fix(approvals): actor_id records the person who decided, acted_as the slot it was taken as; stored slot literals move out at boot

[objectstack-fleet]

Fixes #21411
Clause-②: no

What changes

sys_approval_action.actor_id is a lookup to sys_user. Under ADR-0118 D1 it holds a user id or nothing. A slot-gated action recorded the SLOT it took there instead: the literal position:P for a position staffed after the request opened, or an email for a user approver authored as one.

Measured on a booted showcase at main 53fd35e3e: a position-slot approve wrote 8 rows across both databases, and none of them names the person who decided. The 3 audit-ledger rows carry user_id / actor null, the 3 activity rows carry actor_id null, and the action row carries actor_id position:finance. That is the raise to p1.

Triage ruling B (5954022700), with its retriage answers (5960329631), puts the person and the slot in two columns.

• Column. sys_approval_action.acted_as (text, max 255) holds the pending-approver slot the action was taken as, in its stored spelling. actor_id gains its ADR-0118 D1 describe. Both are in highlightFields and the recent / all_actions grids, as via_override was. Translations are regenerated, with zh-CN, ja-JP and es-ES written by hand.
• Writes.
  • Every insert in decideNode, sendBack (revise and auto-reject), reassign, requestInfo, comment, recall, resubmit and remind records actor_id: recordedActor(actorId, context). That is the person the context vouches for, never an address the caller named. The slot-gated ones also record acted_as: slot ?? null.
  • A system context that vouches for nobody records nobody. One exception, kept by name: the SLA sweep's reserved system:sla, which is approvals: ADR-0118 D1 family — sys_user actor columns still hold sentinels and slot literals (system:sla / system:dead-run, reassign_from / reassign_to, sys_notification.actor_id) #21455's.
  • The action link (redeemActionToken) now puts the person behind the token's slot on the context: a user id as itself, an email as the ONE account carrying it, otherwise nobody. Before, it put the slot itself there, so an email-bound link recorded the email as the acting user, on the action row and in the status mirror alike.
• Readers.
  • The multi-approver tally and decision_progress read acted_as. There is no ?? to actor_id anywhere.
  • The already-acted probe follows Q3 = A: actor_id equals the caller's user id, OR acted_as is in actingAddresses(caller). These are two facts, each compared only with its own identity kind.
  • listActions maps acted_as onto ApprovalActionRow.acted_as, which spec(contracts): ApprovalActionRow declares the slot an approval action was taken as (an optional member), so the action log can show it once actor_id holds the person (#21411 ruling B) #21458 declared and which merged first, beside actor_id and actor_name. A row no slot admitted omits the member.
• Boot-time backfill (action-slot-backfill.ts). It is hooked on kernel:ready beside backfillApproverIndex, the plugin's existing boot-time repair, so it ships as code and no agent runs a bulk write.
  • Pass 1 covers the whole history. A row whose actor_id holds a slot address (contains : or @, and is not one of the reserved system:sla / system:dead-run) gets acted_as := actor_id and actor_id := null. No stored record names its decider, so null is ADR-0118's value, not a guess.
  • Pass 2 covers the approve votes, at step 0, of still-pending requests whose acted_as is empty: they get acted_as := actor_id. This is exact, not a guess, because an override finalizes the node.
  • No other row is stamped. Finished user-id rows are read by their person.
  • It is idempotent (both predicates are empty after a run). A failure logs at error with the consequence and the fix.
• Docs. content/docs/automation/approvals.mdx: the sentence that said the decision is recorded in actor_id under its slot now names both columns.

Pins

• approval-service.test.ts:
  • The approvals: a holder of a position whose slot reads position:<p> can see the request but cannot decide it with the default actor (can_act false, approve 403), and loses sight of it after deciding (404) #21379 slot tests now assert [actor_id, acted_as].
  • A approvals: sys_approval_action.actor_id (a sys_user lookup) records the slot literal (position:<p>, or an email) instead of the deciding user, so the person who decided is on no column (ADR-0118 D1) #21411 block covers:
    • position, email and user-id slots through the session door;
    • the action link: user id, an email with an account, an email with none;
    • an override recording the admin and no slot, even when the admin named a position address;
    • a system context recording nobody, with the SLA sentinel kept;
    • the per_group tally and decision_progress counting slots;
    • both probe halves, including a holder who lost the position, a legacy row found by its person, and a literal left in actor_id never read as a slot;
    • the action log showing person, name and slot, and a backfilled row showing the slot alone.
• approver-address-readers.test.ts, the enumeration pin: acted_as replaces actor_id among the slot columns. Every read of actor_id in the package is classified with its count. The only comparison is with the caller's user id (actor_id: uid in visibleRequestIds).
• action-slot-backfill.integration.test.ts runs on a real ObjectQL with SqlDriver (better-sqlite3, in-memory) and the real DDL:
  • literals are moved;
  • a pending vote is stamped;
  • a finished user-id row, the sentinels, a system row and a new-style row are untouched;
  • a second run writes {0, 0};
  • end to end, an in-flight unanimous request still finalizes on the votes the old writer recorded.
• action-slot-backfill-wiring.test.ts: the plugin runs the repair once at kernel:ready (never at start()), against its own engine, and logs a failure at error.
• packages/qa/dogfood/test/position-address-readers.dogfood.test.ts, a cross-lane domain:cli addition declared on [PM seat] domain:cli — ⏳ vacant #6024: recorded() reads actor_id, acted_as and via_override, and asserts the person plus the slot, and the admin plus no slot on an override.

Ablations (committed first; every leg through scripts/ablation-replace.mjs, with the anchor hit once, the blob changed, the restore proven equal to HEAD and git diff HEAD empty)

Legs a to f ran at b668cf134; leg g ran at dc6d72a9c. Each direction was predicted before its run.

leg	mutation	predicted	observed
a	decideNode writes no acted_as	red, more than the slot pins (the tally too)	24 failed / 304 passed: every slot pin, plus the unanimous, quorum and per_group tallies and the probe. ⚠ The first attempt was a no-op: the tool refused it because the replacement text was a substring of the anchor (count 1 to 1), and it restored. Re-run with a distinct replacement.
b	tally reads a.actor_id	red: tally pins, backfill end to end, enumeration pin	14 failed / 320 passed across all 3 files
b2	decision_progress reads a.actor_id	red: per_group progress pin and enumeration pin	4 failed / 328 passed
c	backfill pass 2 writes nothing	red: backfill pin	1 failed / 1 passed
d	backfill pass 1 writes nothing	red: backfill pin	1 failed / 1 passed
e	the kernel:ready hook is removed	red: wiring pin	2 failed
f	the probe's { actor_id: uid } half is removed	red: person pins and enumeration pin	3 failed / 329 passed
g	listActions stops mapping acted_as	red: action-log pin only	1 failed / 328 passed

Gates, at dc6d72a9c

• dispatch-gates --repo objectstack-ai/objectstack --commands derived 97 commands. All 97 ran with exit 0, plus the 4 in-path roster gates it flagged (check-changeset-fixed, check:authz-resolver, check:error-code-casing, check:filter-alias-parity), also exit 0. --ran with exit codes answers: 97 derived, 97 run, 0 NOT-MEASURED (a derived zero).
• pnpm --filter @objectstack/plugin-approvals typecheck: exit 0 (check:test-typecheck OK). pnpm --filter @objectstack/plugin-approvals test: 58 files, 868 tests passed.
• The dogfood file: 1 of 1 passed. pnpm --filter @objectstack/dogfood typecheck: exit 0.
• Earlier reds, each repaired in this PR:
  • check:engine-double-contract: the wiring test's double now declares no write verbs.
  • check-tenant-audit-census: the backfill adds 2 elevated update sites (229 to 231 write call sites, 110 to 112 elevated). node scripts/tenant-audit-census.mjs --write regenerated both census tables, and the page's hand-written prose counts were updated to match. This adds content/docs/permissions/tenant-audit-census.mdx and docs/audits/2026-08-tenant-audit-write-call-sites.counts.md to the diff, declared here.

Changeset

@objectstack/plugin-approvals patch, Clause-②: no (the spec widening is #21458's). It states that it supersedes the "What is recorded" sentence of the unreleased .changeset/21379-position-address-readers.md, which this PR does not edit. Both ship in one release.

Acceptance notes

• One widening, by ruling (Q3 = A): a person who decided under a position they later lost keeps sight of that request. The old probe hid it.
• Behaviour narrowed for system contexts: a machine caller that names an actor without vouching for a person on the context now records actor_id null, with the slot still taken and recorded. The only first-party machine callers are the SLA sweep (kept by name) and the action link (which now vouches for the resolved person). Test fixtures that decided from a bare system context and asserted actor_id moved to a vouching context.
• Out of this PR, the ADR-0118 D1 family on approvals: ADR-0118 D1 family — sys_user actor columns still hold sentinels and slot literals (system:sla / system:dead-run, reassign_from / reassign_to, sys_notification.actor_id) #21455: the SLA and dead-run sentinels in actor_id, reassign_from / reassign_to holding slot literals and emails (reassign_from still records the slot handed over; a pin here names it), and sys_notification.actor_id fed by notify.
• Console: rendering acted_as beside the actor's name in the timeline is objectui work. Until it lands, a backfilled historical row shows no actor, and the slot is on the wire.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

This PR changes 1 package(s): @objectstack/plugin-approvals, touching 39 documentable anchor(s).

22 hand-written doc(s) name something this change touched — list omitted above 15 rows. Re-derive on the tree named below: node scripts/docs-audit/affected-docs.mjs --json 100c394f6fd0113377bccdabed6ecf3b04b720da.

⛔ 8 release-owned page(s) also affected — read-only, see AGENTS.md Documentation Guardrails.

What this run could not see

• 2 anchor(s) matched too much of the corpus to be a work list: created_at (literal, 36 pages), sys_user (literal, 37 pages)
• 8 name(s) were too generic to anchor anything (single lowercase words)
• the SDK route bridge reached 54 of 206 client-bound route-ledger rows — the other 152 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 152: 0 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 55 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 97 are undecided (no in-repo declaration, on a ledger that has other in-repo registrars — absence and an unreadable spelling are not distinguishable here). The rows themselves: node scripts/docs-audit/affected-docs.mjs --bridge-coverage
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 6 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 100c394f6fd0113377bccdabed6ecf3b04b720da → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 0666a388f7a4731b50ffc3acae825ee9d75d3ac2 — the merge of head dc6d72a9cdf3d1bdf2bc5f0321b0ad3b0448a26a into base 100c394f6fd0113377bccdabed6ecf3b04b720da, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin 0666a388f7a4731b50ffc3acae825ee9d75d3ac2 && git checkout 0666a388f7a4731b50ffc3acae825ee9d75d3ac2
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin 100c394f6fd0113377bccdabed6ecf3b04b720da dc6d72a9cdf3d1bdf2bc5f0321b0ad3b0448a26a && git checkout -B drift-repro 100c394f6fd0113377bccdabed6ecf3b04b720da && git merge --no-ff dc6d72a9cdf3d1bdf2bc5f0321b0ad3b0448a26a

node scripts/docs-audit/affected-docs.mjs --json 100c394f6fd0113377bccdabed6ecf3b04b720da

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.

Advisory only, and a precision-first one (#9192): a page is listed because it names a
symbol, wire route or SDK method this diff touched — not because it mentions a changed
package. Each row says which anchor put it there, so a wrong row is reportable rather than
merely annoying. To re-verify, run the docs-accuracy-audit workflow scoped to these files:
node scripts/docs-audit/affected-docs.mjs 100c394f6fd0113377bccdabed6ecf3b04b720da → pass the list as
args.docs, on the commit named under Which tree this was computed on.