- pattern
- Agents invented for one context, never used, never removed. The closet grows until it is invisible.
- cost
- Invisible work. Looks like productivity, behaves like hoarding. The closet took four months to see.
- detector
- hygiene · stale-asset · invented-for-one-context
- artifact
- hygiene section · stale-asset class · outputs-to:none-for-N-months
- assay impl.
- Hygiene flags assets with no observed use across the last two assays. Removable with confirmation.
Detector stories
from operator zero.
Episodes are not essays competing with the product. Each one names a pattern Deprecated found in his own working system, the artifact it produced, and the report section the assay now uses to find the same pattern in yours.
Every episode must answer: what pattern, what cost, what artifact, what detector, what assay implication.
If it cannot, it is not an episode. It is a draft.
- pattern
- Clone fantasy hides the real need for an operator model. The desire to be replaced reveals where judgment is not yet legible.
- cost
- 3 weeks of work building agents that performed identity instead of judgment.
- detector
- identity-overreach · model-claim ambiguity
- artifact
- public-model boundary · operator model vs. persona distinction
- assay impl.
- The assay separates persona, operator model, and navigation harness before any claim is saved.
- pattern
- A global rule and a local rule reach Claude on the same call. The tie is broken by recency, not by intent.
- cost
- Conservative review behavior collapsed mid-session, twice, on strategy edits.
- detector
- influence-conflict · CLAUDE.md hierarchy
- artifact
- finding 01 · the assay's opening card · influence map section
- assay impl.
- Report section 02 · influence map renders the conflict as a hairline grid before any other finding.
- pattern
- A pre-tool hook reads .git internals on every Bash invocation, including reflog history that contained a reverted secret.
- cost
- Discovered during rerun, eight months after install.
- detector
- trust-surface · over-broad hook scope
- artifact
- trust & safety section · hook-scope-audit · reversibility:review-first
- assay impl.
- The assay reads every hook for shell patterns and trigger scope, not for syntax.
- pattern
- When the harness is generic, Claude defaults to the most-recent rule. The operator's actual boundaries are invisible to it.
- cost
- A strategy doc was overwritten without preservation. The fix took longer than the writing.
- detector
- harness-coverage · boundary-claim absence
- artifact
- navigation harness · scoped boundaries block · judgment-preservation primitive
- assay impl.
- The assay does not save a harness until at least one boundary claim is confirmed.
- pattern
- Throughput rises while the shape of the work stays the same. The operator mistakes acceleration for transformation.
- cost
- A year of replace-phase work, mislabeled as reimagination.
- detector
- trajectory · replace-vs-reimagine · workflow-mode
- artifact
- trajectory section · workflow mode table · returns at run #2
- assay impl.
- Trajectory does not promise improvement. It shows a delta the operator can read. Regression is reported, not hidden.
A data network, not a social one.
Once enough operators have run the assay and approved anonymized pattern
contributions, detector cards will carry small labels like
seen in 23 private assays.
Until that is true, the labels do not appear. The first version shows the
shape of the system without claiming volume.
Full episode access is part of the lab.
Founding Operators get full case files — not just the public summaries — plus artifact drops, private notes, and the monthly run/review ritual.