaining. Under ADR 066 both landings carry their own guards.
Its independence from note 7 is developed in both directions and is the reason both notes exist rather than one. If a pass on 7 or 8 weakens that demonstration, check the other before accepting it.