[junixlabs]

Two things, because the second was forced by the first landing.

1. A far goal, with kill conditions

TraceOS makes software behavior accountable to intent.

A claim about behavior is accountable when it can be traced back to the decision that
asked for it, and forward to an outcome capable of contradicting it.

docs/DIRECTION.md carries the north star, four horizons — each naming the measurement
that would kill it — what the direction refuses, and the one number that governs all of
it (#10, the cost of keeping a model current). docs/decisions/015-intent-behavior-outcome.md
is the ADR. The README leads with the ambition and puts a status table directly under it,
because a large goal in a README is ordinary practice and hiding it in docs/ made the
project read smaller than it is — what is not ordinary is letting a reader guess which
parts exist.

Why Outcome and not Intent is the load-bearing layer: #45 established that claim is
the only field carrying meaning and nothing reads it. Intent and Behavior are both
statements, and two statements can only disagree about words. Outcome is the only layer
where reality can contradict the model.

2. An observation must outlive the commit it was made on

Symbol narrowing (#38) asked git log -L :symbol:path <observed_ref>..HEAD. That needs
the branch's commits to still exist, and a squash merge deletes them.

Measured on this repository: merging #38 left 7 of 25 observed refs unreachable, and
merging #44 orphaned more. It reproduced only in CI — a developer's clone keeps the
commits alive in the reflog and in stale branches, so the failure was invisible exactly
where the model is authored and total where it is checked. It cost roughly an hour of
believing CI was flaky.

Fix: an observation now records observed_symbol, a hash of the cited symbol's text
at the moment it was read. The line range comes from git's own hunk header at check time
and is never stored, so INV-022 is untouched. Order of answers, strongest first:

	Needs history	Answers
whole-file hash unchanged	no	not changed
observed_symbol unchanged	no	symbol not changed
git log -L <ref>..HEAD	yes	whether the symbol moved in that range
none of the above	—	could not establish → whole-file, loud, and it says why

Also: OBSERVED_REF_UNREACHABLE reports observations stranded by an earlier merge, the
ratchet now names the uncertain assertions behind its count, and narrowing prints
NO NARROW <ref>: <reason> when it cannot answer.

That last group is the session's recurring lesson: three separate blockers today were
diagnosed only after a silent fallback was made to speak. A count with no names is
silence wearing a number.

Checks

37/37 groups, 172/172 assertions — three new, including one that orphans a test
repository's history mid-test and proves the content hash still narrows when the ref is
gone. Verified the suite fails when the hash comparison is stubbed. check_docs 26/26,
check_locators 34/34, ratchet exit 0.

Observations were not mass re-observed to backfill the new field. That would be the
rubber stamp INV-024 exists to prevent; the mechanism degrades loudly and self-heals as
references are genuinely re-checked.