@oneharness/sdk 0.12.0 → 0.12.2

package/dist/generated/contracts.d.ts

@@ -127,8 +127,9 @@ export type ExecutionTelemetry = {
  *
  * - **Tool events.** A recorded tool call is the harness acting on the task.
  * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
- *   `signals::record_reports_work` classifies a raw harness record with, so the
- *   two readings of "billed" are one contract with one implementation.
+ *   `signals::record_work_evidence` reads a raw harness record with, and it answers
+ *   in this same type, so the two readings of "billed" are one contract with one
+ *   implementation.
  *
  * It is also a **published reading**, not only an internal one: a run that
  * failed with nothing to classify carries it as [`RunResult::work`] and in its
@@ -324,13 +325,16 @@ export interface FallbackReport {
      * [`RunResult::work`] read [`RunWork::None`] on a failure no classifier
      * recognized.
      *
-     * The chain still stops there, deliberately: re-running a task that may
-     * genuinely have failed for free would burn the next identity's quota on
-     * the same failure, which is worse than stopping (see
+     * The chain still stops there, deliberately — decided, not inherited:
+     * empty accounting is the absence of work evidence, not proof nothing was
+     * spent, so handing the candidate on could bill a task twice (the decision
+     * and its reasoning are stated once, at
      * [`crate::domain::fallback::startup_failure_reason`]). What this flag adds
      * is the *attribution* — without it a candidate that never got started reads
      * in the report exactly like one that tried the task and failed it, and the
-     * remaining candidates look untried for a reason nobody can name.
+     * remaining candidates look untried for a reason nobody can name — and the
+     * failure summary carries the candidate's own words beside it, so the
+     * reader left holding that one line can act on the cause.
      *
      * `false` whenever `ran` is `null` (no candidate ran at all — every one is
      * in `fell_through`, each with its reason).

package/dist/generated/history-line.d.ts

@@ -377,8 +377,9 @@ export type PermissionMode = "read-only" | "plan" | "default" | "edit" | "auto"
  *
  * - **Tool events.** A recorded tool call is the harness acting on the task.
  * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
- *   `signals::record_reports_work` classifies a raw harness record with, so the
- *   two readings of "billed" are one contract with one implementation.
+ *   `signals::record_work_evidence` reads a raw harness record with, and it answers
+ *   in this same type, so the two readings of "billed" are one contract with one
+ *   implementation.
  *
  * It is also a **published reading**, not only an internal one: a run that
  * failed with nothing to classify carries it as [`RunResult::work`] and in its

package/dist/generated/history-records.d.ts

@@ -1279,8 +1279,9 @@ export type Status = "ok" | "nonzero" | "timeout" | "cancelled" | "spawn-error"
  *
  * - **Tool events.** A recorded tool call is the harness acting on the task.
  * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
- *   `signals::record_reports_work` classifies a raw harness record with, so the
- *   two readings of "billed" are one contract with one implementation.
+ *   `signals::record_work_evidence` reads a raw harness record with, and it answers
+ *   in this same type, so the two readings of "billed" are one contract with one
+ *   implementation.
  *
  * It is also a **published reading**, not only an internal one: a run that
  * failed with nothing to classify carries it as [`RunResult::work`] and in its

package/dist/generated/history-stream-envelope.d.ts

@@ -1350,8 +1350,9 @@ export type Status = "ok" | "nonzero" | "timeout" | "cancelled" | "spawn-error"
  *
  * - **Tool events.** A recorded tool call is the harness acting on the task.
  * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
- *   `signals::record_reports_work` classifies a raw harness record with, so the
- *   two readings of "billed" are one contract with one implementation.
+ *   `signals::record_work_evidence` reads a raw harness record with, and it answers
+ *   in this same type, so the two readings of "billed" are one contract with one
+ *   implementation.
  *
  * It is also a **published reading**, not only an internal one: a run that
  * failed with nothing to classify carries it as [`RunResult::work`] and in its

package/dist/generated/history.d.ts

@@ -1336,8 +1336,9 @@ export type Status = "ok" | "nonzero" | "timeout" | "cancelled" | "spawn-error"
  *
  * - **Tool events.** A recorded tool call is the harness acting on the task.
  * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
- *   `signals::record_reports_work` classifies a raw harness record with, so the
- *   two readings of "billed" are one contract with one implementation.
+ *   `signals::record_work_evidence` reads a raw harness record with, and it answers
+ *   in this same type, so the two readings of "billed" are one contract with one
+ *   implementation.
  *
  * It is also a **published reading**, not only an internal one: a run that
  * failed with nothing to classify carries it as [`RunResult::work`] and in its

package/dist/generated/run-stream-envelope.d.ts

@@ -144,8 +144,9 @@ export type ExecutionTelemetry = {
  *
  * - **Tool events.** A recorded tool call is the harness acting on the task.
  * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
- *   `signals::record_reports_work` classifies a raw harness record with, so the
- *   two readings of "billed" are one contract with one implementation.
+ *   `signals::record_work_evidence` reads a raw harness record with, and it answers
+ *   in this same type, so the two readings of "billed" are one contract with one
+ *   implementation.
  *
  * It is also a **published reading**, not only an internal one: a run that
  * failed with nothing to classify carries it as [`RunResult::work`] and in its
@@ -398,13 +399,16 @@ export interface FallbackReport {
      * [`RunResult::work`] read [`RunWork::None`] on a failure no classifier
      * recognized.
      *
-     * The chain still stops there, deliberately: re-running a task that may
-     * genuinely have failed for free would burn the next identity's quota on
-     * the same failure, which is worse than stopping (see
+     * The chain still stops there, deliberately — decided, not inherited:
+     * empty accounting is the absence of work evidence, not proof nothing was
+     * spent, so handing the candidate on could bill a task twice (the decision
+     * and its reasoning are stated once, at
      * [`crate::domain::fallback::startup_failure_reason`]). What this flag adds
      * is the *attribution* — without it a candidate that never got started reads
      * in the report exactly like one that tried the task and failed it, and the
-     * remaining candidates look untried for a reason nobody can name.
+     * remaining candidates look untried for a reason nobody can name — and the
+     * failure summary carries the candidate's own words beside it, so the
+     * reader left holding that one line can act on the cause.
      *
      * `false` whenever `ran` is `null` (no candidate ran at all — every one is
      * in `fell_through`, each with its reason).

package/dist/generated/schemas.json

@@ -1945,7 +1945,7 @@
                         ]
                     },
                     "stopped_without_work": {
-                        "description": "Whether the candidate named by `ran` stopped the chain **having shown no\nevidence it did the task's work, and with nothing to say why** — its\n[`RunResult::work`] read [`RunWork::None`] on a failure no classifier\nrecognized.\n\nThe chain still stops there, deliberately: re-running a task that may\ngenuinely have failed for free would burn the next identity's quota on\nthe same failure, which is worse than stopping (see\n[`crate::domain::fallback::startup_failure_reason`]). What this flag adds\nis the *attribution* — without it a candidate that never got started reads\nin the report exactly like one that tried the task and failed it, and the\nremaining candidates look untried for a reason nobody can name.\n\n`false` whenever `ran` is `null` (no candidate ran at all — every one is\nin `fell_through`, each with its reason).",
+                        "description": "Whether the candidate named by `ran` stopped the chain **having shown no\nevidence it did the task's work, and with nothing to say why** — its\n[`RunResult::work`] read [`RunWork::None`] on a failure no classifier\nrecognized.\n\nThe chain still stops there, deliberately — decided, not inherited:\nempty accounting is the absence of work evidence, not proof nothing was\nspent, so handing the candidate on could bill a task twice (the decision\nand its reasoning are stated once, at\n[`crate::domain::fallback::startup_failure_reason`]). What this flag adds\nis the *attribution* — without it a candidate that never got started reads\nin the report exactly like one that tried the task and failed it, and the\nremaining candidates look untried for a reason nobody can name — and the\nfailure summary carries the candidate's own words beside it, so the\nreader left holding that one line can act on the cause.\n\n`false` whenever `ran` is `null` (no candidate ran at all — every one is\nin `fell_through`, each with its reason).",
                         "type": "boolean",
                         "default": false
                     }
@@ -2257,7 +2257,7 @@
                 ]
             },
             "RunWork": {
-                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_reports_work` classifies a raw harness record with, so the\n  two readings of \"billed\" are one contract with one implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
+                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_work_evidence` reads a raw harness record with, and it answers\n  in this same type, so the two readings of \"billed\" are one contract with one\n  implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
                 "oneOf": [
                     {
                         "description": "The candidate ran the task, whatever its terminal record then said.",
@@ -3016,7 +3016,7 @@
                         ]
                     },
                     "stopped_without_work": {
-                        "description": "Whether the candidate named by `ran` stopped the chain **having shown no\nevidence it did the task's work, and with nothing to say why** — its\n[`RunResult::work`] read [`RunWork::None`] on a failure no classifier\nrecognized.\n\nThe chain still stops there, deliberately: re-running a task that may\ngenuinely have failed for free would burn the next identity's quota on\nthe same failure, which is worse than stopping (see\n[`crate::domain::fallback::startup_failure_reason`]). What this flag adds\nis the *attribution* — without it a candidate that never got started reads\nin the report exactly like one that tried the task and failed it, and the\nremaining candidates look untried for a reason nobody can name.\n\n`false` whenever `ran` is `null` (no candidate ran at all — every one is\nin `fell_through`, each with its reason).",
+                        "description": "Whether the candidate named by `ran` stopped the chain **having shown no\nevidence it did the task's work, and with nothing to say why** — its\n[`RunResult::work`] read [`RunWork::None`] on a failure no classifier\nrecognized.\n\nThe chain still stops there, deliberately — decided, not inherited:\nempty accounting is the absence of work evidence, not proof nothing was\nspent, so handing the candidate on could bill a task twice (the decision\nand its reasoning are stated once, at\n[`crate::domain::fallback::startup_failure_reason`]). What this flag adds\nis the *attribution* — without it a candidate that never got started reads\nin the report exactly like one that tried the task and failed it, and the\nremaining candidates look untried for a reason nobody can name — and the\nfailure summary carries the candidate's own words beside it, so the\nreader left holding that one line can act on the cause.\n\n`false` whenever `ran` is `null` (no candidate ran at all — every one is\nin `fell_through`, each with its reason).",
                         "type": "boolean",
                         "default": false
                     }
@@ -3492,7 +3492,7 @@
                 ]
             },
             "RunWork": {
-                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_reports_work` classifies a raw harness record with, so the\n  two readings of \"billed\" are one contract with one implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
+                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_work_evidence` reads a raw harness record with, and it answers\n  in this same type, so the two readings of \"billed\" are one contract with one\n  implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
                 "oneOf": [
                     {
                         "description": "The candidate ran the task, whatever its terminal record then said.",
@@ -5955,7 +5955,7 @@
                 ]
             },
             "RunWork": {
-                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_reports_work` classifies a raw harness record with, so the\n  two readings of \"billed\" are one contract with one implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
+                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_work_evidence` reads a raw harness record with, and it answers\n  in this same type, so the two readings of \"billed\" are one contract with one\n  implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
                 "oneOf": [
                     {
                         "description": "The candidate ran the task, whatever its terminal record then said.",
@@ -8903,7 +8903,7 @@
                 ]
             },
             "RunWork": {
-                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_reports_work` classifies a raw harness record with, so the\n  two readings of \"billed\" are one contract with one implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
+                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_work_evidence` reads a raw harness record with, and it answers\n  in this same type, so the two readings of \"billed\" are one contract with one\n  implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
                 "oneOf": [
                     {
                         "description": "The candidate ran the task, whatever its terminal record then said.",
@@ -12003,7 +12003,7 @@
                 ]
             },
             "RunWork": {
-                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_reports_work` classifies a raw harness record with, so the\n  two readings of \"billed\" are one contract with one implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
+                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_work_evidence` reads a raw harness record with, and it answers\n  in this same type, so the two readings of \"billed\" are one contract with one\n  implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
                 "oneOf": [
                     {
                         "description": "The candidate ran the task, whatever its terminal record then said.",
@@ -14957,7 +14957,7 @@
                 ]
             },
             "RunWork": {
-                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_reports_work` classifies a raw harness record with, so the\n  two readings of \"billed\" are one contract with one implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
+                "description": "Whether a candidate's normalized result carries **evidence it did the task's\nwork** — the first thing [`startup_failure_reason`] consults.\n\nTwo independent witnesses, either of which is decisive:\n\n- **Tool events.** A recorded tool call is the harness acting on the task.\n- **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition\n  `signals::record_work_evidence` reads a raw harness record with, and it answers\n  in this same type, so the two readings of \"billed\" are one contract with one\n  implementation.\n\nIt is also a **published reading**, not only an internal one: a run that\nfailed with nothing to classify carries it as [`RunResult::work`] and in its\nhistory record, because there the question \"did this candidate do anything?\"\nis the only one left. One type for both, so the value a reader sees is the\nsame value the fall-through verdict consulted.\n\n[billed]: crate::domain::signals::Usage::reports_billed_work",
                 "oneOf": [
                     {
                         "description": "The candidate ran the task, whatever its terminal record then said.",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oneharness/sdk",
-  "version": "0.12.0",
+  "version": "0.12.2",
   "description": "Typed Node.js SDK for oneharness",
   "type": "module",
   "main": "dist/index.js",
@@ -19,7 +19,7 @@
     "node": ">=20"
   },
   "dependencies": {
-    "oneharness-cli": "0.12.0",
+    "oneharness-cli": "0.12.2",
     "zod": "^4.4.3"
   }
 }