test(aterm): assert against the live roster, not only the fixtures (#1259)

coilyco-ops commented

2026-08-26 08:20:27 +00:00

Member

Closes #1259. Tier 3 of the #1245 dispatch, deliberately last because it is only meaningful once tiers 1 and 2 consumed the fields.

`just aterm-contract`

--- PASS: TestLiveOverlayDiscardsNoField                    (15 role/seat pairs)
--- PASS: TestLiveRosterEveryLaunchableSeatResolvesToAPlan   (15 role/seat pairs)
--- PASS: TestLiveRosterRefusesEveryNonNativeSeat
--- PASS: TestLiveRoleBackgroundsStaySeparable
        closest background pair eval/sysadmin at dE 3.25 across 7 roles
--- PASS: TestLiveDefaultSeatBelongsToItsRole
--- PASS: TestLiveSoundMarksAllHaveASample

The one the issue says would have caught this whole backlog

Decode the live overlay into a generic map, decode it into the typed struct, re-encode, diff leaf by leaf. Anything present upstream and absent downstream is a failure. The waiver list is empty and an entry in it needs a reason, not a shrug.

Verified against a negative control rather than assumed: retagging motif as json:"-" fails the test 30 times across the live roster, once per personality per role/seat pair, each naming the dropped path. That is what turns sound_mark and form sitting unread into a failure on the day they are added.

The colour floor is a ratchet, not a target

Backgrounds are compared in CIELAB, because hex distance says nothing about whether an eye can separate two windows. Measured today across the seven roles:

21 pairs, closest eval/sysadmin at dE 3.25, furthest at dE 13.59
0 pairs under dE 2, 4 pairs under dE 5

The floor is dE 3.0, which today's roster clears by 0.25. agent-compose#358 raises the ceiling; this stops the floor falling, which is the half aterm can hold on its own. labDistance is a real addition to brand.go rather than test-only arithmetic, since #358's fix will want it too.

Skipping is loud

CI carries no agent-compose, so just aterm-test skips these and stays green. The skip names the escape hatch, and setting it makes the same absence a failure:

$ ATERM_LIVE_ROSTER=1 go test -run TestLive .
    ATERM_LIVE_ROSTER is set but agent-compose is not on PATH: ...
    FAIL

$ go test -run TestLive -v .
    no agent-compose on PATH, so the live roster cannot be read.
    Set ATERM_LIVE_ROSTER to make this a failure.
    --- SKIP

just aterm-contract sets it, so the recipe is the enforcing surface and a plain test run is the permissive one. Both behaviors exercised above.

Also in here

docs/aterm.md had room again after the status-line composer moved out, so this corrects a sentence #1251 left behind: the overlay's non-glyph fields no longer "reach nothing", they reach the identity card. The contract recipe is documented beside it.

Verification

just aterm-test, just aterm-contract, just aterm-fmt, just aterm-lint clean; pre-commit run --all-files clean
the negative control and both skip behaviors above, run rather than reasoned about

Closes #1259. Tier 3 of the #1245 dispatch, deliberately last because it is only meaningful once tiers 1 and 2 consumed the fields. ## `just aterm-contract` ``` --- PASS: TestLiveOverlayDiscardsNoField (15 role/seat pairs) --- PASS: TestLiveRosterEveryLaunchableSeatResolvesToAPlan (15 role/seat pairs) --- PASS: TestLiveRosterRefusesEveryNonNativeSeat --- PASS: TestLiveRoleBackgroundsStaySeparable closest background pair eval/sysadmin at dE 3.25 across 7 roles --- PASS: TestLiveDefaultSeatBelongsToItsRole --- PASS: TestLiveSoundMarksAllHaveASample ``` ## The one the issue says would have caught this whole backlog Decode the live overlay into a generic map, decode it into the typed struct, re-encode, diff leaf by leaf. Anything present upstream and absent downstream is a failure. The waiver list is empty and an entry in it needs a reason, not a shrug. Verified against a negative control rather than assumed: retagging `motif` as `json:"-"` fails the test **30 times** across the live roster, once per personality per role/seat pair, each naming the dropped path. That is what turns `sound_mark` and `form` sitting unread into a failure on the day they are added. ## The colour floor is a ratchet, not a target Backgrounds are compared in CIELAB, because hex distance says nothing about whether an eye can separate two windows. Measured today across the seven roles: ``` 21 pairs, closest eval/sysadmin at dE 3.25, furthest at dE 13.59 0 pairs under dE 2, 4 pairs under dE 5 ``` The floor is dE 3.0, which today's roster clears by 0.25. agent-compose#358 raises the ceiling; this stops the floor falling, which is the half aterm can hold on its own. `labDistance` is a real addition to `brand.go` rather than test-only arithmetic, since #358's fix will want it too. ## Skipping is loud CI carries no `agent-compose`, so `just aterm-test` skips these and stays green. The skip names the escape hatch, and setting it makes the same absence a failure: ``` $ ATERM_LIVE_ROSTER=1 go test -run TestLive . ATERM_LIVE_ROSTER is set but agent-compose is not on PATH: ... FAIL $ go test -run TestLive -v . no agent-compose on PATH, so the live roster cannot be read. Set ATERM_LIVE_ROSTER to make this a failure. --- SKIP ``` `just aterm-contract` sets it, so the recipe is the enforcing surface and a plain test run is the permissive one. Both behaviors exercised above. ## Also in here `docs/aterm.md` had room again after the status-line composer moved out, so this corrects a sentence #1251 left behind: the overlay's non-glyph fields no longer "reach nothing", they reach the identity card. The contract recipe is documented beside it. ## Verification * `just aterm-test`, `just aterm-contract`, `just aterm-fmt`, `just aterm-lint` clean; `pre-commit run --all-files` clean * the negative control and both skip behaviors above, run rather than reasoned about

coilyco-ops added 1 commit

2026-08-26 08:20:28 +00:00

test(aterm): assert against the live roster, not only the fixtures (#1259 )

ci / ward-doctor (pull_request) Successful in 8s

Details

ci / aos-cli-tests (pull_request) Failing after 28s

Details

ci / aos-eval-tests (pull_request) Successful in 48s

Details

ci / gate (pull_request) Successful in 54s

Details

3e39070110

The fixtures prove aterm still works on the roster it was written for. The
failure that happens is upstream drift, and nothing caught it.

`just aterm-contract` walks the live roster. Every launchable seat resolves to
a plan whose title leads with its workspace and whose card carries every
personality. Every catalogue seat outside the native set refuses with exit 3
and says which of the two checks it failed. Every role's default seat is one
that role can launch, asked of aos rather than parsed a second time. Every
timbre the roster ships has a committed sample.

The one worth the recipe decodes the live overlay, re-encodes it through the
typed struct, and diffs leaf by leaf, so a field Agent Compose ships and aterm
drops fails on the day it is added rather than sitting unread for months.
Nothing is waived; an entry in that list needs a reason. Verified against a
negative control: retagging one field as discarded fails it 30 times across the
live roster.

Backgrounds get a dE 3.0 floor on the closest pair, measured in CIELAB because
hex distance says nothing about whether an eye can separate two windows. That
is a ratchet rather than a target: the closest pair today is eval/sysadmin at
dE 3.25, and agent-compose#358 raises the floor.

A host without agent-compose skips and names ATERM_LIVE_ROSTER, which turns the
same absence into a failure where the roster was supposed to be. `just
aterm-test` therefore stays green in CI, which carries no roster.

Closes #1259

Co-authored-by: Kai Siren <coilysiren@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Agent-Role: platform

coilysiren added 1 commit

2026-08-26 08:25:51 +00:00

Merge branch 'main' into aos/claude/mu55-contract-tests

ci / aos-cli-tests (pull_request) Failing after 42s

Details

ci / aos-eval-tests (pull_request) Successful in 9s

Details

ci / ward-doctor (pull_request) Successful in 7s

Details

ci / gate (pull_request) Has been cancelled

Details

22935be34b

coilyco-ops added 2 commits

2026-08-26 08:27:25 +00:00

fix(aterm): let the live default-seat check meet an aos too old for the verb (#1259 ) 6f3abb6ac6

The CI image carries agent-compose, so the live contract tests do run there,
which is better than assumed. Its aos is older than main and answers
`_launch-agent` with "integrated launch needs --role", which failed the check.

aterm itself degrades to the catalogue order against an aos too old for the
verb, so this now does the same: every role refusing is an old binary and skips
with the refusal quoted, and some roles refusing while others answer is the
defect the check exists for and still fails. Both branches exercised against
stub binaries.

Co-authored-by: Kai Siren <coilysiren@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Agent-Role: platform

Merge remote-tracking branch 'origin/aos/claude/mu55-contract-tests' into aos/claude/mu55-contract-tests

ci / aos-eval-tests (pull_request) Successful in 6s

Details

ci / ward-doctor (pull_request) Successful in 7s

Details

ci / aos-cli-tests (pull_request) Successful in 28s

Details

ci / gate (pull_request) Failing after 53s

Details

98bf6f98fa

coilyco-ops added 1 commit

2026-08-26 08:30:59 +00:00

docs(aterm): fit the contract paragraph back under the char cap (#1259 )

ci / ward-doctor (pull_request) Successful in 12s

Details

ci / aos-cli-tests (pull_request) Successful in 34s

Details

ci / gate (pull_request) Successful in 1m5s

Details

ci / aos-eval-tests (pull_request) Successful in 3m38s

Details

cc22cf5ed0

The #1279 bundle fix and this branch both added to docs/aterm.md inside the
same hour and the page went 481 chars over. Trimmed to fit. Fourth instance
on #1274.

Co-authored-by: Kai Siren <coilysiren@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Agent-Role: platform

coilyco-ops merged commit 6ef25af844 into main

2026-08-26 08:35:51 +00:00

coilyco-ops referenced this pull request from a commit

2026-08-26 08:35:52 +00:00

test(aterm): assert against the live roster, not only the fixtures (#1259) (#1278)

Rows
Columns

test(aterm): assert against the live roster, not only the fixtures (#1259) #1278

just aterm-contract

The one the issue says would have caught this whole backlog

The colour floor is a ratchet, not a target

Skipping is loud

Also in here

Verification

`just aterm-contract`