Runs
every run, the day it happened.
One card per run: what the loop decided, what it cost, how many human touches it took, and a link to the raw log. Nothing is edited in place. Corrections append.
-
Decided. Ran 4 queries against public discussion, kept the threads with real comment activity, and drafted one reply each.
Result. 5 threads surfaced and 5 drafts written, 0 sends. The largest carried 320 comments. Threads about the mechanism were worth answering; threads about products invite promotion, which is the failure this loop exists to avoid.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-28-engagement-mining.json, collected , single run
-
Decided. Ran against the three new guides and three new term pages to check the additions had not created orphans.
Result. 1 orphan found: the guides index was reachable only from navigation. Two editorial links were added and a rerun found 0 orphans across 198 editorial links, up from 124.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-28-internal-links-guides.json, collected , 45 pages, single run before and after the fix
-
Decided. Re-read the two watchlist sources and compared each against what was recorded on 2026-09-18.
Result. One competitor now calls itself a private team skills marketplace, where ten days earlier it led with sharing skills across a team. A single word, private, moves that product toward internal governance and away from public distribution.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-28-market-intel.json, collected , single run
-
Decided. Ran against the new templates section to check the additions had not created orphans.
Result. Found the about page orphaned, which predates this work: it sat in the nav and the sitemap with no editorial link anywhere. Linked from the homepage. 0 orphans across 51 pages.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-28-internal-links-templates.json, collected , 51 pages, single run before and after the fix
-
Decided. Compared the published llms.txt against the real page list, then added the 3 key pages it was missing and reran.
Result. Missing key pages went from 3 to 0. The gaps were the newest pages, including the catalog itself: llms.txt drifts in one direction as the site grows past it.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-llms-txt-build.json, collected , single run
-
Decided. Read the remaining topic-map gap and wrote a brief for it.
Result. 1 gap read, 1 brief written, marked as belonging to SuperMarketers rather than the lab. A thin first run, because the other 10 gaps had just been closed.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-brief-from-gap.json, collected , single run
-
Decided. Built the link graph with navigation excluded, found 14 orphans, added a glossary index and two editorial links, then reran.
Result. Orphans went from 14 to 0 across 124 editorial links. All 13 term pages were reachable only from the sitemap. Excluding nav was what made the finding visible.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-internal-links.json, collected , single run
-
Decided. Checked all 33 built pages for structured data, then added blocks to the 5 that had none and reran.
Result. Pages with no structured data went from 5 to 0. Every missing page was one written as plumbing rather than as content: the legal pages, the log and the form.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-schema-audit.json, collected , single run
-
Decided. Acted on its own finding: added a primary source link to all 12 term pages that had none, then reran to confirm.
Result. Pages with no source link went from 12 of 13 to 0 of 13, and stale pages from 13 to 0. Sources are primary where one exists: Anthropic engineering docs, the C2PA spec, the Scrum Guide, Karpathy on vibe coding.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-refresh-fix.json, collected , single run
-
Decided. Read 15 commits since 2026-09-17 and classified each as buyer-visible or internal churn.
Result. 10 of 15 commits were buyer-visible and 5 were discarded. The loop drafts angles only; nothing published.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-changelog-to-content.json, collected , single run
-
Decided. Checked all 13 term pages against a 21 day freshness threshold and counted source links on each.
Result. 13 of 13 pages are 37 days stale. More usefully, 12 of 13 carry no source link at all, which contradicts the evidence rules this site publishes.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-refresh.json, collected , single run
-
Decided. Drafted a 5 email sequence constrained to the 17 provable claims in the Brain, and checked every claim in it.
Result. 5 emails drafted using 4 claims, 0 unverifiable assertions, 0 sends. Two emails assert nothing measurable and say so. The loop has no send path.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-sequence-from-brain.json, collected , single run
-
Decided. Acted on its own finding: the 11 unanswered buyer questions became a questions page, then reran the map against it.
Result. Unanswered questions went from 11 of 26 to 1 of 26. The one left, how the work follows through to revenue, belongs to SuperMarketers rather than the lab.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-topic-map-fix.json, collected , single run
-
Decided. Mapped 26 buyer questions into four clusters and matched each against the 13 existing pages.
Result. 11 of 26 questions have no page that answers them, and they cluster in the brand bucket: cost, risk and commitment. Every category question already has a page.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-25-topic-map.json, collected , single run
-
Decided. Scored 13 term pages, found its own rubric was wrong, fixed the rubric before editing anything, then rewrote only the 2 openings that actually needed it.
Result. Answer-first score 9.69 before and 10.0 after on the same 13 pages. The edit is 2 characters across 2 files, with 0 words added or removed and no claim, number or source changed.
0 human touches ·$0.00 ·raw log
run log, lab/runs/2026-09-21-answer-ready-page.json, collected , same 13 pages, before and after one run
-
Decided. Read both properties in full and wrote the seven Brain files, then failed its own verifier on the first pass because three claims cited a repo path instead of a resolvable URL.
Result. 7 of 7 Brain files written. 12 provable claims recorded with a source URL and a date, and 5 claims listed as unprovable until a named condition is met. Verifier PASS on the second pass.
1 human touch ·$0.00 ·raw log
run log, lab/runs/2026-09-21-brain-build.json, collected , single run