Skip to content

Commit 5896343

Browse files
stroland02claude
andcommitted
docs: refresh what B7's cheap check actually finds today
The entry claimed eight nodes. The graph builds ten, and four of them postdate the last acceptance run -- which strengthens rather than weakens the case the entry makes, since every one of them sits on the path nothing has exercised. Also records that build_graph refuses a store lacking record_migration_outcome. That check is why three of the five quality axes read zero rather than reading wrong. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
1 parent bc1afdb commit 5896343

1 file changed

Lines changed: 10 additions & 4 deletions

File tree

docs/superpowers/BACKLOG.md

Lines changed: 10 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -126,10 +126,16 @@ range, branch deletion on abandonment, checkpoint serialiser registration, the
126126
dependency-edit guard, staged-new-file support, and dependency-tree discarding. Every one of
127127
those sits on the acceptance path.
128128

129-
Checked cheaply and it is not obviously broken: the test still collects, and the production
130-
graph compiles with the real `StripeAdapter`, `TypeScriptAdapter`, `TieredRemediator`,
131-
`GitHubForge` and store, exposing all eight nodes. That establishes the wiring survived. It
132-
establishes nothing about behaviour.
129+
Re-checked cheaply on 2026-07-30 at `bc1afdb` and it is still not obviously broken: the test
130+
collects, and the production graph compiles with the real `StripeAdapter`, `TypeScriptAdapter`,
131+
`TieredRemediator`, `GitHubForge` and a store. It now exposes **ten** nodes rather than the eight
132+
this entry used to claim — `locate`, `prepare`, `patch`, `static_verify`, `push_branch`,
133+
`await_ci`, `replay`, `open_pr`, `report`, `abandon`. Four of those postdate the last acceptance
134+
run, which is the point: the wiring survived, and that establishes nothing about behaviour.
135+
136+
`build_graph` also refuses a store that cannot record a migration outcome, naming the missing
137+
`record_migration_outcome(outcome)` and calling it the single write every benchmark axis reads
138+
from. So the corpus wiring is checked at construction rather than at the end of a run.
133139

134140
**Run it with `-n0`.** `addopts` now carries `-n auto`, which applies to the e2e test too.
135141

0 commit comments

Comments
 (0)