/renew → resumed mid-epic (lf4b 4/11). Orchestrated five
epic-orchestrator runs end-to-end (lead never implements; pi subagents
named persisting sessions, oracle-gated verify): lf4b.5 worked example
(pushed 17cbeb9..33624fb), lf4b.6 guarantee-ladder (33624fb..1a0e91b),
lf4b.7 cold-reader glossary (358334e..5ea861d), lf4b.8
when-to-use-command notes (…1dd22a8), lf4b.9 tool-relationships
(…2e14e9a), lf4b.10 docs-accuracy gate (…1f0df22), lf4b.11 stamped
artifacts (…a011dd5). Epic specodelic-lf4b closed
11/11; all four external reviews' actionable items landed.
Fixed the pretender gate first (user-authorized exclusion
disposition, specodelic-m1rk closed): root cause = d864425's
./-anchoring lesson applied incompletely — vendor/** lacked its
./vendor/** twin, vendored mermaid.min.js leaked 104 findings; fix
a59448e (2 lines), just test FULLY GREEN at HEAD for the first time
since vendor landed. Known-red disposition retired from future
verifies.
Filed upstream pretender-094 (pretender repo, embedded-dolt — .beads
gitignored there, no export commit): honor .gitignore by default in walk
scans so consumer exclude lists stay minimal; includes specodelic's
11-pattern list as motivating evidence + acceptance criteria.
lf4b.10 spawn lesson: first spawn did 50min of good research
(genesis 0.11 Output::emit — JSON envelope = stable comparison surface)
then died in a tool-call abort, zero commits. Re-spawned with the
conclusion seeded into the brief + 15-min research timebox → landed
fast. Pattern now twice-confirmed (lf4b.4 #1 same shape): deep-research
spawns are the death window; seed + timebox proactively.
Verify discipline held: every run verified by gates not reports —
including independent meter reproductions (lf4b.10 corrupt-fixture drill
exit 1→0; lf4b.6 py:/ts: compile.emission_failure reproduced from the
page's embedded demo spec; lf4b.7 scope_sha256 definition cross-checked
vs specs/model_check.md).
Machine summaries disambiguated (lf4b.11): llm.txt regenerated from
llms.txt body + SUMMARY.md map with divergence-proof test; release.md
commit-stamped by release.yml at release.