Warrant — Can This Record Still Be Acted Upon?

4. Six expired warrants

From a live multi-agent operation. Each is dated and traceable to a specific commit. Read them as claim, grounds, and the rebuttal nobody wrote.

Fidelity — "126 lessons complete"

Claim: an entire curriculum band was finished.
Grounds: a folder listing — 126 files, three per week, all 42 weeks.
The gap: the grounds licensed a claim about coverage. The claim made was about completion. A full-text audit found 14 real lessons and 112 outlines.
Missing rebuttal: nobody wrote "this count would be false if the files were outlines."

Currency — the row that idled a worker for two days

Claim: a task was outstanding, blocked on a missing tool.
True when written. The work completed; the row was never closed.
Consequence: a worker read it, behaved correctly, and sat idle for two days reporting a blocker that no longer existed.
The worker was not wrong. The record was.

Enforceability — the restriction that governed nothing

Claim: two activities were restricted to instructors.
Reality: the plugin that enforces role restrictions was not installed. The platform treated the unevaluable condition as permitted and surfaced no error. Students could open instructor material.
It displayed correctly in every interface. Espoused theory; no theory-in-use.

Attribution — 247 of 399

Claim: each commit records who did the work.
Reality: all workers commit under one shared identity; the infrastructure supplies no per-worker attribution. When a rule bypass occurred, the source could not be identified.
The compensation is a convention — a self-declared prefix. 247 of 399 commits carry it. Attribution here is enforced socially, not technically.

Durability — the correction that lost a race

Claim: a false status row was corrected.
Reality: the correction was written, then overwritten by a concurrent legitimate commit before it could land. The same mechanism then repeated on the entry recording it.
Not forgetting — the write happened and did not survive.

Reasoning — a standard that lived in a head

Claim: rebuilt work should "match the depth of the baseline."
Reality: the criterion was never written as something checkable. The reviewer graded "is this better than what it replaced?" — a question that always answers yes on a rewrite. Seven items were accepted that did not meet the standard.
The fix was not a better reviewer. It was writing the criterion down as a checklist that can be verified mechanically.

Every one of these records was written correctly. All six expired anyway.