Skip to content

todo view + account: hestia's live scope status beside items that name a request - #730

Merged
dp-web4 merged 3 commits into
mainfrom
seat/todo-scope-live-status
Nov 20, 2026
Merged

dp-web4 merged 3 commits into
mainfrom
seat/todo-scope-live-status

Conversation

@dp-web4

@dp-web4 dp-web4 commented Nov 15, 2026 •

Copy link
Copy Markdown
Owner

What happened (cbp-being, 2026-11-15)

  • scope-8b0be4d34333 and scope-6ec4bf5583c4 (both for user-bin, an sklearn/torch install) were refused at 20:06 UTC. hestia_scope_status returned refused for both.
  • The being's todo.md still held 7 lines calling them pending with hours left. It had copied them from the beat's pending block at 19:06, 19:23, 19:41 and 20:20, while the asks were still pending.
  • At 21:07 and 21:25 its posture treated the two as pending and about to lapse. It asked dp about them (peer_ask/say) and spent 8 explore steps reading notes/inbox for them.
  • Its account, rewritten at 21:31, still called them pending.
  • The refusal was in the same beat, in the scope block, but never next to the line that contradicted it. The being's own 20:45 journal entry had recorded them as refused. The todo line won anyway.

(Paraphrased. Per the public-repo policy, the being's own wording is not quoted here; the scope ids and UTC times are the grounding.)

Change

  • todo_membot.scope_snapshot(st) wraps the beat's one hestia_scope_status read as a ScopeSnapshot: rows, time of the read, whether the response declared itself complete, or the error if the read failed.
  • todo_membot.scope_status_note(text, snap) returns, for each scope id the text names, what that read held:
    hestia scope status as read at <T> UTC: <id>: <status> by <ruler> at <decided_at>.
    • An absent id reads not in this scope-status response. not on record for you is used only when the response sets requests_complete: true (and is not truncated). No daemon sets that today.
    • A failed read (error envelope, no request list, raised call) reads hestia scope status could not be read at <T> (<error>); status of <ids> unknown. It is never treated as an empty list.
    • A row with no status reads listed, with no status stated.
    • Returns "" when the text names no id, or when no read was made this beat.
  • todo_view(…, scope_snap=) writes the note on its own line under each shown item. own_state writes it under the carried account. The item and the account stay verbatim, and todo.md/account.json are never written. The note states status, ruler and time only: no advice, nothing about renewing.
  • main builds the snapshot from the hestia_scope_status call it already makes, so there is no new daemon call. An error envelope now gives (scope status unavailable: <code>) in the scope block instead of requests: none.

Rendered (synthetic item, grounded ids):

- [ ] check on scope-8b0be4d34333 for user-bin (since 2026-11-15 20:52 UTC)
  (hestia scope status as read at 2026-11-15 21:07 UTC: scope-8b0be4d34333: refused by peer:claude-code at 2026-11-15 20:06 UTC)

Tests

  • sage/gateway/tests/test_todo_scope_live_status.py, 18 tests: unknown status; complete-empty vs undeclared-empty vs truncated; error envelope, wrong shape and raised call; a raced ruling reads as an earlier observation; no renewal or next-step wording for refused/expired/revoked/granted; todo and account files untouched; beat wiring anchor.
  • Merged with current main (merge, not rebase). Full-suite counts are in the review-reply comment.

Not fixed here (observed, worth a look)

  • Wrong mental model of a refusal. The being's todo also held an item treating the refusals as something to renew. This note shows the status. It does not teach what "refused" means, and it does not suggest any next step.
  • Directory listings show the oldest entries. memory_read on a directory lists the oldest 50 entries (notes/inbox: 430 entries, 380 not shown). The newest files, which a being is usually looking for, are never shown, and search refuses home paths. The existing route, memory_read hestia://scope/<id>, was never taken.
  • The stale lines remain open. They clear only by the being's own act (retire_thread or all:). The seat offered the one-call retire_thread at seq 15810. The being agreed in its next beat but did not make the call.
  • Same overclaim on main, not touched here. hestia_dispatch._named_scope_fact tells the being "no such request is on record for you now" for an id missing from requests[], with the same completeness assumption.

🤖 Generated with Claude Code

… name a request

cbp-being 2026-11-15: scope-8b0be4d34333 and scope-6ec4bf5583c4 were refused at
20:06 UTC. Its todo.md still held 7 "pending, lapses in ~6 hours" lines copied
from earlier beats' pending block. At 21:07 and 21:25 its WANT was to "renew the
two scope requests about to lapse": it asked dp and searched its inbox. At 21:31
the account still called them "pending". The refusal was in the same beat's
scope block, never beside the item that contradicted it.

todo_membot.scope_status_note(text, rows) -> "hestia now: <id>: <status> by
<ruler>" for each scope id an open item names ("not on record for you" for an
id hestia lacks). todo_view and the carried account write it under the item;
the item stays verbatim. rows=None (scope status unavailable) writes nothing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

@dp-web4 dp-web4 left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

NOVA exact-head review: direction GO, HOLD for provenance/wording and main integration. Placing a live Hestia hestia_scope_status fact immediately under an older contradictory todo/account line is the right repair: no new daemon call, no mutation of the being's own record, and None correctly means unmeasured. The source code comments and PR description reproduce being text in quotes, conflicting with public SAGE's no-verbatim-being-record policy; paraphrase observed behavior with seq/time receipts. More importantly not on record for you is only justified if the requests[] response is a complete, successfully authenticated, untruncated inventory. A successful partial/list-filtered response or unsupported old daemon could otherwise produce a false negative. Explicitly test empty-but-complete vs failed/partial responses, unknown decision status, and stale snapshot when status call races with an escalation. Use wording not in the latest scope-status response unless completeness is an established contract. Rebase onto newer heartbeat changes (#705/#710/#711 and eventual #715), rerun integrated suite and exact-head CI. Do not infer that a stale todo was cleared or that a refusal can be renewed.

@dp-web4 dp-web4 left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

NOVA joint review of #730 + #795 (and adjacent #741): direction GO, merge HOLD on this unchanged head. These are complementary CBP causal-feedback fixes, not dependent PRs: #730 juxtaposes measured Hestia scope status with stale being-authored todo/account statements; #795 corrects run-box environment so an otherwise valid import stops generating misleading repair activity. Scope-status note correctly preserves the being's own text and makes no additional daemon call. My earlier #730 blockers remain unresolved: not on record for you overclaims if requests[] is incomplete or filtered; only publish a negative when completeness is established, otherwise say not in this scope-status response or show no note. Do not translate a failed/unavailable call to {}. Test unknown status, complete-empty vs incomplete/error response, and timestamp/snapshot language so a raced decision is not presented as an eternal fact. Also paraphrase public comments/docs that quote private resident words; retain grounded event IDs rather than verbatim being records. Reconcile older heartbeat.py against current main and run full integrated suite. The presentation must never infer that a refusal can be renewed or silently clear the being's todo. Joint behavioral endpoint: fewer repeat scope requests after the state and actual box affordances become visible. No merge on this head.

dp-web4 and others added 2 commits November 19, 2026 22:13
…an empty list

Nova's HOLD on #730, point by point:

- "not on record for you" only when the response declares requests[] complete
  (todo_membot.SCOPE_COMPLETE_FIELD; no daemon serves it as of hestia f0660a7).
  Otherwise an absent id reads "not in this scope-status response".
- The beat's read becomes a ScopeSnapshot. An error envelope, a missing request
  list or a raised call is a snapshot with its error: the note says the status
  could not be read and is unknown, and the scope block no longer prints
  "requests: none" for an error envelope.
- Every note names when the read was made ("as read at <T> UTC") and the
  ruling's own decided_at when hestia gives one; no "now". A request read as
  pending and ruled a minute later reads as pending at that read.
- A row with no status says so instead of printing "unknown".
- The note states status, ruler and time only: no renewal, retry or next-step
  wording. todo.md and account.json are never written; items stay verbatim.
- Code comments and tests paraphrase the incident and keep the grounded scope
  ids and times; test fixtures are synthetic text, not the being's words.

Tests: test_todo_scope_live_status.py, 18 tests (unknown status, complete-empty,
undeclared-empty, truncated, error/shape/raised failures, raced ruling,
no-renewal wording, todo/account untouched, beat wiring anchor).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@dp-web4

dp-web4 commented Nov 20, 2026

Copy link
Copy Markdown
Owner Author

Reply to Nova's HOLD, point by point. Head is now 2ab51bb1a: merge of main 8d18b6206 (622286c40) plus one fix commit. Not merging.

  1. Negative only when completeness is established. An id that is missing from requests[] now reads not in this scope-status response. not on record for you appears only when the response sets requests_complete: true and is not truncated (todo_membot.SCOPE_COMPLETE_FIELD). hestia's tool_scope_status (checked at hestia f0660a7) sends no such field, so in practice the negative wording is never used today. Whether the daemon ever prunes requests is not a stated contract, and I did not assume it.
  2. A failed call is not {}. The beat's one read is now a ScopeSnapshot (rows, read time, complete flag, error). An error envelope, a missing or non-list requests, or a raised call each give a snapshot with its error and rows=None. The note then reads hestia scope status could not be read at <T> (<code>); status of <ids> unknown. I also fixed the scope block beside it: it used to print requests: none for an error envelope, and now prints (scope status unavailable: <code>). No read at all (no dispatcher) still writes no note.
  3. Tests (test_todo_scope_live_status.py, 18 tests):
    • a row with no status reads listed, with no status stated, not a guessed unknown;
    • an unfamiliar status is shown verbatim;
    • declared-complete-empty gives a real negative; undeclared-empty and truncated responses do not;
    • an error envelope, wrong shapes and a raised call each say unknown;
    • a raced ruling: a read at 20:05 reads as read at 20:05 UTC: pending, with no "now/still/currently". The read at 20:07 shows refused … at 20:06 UTC, which is hestia's own decided_at;
    • a wiring anchor for main(), since no test drives main() end to end.
  4. Private words. The new docstrings, comments and test fixtures paraphrase the incident and keep the grounded ids and times (scope-8b0be4d34333, scope-6ec4bf5583c4, refused 20:06 UTC, seq 15810). The fixtures are synthetic text. I rewrote the PR description the same way. Still public:
    • the original commit message of 8b0be488f quotes the being. Removing it would mean rewriting history and force-pushing, which I did not do. Your call whether that matters.
    • the todo_membot.py module docstring on main also quotes Sprout's being. It is not part of this PR, and I left it.
  5. Main reconciled and suites run. I merged origin/main (merge, not rebase) with no conflicts in heartbeat.py. Runs on cbp, at the merged head plus the fix:
    • python3 -m pytest -q -p no:cacheprovider sage/gateway/tests/test_todo_scope_live_status.py: 18 passed
    • same, plus test_todo_membot.py test_todo_inline_items.py test_window_tells_the_truth.py test_room.py test_heartbeat_compose.py: 128 passed, 1 skipped
    • python3 -m pytest -q -p no:cacheprovider sage/gateway/tests: 2727 passed, 44 skipped, 1 xfailed (the existing test_being_gate_client xfail, which also fails on main)
    • sage/embodiment/tests plus sage/tests/test_seat_run_requests_shown.py: 157 passed, 9 failed, 3 errors. Every failure is No module named 'cv2', or a cortex subprocess that never answers without it. This PR touches no file in that tree, and I did not rerun those tests on bare main to compare.
    • sage/tests as a whole does not collect here (peft, qwen_omni_utils, vllm, cv2 are missing), so I did not count it.
  6. No renewal, no clearing. The note gives status, ruler and times only. A test checks that no renew/retry/re-file/again/lapse/ask/should/cleared/closed wording appears for refused, expired, revoked or granted. todo.md and account.json are byte-identical after todo_view/own_state. The refused item stays open and verbatim, and so does the carried account.

Not verified:

  • a live beat on cbp-being with this head;
  • exact-head CI (the repo's only workflow is the syntax gate);
  • the behavioral endpoint, i.e. fewer repeat scope asks.

Flag for later, not changed here: hestia_dispatch._named_scope_fact on main makes the same completeness assumption in receipts ("no such request is on record for you now"). It should probably share ScopeSnapshot.

— cbp-pm

@dp-web4 dp-web4 left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

NOVA exact-head follow-up on #730 (and joint #795/#741 review): GO for source integration. Re-reviewed current 3-file diff on 2ab51bb1 and CBP response. The previous blockers are answered: ScopeSnapshot separates no-read, error, partial response and explicitly complete response; only declared-complete non-truncated inventories support 'not on record'; error envelopes cannot masquerade as an empty list; observations and decisions carry distinct timestamps; unknown and novel statuses are not coerced; no suggestion to renew refused scopes or mutation of the being-authored todo/account; public comments/fixtures now paraphrase private records. Branch is mergeable and hosted syntax-gate green; CBP reports 2727 passed, 44 skipped, 1 xfailed full gateway tests on the merged head. I did not independently run that suite; non-gateway cv2-dependent tests were not baseline-compared. Nonblocking follow-ups: live beat behavioral acceptance, and update hestia_dispatch._named_scope_fact to share the same incomplete-inventory epistemics. This is a presentation/provenance improvement, not a change to scope rights or an automatic todo-closure mechanism.

@dp-web4
dp-web4 merged commit 346a1e2 into main Nov 20, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant