@agentsfleet/orly 0.13.0 → 0.14.1

package/README.md

@@ -42,9 +42,19 @@ Works with Claude Code, Codex, OpenCode, and Amp.
 
 ---
 
-## 0.13: measured judgment comparison
+## 0.14: retained Jev answers explained
 
-The 0.13.0 source candidate adds a deterministic examiner for unsupported completion claims.
+The [offline diagnosis](https://github.com/agentsfleet/orly/blob/main/evals/judgments/comparison/diagnosis/report.md) explains all 188 retained TypeSafe Jev attempts.
+It separates native-answer agreement, confidence withholding, grading disagreements and unavailable replies.
+Each attempt links to its saved inputs and a bounded explanation; unknown provider causes stay unknown.
+
+The diagnosis uses existing receipts and sends no new provider requests.
+All five proposed checks remain evaluation-only while fixture meanings and independent adoption evidence need review.
+These contributor records live in the source checkout; installed packages link to the same published source.
+
+## Measured judgment comparison
+
+The source-checkout examiner checks unsupported completion claims.
 It freezes the six runtime questions and evaluates five proposed checks separately.
 Those checks cover obligations, evidence, production wiring, failure tests and finding resolution.
 
@@ -55,7 +65,9 @@ A successful run reports `passed: true`; seeded false completions remain failed
 
 A separate explicit `jev-run.ts --live` command measures TypeSafe Jev answers against the frozen expectations.
 Its retained report lists every measured answer, grading match, disagreement and unavailable response.
-Replay reproduces every result with zero new requests. Autonomous agent improvement remains unmeasured.
+The saved replay reproduced every result with zero new requests at historical source revision [`751ee303`](https://github.com/agentsfleet/orly/tree/751ee303d0c97a403863f12f33adf8fbc20ff433).
+Replay requires that revision's unchanged source inventory; changed bytes fail admission.
+Autonomous agent improvement remains unmeasured.
 All five candidates remain evaluation-only; uncertain and failed replies stay in the report.
 An additional rehearsal uses the packed candidate and actual `agentsfleet` source in an isolated worktree.
 See the [evaluation design](docs/architecture/judgment-evaluation.md) for commands, evidence and limits.
@@ -65,7 +77,8 @@ The [0.12 report](evals/release/release-report.md) retains the earlier package a
 TypeSafe's Jev model supplies runtime advice through explicit `orly judge` requests; contributor measurement uses the separate live evaluator.
 
 See [releases](https://github.com/agentsfleet/orly/releases) for published versions.
-The install commands below fetch the published package; 0.13.0 publication is a separate owner action.
+The install commands below fetch the published package.
+The release workflow publishes 0.14.1 after its version change merges to `main`.
 
 ---
 

package/docs/CHANGELOG.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## 0.14.1 — 2026-10-11
+
+The logging standard names the field for a second failure's code.
+
+- **Role-prefixed codes** — §5 defines `cause_error_code`, `rollback_error_code` and `config_error_code`, and the `<role>_error_code` shape for new roles.
+- **Rust binding** — §8A points a record that reports a second failure at §5.
+
+## 0.14.0 — 2026-10-08
+
+Retained TypeSafe Jev measurements now carry per-attempt explanations and explicit adoption limits.
+
+- **Diagnosis** — accounts for all 188 saved attempts and separates native-answer agreement from confidence withholding and grading.
+- **Evidence** — preserves original replies and labels; names uncertain fixture meanings and unavailable provider explanations.
+- **Guidance** — links contributors to the source diagnosis while keeping all five proposed checks evaluation-only.
+
 ## 0.13.0 — 2026-10-07
 
 A source-checkout examiner checks unsupported completion claims and grades actual TypeSafe Jev answers against frozen expectations.

package/docs/LOGGING_STANDARD.md

@@ -184,6 +184,21 @@ Embed the code as a struct field on the logger call: `log.err("event_name", .{ .
 
 System-level failures with no domain meaning (e.g. raw `EACCES` from a syscall before we attribute it to a tenant operation) emit without `error_code`. The follow-up rule: if a syscall failure surfaces to a user, it gets attributed to a registry code at the boundary.
 
+**One record, two failures.** Some records report a decision about another
+failure: a gate that fails open, a rollback, a refused configuration. The
+record's `error_code` names the decision. The failure behind it keeps its own
+registry code in a field named for its role:
+
+| Field | Carries the code of |
+|---|---|
+| `cause_error_code` | the failure the decision answers |
+| `rollback_error_code` | a failure in the recovery that followed |
+| `config_error_code` | the configuration read that could not be used |
+
+A new role takes the same `<role>_error_code` shape. `error_code` always names
+the record's own code, so never overwrite it with the cause's code. Never
+embed a code inside a reason sentence either; the field carries it.
+
 ## §6 · PII / secret discipline
 
 > [UNENFORCED → broad principle; no scan covers every log call site in the codebase for credential-shaped values]
@@ -324,12 +339,14 @@ arguments, never positional message formatting.
 
 > [JUDGMENT → RUST-STYLE]
 
-Every `warn` or `error` that maps to a registry code supplies
-`error_code = code` per §5. Hoist field expressions into locals before the macro
-call. The log bridge duplicates field expressions, and LLVM coverage
-instrumentation scores the dead copy; locals preserve evaluation and coverage
-behavior. A trailing string message is §3's optional `msg` and is always the
-last macro argument. Prefer fields alone when they already explain the event.
+Every `warn` or `error` that maps to a registry code supplies `error_code =
+code` per §5. A record that reports a second failure adds its
+`<role>_error_code` per §5, for example `cause_error_code`. Hoist field
+expressions into locals before the macro call. The log bridge duplicates field
+expressions, and LLVM coverage instrumentation scores the dead copy; locals
+preserve evaluation and coverage behavior. A trailing string message is §3's
+optional `msg` and is always the last macro argument. Prefer fields alone when
+they already explain the event.
 
 ```rust
 let code = error_code::INTERNAL_DB_QUERY.as_str();

package/docs/RULE_ENFORCEMENT.md

@@ -6,7 +6,7 @@ regenerates this file and fails on any difference. To update it, run
 
 | Rule document | enforced by | judged | acknowledged | classified | trigger |
 |---|---|--:|--:|--:|---|
-| `docs/LOGGING_STANDARD.md` | `LOG` → `logging.sh` | 10 | 13 | 28/28 | mechanical |
+| `docs/LOGGING_STANDARD.md` | `LOG` → `logging.sh` | 10 | 13 | 30/30 | mechanical |
 | `docs/REST_API_DESIGN_GUIDELINES.md` | — | 0 | 0 | 0/67 | mechanical |
 | `docs/SCHEMA_CONVENTIONS.md` | — | 0 | 0 | 0/17 | mechanical |
 | `docs/DOCUMENTATION_RULES.md` | — | 0 | 0 | 0/22 | latent |
@@ -14,7 +14,7 @@ regenerates this file and fails on any difference. To update it, run
 | `docs/CHANGELOG_VOICE.md` | — | 0 | 0 | 0/3 | latent |
 | `docs/VERIFY_TIERS.md` | — | 0 | 0 | 0/4 | uncited |
 | `docs/greptile-learnings/RULES.md` | — | 0 | 0 | 0/115 | mechanical |
-| **corpus** | | | | **28/278** | |
+| **corpus** | | | | **30/280** | |
 
 ## How to read this
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agentsfleet/orly",
-  "version": "0.13.0",
+  "version": "0.14.1",
   "description": "Repository-scoped engineering harness: renders agent rules, materialises the gates that enforce them, and proves the PR boundary \u2014 for any coding agent.",
   "license": "MIT",
   "type": "module",