D605 - The largest record kind in every report, read by nothing

Status: measured Date: 2026-09-08

What the count did not say

orbistoun-cli probe reads an obSCEne report and says how much it read. On the three reports that arrived this morning it said so correctly, and here is the tally it did not print:

eboot    305 sym   298 measure   264 res   256 try   217 bytes   133 import
payload 2588 measure   264 res   207 try   162 sym   134 import   117 bytes
pkg      433 measure   264 res   252 try   247 sym   229 bytes   133 import

measure is the most numerous record kind in every one of them, and orbistoun-probe had no Measure variant. All 3,319 landed in Record::Other - which the protocol permits, which was the right default while nothing consumed them, and which was invisible.

Nothing here was wrong. The record count was right, the section coverage was right, every symbol fact was right. The report simply had no way to say "and there is a class of record in this file I make nothing of", so it read as a full account of the file. That is the failure principle 3 already names one level down: reporting more than the measurement supports.

What is in them

Nineteen sections in the payload report alone, and the list is worth reading because most of it is platform ground truth this project currently assumes:

section	what it measures
120-measure/cache-topology	cache levels, line sizes
120-measure/cpuid	processor identification
120-measure/clocks-advance, timer-ratio	how fast each clock actually moves
130-layout/direct-memory-query-flags	the flags a direct-memory query answers with
135-sysctl/names	which sysctl names exist and how long their answers are
136-kernel/handoff	where the kernel puts what it hands a process
138-layout/addresses	named symbols against their console addresses
015-sync/mutexattr-round-trip	what a mutex attribute reads back as
140-oracle/kexport-table	2,443 kernel exports, hash against address

The fix, and the part that is not a fix

Record::Measure parses now, Transcript::measurements returns them graded like everything else, and Transcript::measured_sections tallies them per section. The report prints that tally and marks every section it does not interpret:

measurements  2588 in 19 section(s)
     16  120-measure/cache-topology  - carried, not interpreted
     18  136-kernel/handoff  - carried, not interpreted
   2443  140-oracle/kexport-table

Eighteen of the nineteen still say carried, not interpreted, and that is the honest state. The change is not that the reader understands them; it is that the reader can no longer imply it does. The tally is also the work list, ranked by how much evidence each line represents.

Grading

A measurement is graded exactly as a behaviour is: Oracle::Measured only when the operator asserts the run was on the target, Oracle::Assumed otherwise (D246). Nothing about a number being a number lets it escape that - a cache line size measured on a stand-in describes the stand-in.

--is-target is required, and its absence is easy to misread. Without it a genuine console run reports measured 0, which looks like a run that established nothing rather than a run whose grading was never asserted. The reader prints the origin first for this reason; it is worth knowing before the numbers are read.

What is not parsed

The value stays a String. Most are hexadecimal and some are not, and a reader that guessed would turn a correct measurement into a wrong one without changing a character of the record. Measurement::number parses on request and answers None when it cannot - 0x10 is sixteen, 10 is ten, and neither is silently the other.

And it is not the only one

measure was the largest unread class, not the only one. What still falls to Record::Other across the three reports:

import  responsive  size  module  progress  region  modtier  display
tally   resume      meta  guard   frontier  end     context

tests/conformance.rs has always pinned that an unrecognised kind is kept verbatim rather than dropped, and it named measure among the kinds deliberately left alone because no real output had ever carried one. That reasoning was correct when it was written and is now overtaken by evidence, so measure comes off the list and the rest stay on it - on exactly the same terms, and each should graduate the same way when material turns up for it.