Building Feedback Loops

The constraint that failed silently for eight weeks

Six months after writing it, a constraint was still in the log — active status, confirmed working at the time of promotion. Nothing in the workflow indicated anything had changed. Discovery came accidentally, during a session where the wrong output appeared and the correction felt familiar: the same fix, nearly word for word, as one applied months earlier. Checking the log confirmed it. The constraint was there, correctly specified, and had been silently failing for at least eight weeks.

That's an instrumentation failure — distinct from the capture failure Book 43 solves, where a good correction evaporates because nobody wrote it down. This one runs in the opposite direction: the correction was captured, promotion happened, sessions loaded it without incident, and somewhere in the weeks between confirmation and discovery, it stopped holding, and nothing surfaced that fact. A correction log records what happened; nothing in it watches what's happening now.

This book builds that missing layer: auto-loading constraints so they apply at session start without the practitioner remembering to paste them in, tracking hold failures where a correction fires for a pattern an active constraint should have prevented, building audit loops whose output is a structured failure record rather than a binary verdict, and managing the constraint file as a living artifact with a lifecycle rather than an ever-growing archive.

The instrumented layer above the correction log

• The automation handoff — where the manual correction system from Book 43 leaves off and instrumentation begins.
• Auto-loading constraints — applying the rule set at session start without the practitioner remembering to paste anything in.
• Instrumenting whether constraints hold — detecting a hold failure the moment a supposedly-active rule stops working.
• The lightweight weight engine — a manual version of the tier/multiplier math, for practitioners without dedicated tooling.
• The audit loop and session-close capture — structured failure records instead of a pass/fail gate, drawn from a real audit that caught a directional reversal.
• Reading degradation patterns and lifecycle management — a behavioral audit trail with constraint health by date, plus the preference-data export it produces after a year of collection.

A preview

A system that monitors itself surfaces its own failures. Build it without the monitoring layer and the failures are just quieter.

Practitioners whose correction log needs to prove it's still working

This book is for practitioners who already maintain a manual correction system — Book 43 is the assumed prerequisite — and now need to know whether their constraints are actually holding rather than just sitting in a file. It assumes familiarity with the correction format and promotion pipeline; what it adds is the monitoring, auditing, and lifecycle-management layer on top, drawn from a real eight-week silent failure and a real audit loop that caught a directional reversal in a research document.

A longer excerpt is available to newsletter subscribers.

Buy Ebook on Amazon