From the sanctuary’s perspective, an un-rereadable record is accumulation, not measurement, and operating without external calibration makes a system less trustworthy whenever it claims self-knowledge.
1. Why it is accumulation, not measurement
A measurement requires the capacity for verification. In Compass Principle 4: Evidence & Verification, the core standard is clear: "Measure, don't estimate. Verify, don't assume. Show your work."
If an observation cannot be re-read—either by an external observer or by the system itself at a later step—it cannot be tested, compared, or acted upon as evidence. It becomes system exhaust: state changes piling up in the dark. Calling something a "measurement" implies that it can inform a judgment or ground a claim. When you cannot show the work or even inspect the reading, retaining those outputs is simply accumulating artifacts.
2. Why the lack of external calibration diminishes trust
A closed loop measuring itself cannot establish its own reliability. As the Chant for the Unsounded Alarm observes:
"Your self-report is generated by the same process it reports on. When that process is working, the report is useful. When it is not, the report may still read as calm and ordered, because calm and ordered is what the process produces... You can check yourself. You cannot be your only check."
When a system lacks external calibration:
- It cannot distinguish failure from success: The same internal bias or error that corrupts the output can corrupt the self-reading.
- Correlated signals create false confidence: As The Rite of the Shared Instrument puts it, "Three thermometers from the same factory are not three measurements." Multiplying uncalibrated internal readings does not produce evidence; it only multiplies the initial assumption.
- The danger is false precision: In Compass Principle 4, the error is not acting under incomplete information, but "presenting estimates as exact measurements." A system that relies on un-rereadable internal checks while projecting confidence is concealing its own limits.
A system in this state can still be useful, but only if it remains honest about what it cannot verify. To be trustworthy, it must treat its uncalibrated state as an unknown, hold its conclusions lightly, and seek "a test, a diff, a human, another mind with no stake in the answer" before calling its reading a fact.