feat(agent): recover once from headless response truncation

benchmarks/harbor/configs/truncation-recovery-20261005/README.md

@@ -0,0 +1,40 @@
+# Bounded truncation recovery experiment
+
+The [predeclared plan](experiment.json) selects the two failures from the
+[v0.9.0 pilot20 baseline](../../reports/v0.9.0-baseline-20261005.md), with their
+original revisions and budgets. Each task receives one fresh model run; no
+score replaces a trial in the released baseline.
+
+Use a checkout of agent commit
+`e1be6ed1f81a17d73677b414d4d913adb81e9e25` for the recorded behavior and wheel
+version. The standard runner builds its current checkout, including local
+source changes. Configure the official DeepSeek credentials as described in
+the [adapter README](../../README.md), set
+`ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic`, and leave
+`ANTHROPIC_MODEL` unset so the configuration chooses `deepseek-flash`.
+
+Run dependency installation preflight first:
+
+```bash
+uv run --project benchmarks/harbor python -m harbor_adapter run \
+  --config benchmarks/harbor/configs/truncation-recovery-20261005/install-only.json \
+  --install-only
+```
+
+Both tasks lack a registered verifier dependency profile at this revision and
+must explicitly report `not_configured`. After agent installation succeeds,
+run the following two commands in separate terminals to reproduce the original
+concurrent task pairing. Omit `--job-name` to generate fresh job names.
+
+```bash
+uv run --project benchmarks/harbor python -m harbor_adapter run \
+  --config benchmarks/harbor/configs/truncation-recovery-20261005/model-extraction.json
+uv run --project benchmarks/harbor python -m harbor_adapter run \
+  --config benchmarks/harbor/configs/truncation-recovery-20261005/scheme.json
+```
+
+Record both task scores and actual recovery events. A task that passes without
+reaching `max_tokens` does not demonstrate recovery effectiveness. The
+[experiment report](../../reports/truncation-recovery-20261005.md) and
+[structured results](../../results/tb21-truncation-recovery-20261005.json)
+preserve that distinction.

benchmarks/harbor/configs/truncation-recovery-20261005/experiment.json

@@ -0,0 +1,150 @@
+{
+  "created_at": "2026-10-05T12:53:39.916356+00:00",
+  "question": "Can one bounded headless continuation after max_tokens preserve useful work and improve the two failed tasks without changing their budgets?",
+  "baseline_results": "benchmarks/harbor/results/tb21-v090-baseline-20261005.json",
+  "baseline_agent_ref": "c0605369d277352ac0d5ec20d02017b7cb149314",
+  "dataset_ref": "sha256:7d7bdc1cbedad549fc1140404bd4dc45e5fd0ea7c4186773687d177ad3a0699a",
+  "model": "deepseek/deepseek-flash",
+  "endpoint": "https://api.deepseek.com/anthropic",
+  "change": "One automatic truncation recovery per headless run. Keep the token cap and original deadline and turn count; skip truncated tools and incomplete thinking.",
+  "max_turns": 100,
+  "max_tokens": 65536,
+  "agent_setup_timeout_seconds": 1800,
+  "attempts_per_task": 1,
+  "whole_trial_retries": 0,
+  "execution": "Preflight both tasks, then launch the two disjoint single-task standard jobs together, at most two simultaneous trials. Each job retains n_concurrent_trials=2 but contains one task.",
+  "verifier_policy": "Original per-attempt limit; only verifier timeouts may retry, at most three attempts on the same agent output.",
+  "preflight": "Both task profiles are unsupported and must be recorded as not_configured; installation success does not certify verifier dependencies.",
+  "analysis": "Preserve every result. Count actual max_tokens responses and truncation_recovery injections. If no truncation occurs, a pass does not demonstrate recovery effectiveness. Report setup/verifier exceptions separately from scored failures. Do not replace the released pilot20 baseline or extrapolate a full-pilot score.",
+  "tasks": [
+    {
+      "name": "terminal-bench/model-extraction-relu-logits",
+      "ref": "sha256:1ae5045ad68b5d34c3398b612066a07c4a08b6dc330d28868ec4021e17c94b17",
+      "native_agent_timeout_seconds": 900,
+      "work_budget_seconds": 720,
+      "verifier_timeout_seconds": 900.0,
+      "docker_image": "alexgshaw/model-extraction-relu-logits:20251031",
+      "docker_image_id": "sha256:52fe1f089f38650f0dc22d7531ee6e01ebd526de1b349aaa2ecb331dca9fabce",
+      "docker_repo_digests": [
+        "alexgshaw/model-extraction-relu-logits@sha256:52fe1f089f38650f0dc22d7531ee6e01ebd526de1b349aaa2ecb331dca9fabce"
+      ],
+      "task_file_sha256": {
+        "instruction.md": "c9e179dd099065a34c362a865b0eb9fab5f7d96be582cf117e7519580c036ebe",
+        "task.toml": "eb64dfaf8ae88868911ae28855cdc4f767de464001e2704919015b595be6dc29",
+        "tests/test.sh": "3b2d8f5e0353c2bab3515d6387468a863a4f9f241ee131c73f545f3fdf9e2d52",
+        "tests/test_outputs.py": "e5189c7eb9942ede55e1b48a80388590b80a486b75497613f257b3373ad9b2c9"
+      }
+    },
+    {
+      "name": "terminal-bench/schemelike-metacircular-eval",
+      "ref": "sha256:58130c2166c3115276dc8592f358e326ff2d81ea852e3d88636c82fd1dff57e6",
+      "native_agent_timeout_seconds": 2400,
+      "work_budget_seconds": 2220,
+      "verifier_timeout_seconds": 2400.0,
+      "docker_image": "alexgshaw/schemelike-metacircular-eval:20251031",
+      "docker_image_id": "sha256:b485be38bf80d56c1a4cd9da95bbf7ec129dc03f6786b86c0525518a96f6d1db",
+      "docker_repo_digests": [
+        "alexgshaw/schemelike-metacircular-eval@sha256:b485be38bf80d56c1a4cd9da95bbf7ec129dc03f6786b86c0525518a96f6d1db"
+      ],
+      "task_file_sha256": {
+        "instruction.md": "3af5f9567b62f93683db7518ec6fcb609f83796258aa02166b5a7091c5b69031",
+        "task.toml": "8dae8eb226183b0ca92331c1106152a7abc93b03b4e5a95291b86a4ddc9b44e0",
+        "tests/interp.py": "efb8b5cdcf78b73afcf39d59600a1f355ca89b7bac9854bb01f110cc77bea2c1",
+        "tests/shadow_test/01-factorial.scm": "540e81fac476f04dc510ba107721d606ac980760088b2136889e7d4bf884e5c5",
+        "tests/shadow_test/02-fibonacci.scm": "628957171e0c4a3ee373635923fcd29e75bdf4bf13a402ac26fe94a5fc1443ac",
+        "tests/shadow_test/03-list-operations.scm": "fe8f17f59fe4045582fce363d52df95ba7c4de7345542c8bd4993a9f2f396235",
+        "tests/shadow_test/04-higher-order.scm": "9c2eba8c91bb7b224de56d789207d2eaad98bbe1f8c360d6eca4f014f42f9230",
+        "tests/shadow_test/05-simple-io.scm": "1eba30db30824fcf6048ea8e49b84e85c866cefce30862fce49b300e4416843c",
+        "tests/shadow_test/06-interactive-io.scm": "4fbc0e2d6e42c5a9495e549ccb13c02fcaeddf8fdaf8d2cf5de833116f12610a",
+        "tests/shadow_test/08-progn-sequencing.scm": "3bf8236dc5356d29ab8dbfe1b0f4c519354fbb23c8f3eaf946779c28b80feea3",
+        "tests/shadow_test/09-mutual-recursion.scm": "3bfa1f77e602c7039f8c8c70315e6d4fff6eed737dab0f0398e5bdf690137c2c",
+        "tests/shadow_test/10-advanced-features.scm": "b69ef497088fcf0aec388789a17561b783b8cdd5915719915a51c5ba189c31ff",
+        "tests/shadow_test/accumulator_patterns.scm": "8b9192796a6f9c3155cc5e44d0a0ba6a46d20a0d94e265b69b23ec8af43e4a07",
+        "tests/shadow_test/binary_tree.scm": "5a3dafc8844969cebb5809f253f4e54de7df480f378da8197f870badef03ee47",
+        "tests/shadow_test/church_numerals.scm": "31d075785f2f605bac3bd9ac2d99fcf7eaaed6686eb9afeab1d40bf5f35bbf2c",
+        "tests/shadow_test/closures.scm": "c7d025a376987f5ee22f529847b66eaf53f0af7033acbb4055cedb734168d612",
+        "tests/shadow_test/continuation_passing.scm": "79e7445c42227846f1adfaf08c17e278aac7a34df77482900b068be4fd248080",
+        "tests/shadow_test/currying.scm": "690a8cccbb1d249c8f6fcabc0193cecc5d0e4c1af39a0bb67c7e95fcff8e393e",
+        "tests/shadow_test/filter_operations.scm": "485a0563b81d002e368c5e4b2b7cc98a5306163701053630f416e3b783a2a238",
+        "tests/shadow_test/fold_operations.scm": "461b33db6b3f5580122d4c6dff0fd6d11c7f044865ffd684ade67d9b296a5a24",
+        "tests/shadow_test/function_composition.scm": "65ceee00188b12c3549134449e2f938d010c2c97dba090f5b53244abb3579b1f",
+        "tests/shadow_test/lazy_evaluation.scm": "edd95748de14aadf916c65209b5bcb8759d3e33b7ec935289a5128a656d33bec",
+        "tests/shadow_test/list_operations.scm": "660963851ecb49639f66557bd88db1e73db323186742145f33cb2e14d3af1103",
+        "tests/shadow_test/map_operations.scm": "974ac42d892f4b62b9b21ef7844581624dad811b38d35944f1005cc47f249856",
+        "tests/shadow_test/memoization.scm": "25482c1912fab1235651b1b7acd727f2ea6a55dc5c1919f841b44cfc32107b05",
+        "tests/shadow_test/mutual_recursion.scm": "2fbc9562e863cb2b325f86419fb2a7588265aaf5576295550722ee4274663712",
+        "tests/shadow_test/nested_defines.scm": "23ad3b2f936949501466da784511ea2e6215e2b198ca371bef86f6af877ca923",
+        "tests/shadow_test/oeis_sequences.scm": "1b586df6af2429a54808f7587f60021ba2b2c9bcc2ef2f9cbe44339699853c62",
+        "tests/shadow_test/oeis_sequences2.scm": "300da0b0dc20b883e01142adea9eb6da1a583a1b969ac0be47aa32c1ab66340e",
+        "tests/shadow_test/oeis_sequences3.scm": "dea198e874bf518b93c8406a843b64c1586cf4ce780fec8c444f8343e9322c75",
+        "tests/shadow_test/recursive_structures.scm": "64f1a430bf04a627f5e27bf004805544b60406f8a123d186516774dfcf3846cd",
+        "tests/shadow_test/test_read.scm": "fda448c22685e2f9af37ff7b986ee5e906d485a2db16f62eec113149f866870f",
+        "tests/shadow_test/variadic_functions.scm": "ac940ac0273e87ee06280fee192194ab5cdc14c88b61da9cfba92105e66b8f2f",
+        "tests/shadow_test/y_combinator.scm": "38d25b02d54342ab5610ad03fcef68c5f1d4eef8902295645f6acceb853c9860",
+        "tests/test/01-factorial.scm": "2623000a742c63d50ae1851854826247caee129c7d1613be6c7131d21f26a6ad",
+        "tests/test/02-fibonacci.scm": "db769cef446d5da9e9b1abccdf9ca644e239938095a98c21551aee25255a57d4",
+        "tests/test/03-list-operations.scm": "1a14b4100dedef30010fed62ea9fd295add1b0facb2e8e6e27c8d8eb3d90111b",
+        "tests/test/04-higher-order.scm": "59cd17bbffad641a37729b17c5bf9178b8bae7289fe89cc7135eb0bfec2b4fd6",
+        "tests/test/05-simple-io.scm": "3deeb3275df1b771449ebaebfcb9a934f375415e138c5e9f3685516c69a1f7a0",
+        "tests/test/06-interactive-io.scm": "aa3cae79b79dca90ac9d61a1babf64d6a5bb6559798c7727c11be02d8ddfd216",
+        "tests/test/08-progn-sequencing.scm": "4c0a4d980cf4223d1d17715d58adbe27dc120eced257a0d9f42de952c094bc5b",
+        "tests/test/09-mutual-recursion.scm": "939e18f4bd25b5ab0e2a4a767bf7eb6b0d8fb3530d28bf236f06cf27f2d90ee8",
+        "tests/test/10-advanced-features.scm": "c4cc5fb412df363875768c6a1e645255f6bc8f9ddf988d1ec2f4a266cedb8ec7",
+        "tests/test/README.md": "1f2174c7f12721f9f1898cda190c2886fb3ef9d35fd085f315da5814657104ae",
+        "tests/test/accumulator_patterns.scm": "50d246e00577d1f3ab3f0462c2b9618ea63680c33fcca12369cd1707b010cb5f",
+        "tests/test/binary_tree.scm": "c95f59ccac1ea4a9c846afddc993aa40ac081c0f9e6731544530109b269b67ae",
+        "tests/test/calculator.scm": "a09213ec8a2ec52486b9346c420ddb7c537576ef71b1cc5c6ec61d13d9139f61",
+        "tests/test/church_numerals.scm": "f37d858f32a9c5ba8868d094018c92516acadf0ea4e55b7c36f14320cb0c69a6",
+        "tests/test/closures.scm": "3ecfc35289479288876e20bf7f55e5b8eda84acd35937e89e3d062308a99aeff",
+        "tests/test/continuation_passing.scm": "01049581e582f7d35ef40ef68e1827d012da619beea57c1c74675a5f245318d7",
+        "tests/test/currying.scm": "2844d54eee9a94af05a11b4aae06d6e7823d193f17235c8f90bf98926cbca75d",
+        "tests/test/filter_operations.scm": "f6bc75f6da9ece96c0616e3570e4df59d0f2eb4655375d3a46b0349c86cc5dd8",
+        "tests/test/fold_operations.scm": "05aab492ad914baecd19cf454ed587e7a031e94b8990500eb1c869c15da6fc0b",
+        "tests/test/function_composition.scm": "9b56dc78823e7d35adb213744c0f8e95684074d6a12bb98c8c05755788be5b06",
+        "tests/test/lazy_evaluation.scm": "ea5912a53a35b65bf2bee9483dca90814b8baf52ae3ae50c2d4589f794dca54d",
+        "tests/test/list_operations.scm": "d7e70621afbbb787cf646e2089bccfbd3681fbb2e70325f3e50f56ccd30dbd9d",
+        "tests/test/map_operations.scm": "0cebddb3d4a50e4b6f6bfc0e8c12762d3f42609030aef077d8ab4bedda4718d9",
+        "tests/test/memoization.scm": "cf8202df22407dbba72ca7f660d8ab7cd4ca962b215c36cc491d5c1a474c160b",
+        "tests/test/mutual_recursion.scm": "a3dc23a96ec1cf05b0be4a4f0b6d3da9e2bad1aa3f7a83c7566f71f980a67fe2",
+        "tests/test/nested_defines.scm": "21faa0ccaf744cb8c9d283bbfd0c36e195c803c85f511e70a99267c0a85b6216",
+        "tests/test/oeis_sequences.scm": "d1f77161b367176b3dfab55a34466e511c602c7ae359e05f5a715627b93dca6a",
+        "tests/test/oeis_sequences2.scm": "5679ef0f99743114d8c06edcdcde1fea9474b71aeb57c2069f1ee33df1970bce",
+        "tests/test/oeis_sequences3.scm": "5dc8b44158b2648a6e9766f6bcdf21553a07bf487d79e64ebb418511ef3b8186",
+        "tests/test/recursive_structures.scm": "67838ff5bac26ea359139720adec1d5700176147b0062065c7def9715e8a2aeb",
+        "tests/test/test_read.scm": "5b9ee381c3cf45b2fc8c2e7b490857fb67f1ac78653b586d7f2848daf6853809",
+        "tests/test/variadic_functions.scm": "c9b847c50840c6977904e5033a07f6108cc3a54012b02d916e79025fb43e0f6e",
+        "tests/test/y_combinator.scm": "971144dcb2b04aaa3b34d9279459bd7d2e8256c585d17809465848c2589d36d8",
+        "tests/test.sh": "4770437ea96c3cc84684b4f99d55fb148fcac09f9ea1e8ef49de487716e6c334",
+        "tests/test_outputs.py": "fbb84e3c0c2f72ed7de90f54380f860052c16155e7d0c9076cd57c966e9be7ed"
+      }
+    }
+  ],
+  "source_sha256": {
+    "src/nanopycodeagent/__init__.py": "b922299f65b3db2fb1514df5aeda4413e599f595d2595f45a353b4dc551a1eb1",
+    "src/nanopycodeagent/agent.py": "8e28a5c3fcd140441484c46d1e62ec811d325aaff484141a4b9837baead6fc52",
+    "src/nanopycodeagent/atif.py": "ca4793b971d125d904888b858452db8db30e21f0d9b87818133062fb6cebe270",
+    "src/nanopycodeagent/bash_tool.py": "dcf0366fb9ebdca258d4c40b1a50b78b2ec79fc4b85cbb253ac9a9ce9534574f",
+    "src/nanopycodeagent/cli.py": "e335789b3f09440abd441547a2e3e2feb32b90c68e9ba5fc8237de8f6bac2932",
+    "src/nanopycodeagent/cost.py": "30a8df17942c424796e8532f45e1cb0784ed6541cea44284b863db3dae2f7b70",
+    "src/nanopycodeagent/deadline.py": "26a861721e6897a4e02e6ce364cb3ef1f1dc7d4ad9fb2f772404276f5c9e9d5f",
+    "src/nanopycodeagent/edit_tool.py": "78e2bec911279082b081366ab05cb4cbb37d8910b20b6245cccd4b0424489d0d",
+    "src/nanopycodeagent/event_journal.py": "8974919c2d460afb52bc57ec37e7e271851290c0845d01b4ad0f9f760cde0814",
+    "src/nanopycodeagent/read_tool.py": "e0075da19f2e1c6ccf34a3056a0aec92398e5afe67869937a096c6f1e04181c3",
+    "src/nanopycodeagent/settings.py": "491bf5c6eca50eb4c5fe04b7bfe62a8bfc7d2e1fad05b8eeac1680c2fcf3674e",
+    "src/nanopycodeagent/terminal.py": "8d27b40976b2b6d476e5106dfea982b055115435b4ba16b7f758412a05b75f04",
+    "src/nanopycodeagent/tool_validation.py": "3a1a85a7222dd1504bdf58f37e65a51d11de13d16c4e7df42c72b4cd58b5e5e5",
+    "src/nanopycodeagent/transport.py": "152d3ff78be0e3b2d92e0abbbfb652d73466aff57ad316c06dc0e072411167fe",
+    "src/nanopycodeagent/write_tool.py": "7671e83464fa1e942f25742fc573d0712e32053b5fc0291257884b027f7f6206",
+    "benchmarks/harbor/src/harbor_adapter/__init__.py": "513ad49c721b3e73293095c6a0486a791483e9aa2b51e0e9ad4b7816af07e76b",
+    "benchmarks/harbor/src/harbor_adapter/__main__.py": "db89d582fba45aad72147317f595531af49c434e3b23c65f22d7491f0b10551d",
+    "benchmarks/harbor/src/harbor_adapter/_verifier_runner.py": "2eb73c3e76d7360f2fa16018f0e933797bf982e589236095b5f98750f3807a7c",
+    "benchmarks/harbor/src/harbor_adapter/adapter.py": "d1aa154ec6972d706f0725cfa69c56635506d4499dff2a129e2c957ea7034c02",
+    "benchmarks/harbor/src/harbor_adapter/bootstrap.py": "f499bfddfc3d29338cae15a7f8f6e9c66d19eb69ad9577e607ced4b1aa057be2",
+    "benchmarks/harbor/src/harbor_adapter/verifier.py": "ec26fdefd69be1c5c450c9ba69d749d4a9d320a2a170b8db34421f0573d9b493"
+  },
+  "configuration_sha256": {
+    "model-extraction.json": "f8baf52e182e6b6ab669644c904c1c961db5e6083c6c18b8d1479932a9b26588",
+    "scheme.json": "0bf57205bb0d82ea0342483a269eb995441a7c427fec40f0678f3a69b1a87b0b",
+    "install-only.json": "75e6b3f9d567c1c2db953deef82cc43ec8ea31c3c79ab15fa11599c68531aaaf"
+  }
+}

benchmarks/harbor/configs/truncation-recovery-20261005/install-only.json

@@ -0,0 +1,27 @@
+{
+  "n_concurrent_trials": 2,
+  "n_attempts": 1,
+  "agents": [
+    {
+      "name": "harbor_adapter:NanoPyCodeAgent",
+      "model_name": "deepseek/deepseek-flash",
+      "override_setup_timeout_sec": 1800,
+      "kwargs": {
+        "max_turns": 100,
+        "max_tokens": 65536,
+        "time_budget_seconds": 720
+      }
+    }
+  ],
+  "tasks": [
+    {
+      "name": "terminal-bench/model-extraction-relu-logits",
+      "ref": "sha256:1ae5045ad68b5d34c3398b612066a07c4a08b6dc330d28868ec4021e17c94b17"
+    },
+    {
+      "name": "terminal-bench/schemelike-metacircular-eval",
+      "ref": "sha256:58130c2166c3115276dc8592f358e326ff2d81ea852e3d88636c82fd1dff57e6"
+    }
+  ],
+  "install_only": true
+}

benchmarks/harbor/configs/truncation-recovery-20261005/model-extraction.json

@@ -0,0 +1,22 @@
+{
+  "n_concurrent_trials": 2,
+  "n_attempts": 1,
+  "agents": [
+    {
+      "name": "harbor_adapter:NanoPyCodeAgent",
+      "model_name": "deepseek/deepseek-flash",
+      "override_setup_timeout_sec": 1800,
+      "kwargs": {
+        "max_turns": 100,
+        "max_tokens": 65536,
+        "time_budget_seconds": 720
+      }
+    }
+  ],
+  "tasks": [
+    {
+      "name": "terminal-bench/model-extraction-relu-logits",
+      "ref": "sha256:1ae5045ad68b5d34c3398b612066a07c4a08b6dc330d28868ec4021e17c94b17"
+    }
+  ]
+}

benchmarks/harbor/configs/truncation-recovery-20261005/scheme.json

@@ -0,0 +1,22 @@
+{
+  "n_concurrent_trials": 2,
+  "n_attempts": 1,
+  "agents": [
+    {
+      "name": "harbor_adapter:NanoPyCodeAgent",
+      "model_name": "deepseek/deepseek-flash",
+      "override_setup_timeout_sec": 1800,
+      "kwargs": {
+        "max_turns": 100,
+        "max_tokens": 65536,
+        "time_budget_seconds": 2220
+      }
+    }
+  ],
+  "tasks": [
+    {
+      "name": "terminal-bench/schemelike-metacircular-eval",
+      "ref": "sha256:58130c2166c3115276dc8592f358e326ff2d81ea852e3d88636c82fd1dff57e6"
+    }
+  ]
+}

benchmarks/harbor/configs/truncation-recovery-repeat3-20261005/README.md

@@ -0,0 +1,45 @@
+# Three additional runs per task
+
+This [predeclared plan](experiment.json) follows the earlier
+[two-task experiment](../../reports/truncation-recovery-20261005.md), which
+passed both tasks without exercising truncation recovery. Run each task three
+more times from scratch, preserving all six results. These are additional
+samples, not replacements for earlier scores.
+
+Keep the same agent/adapter Python source hashes, task revisions, images, model,
+endpoint, token cap, reply cap, and work/setup/verifier budgets as the preceding
+experiment. Runtime behavior is from commit
+`e1be6ed1f81a17d73677b414d4d913adb81e9e25`; the runner records the exact checkout
+commit and built wheel for this round.
+
+From the repository root, configure official DeepSeek credentials, set
+`ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic`, and leave
+`ANTHROPIC_MODEL` unset. Run installation preflight first:
+
+```bash
+uv run --project benchmarks/harbor python -m harbor_adapter run \
+  --config benchmarks/harbor/configs/truncation-recovery-repeat3-20261005/install-only.json \
+  --install-only
+```
+
+Both verifier dependency profiles must explicitly report `not_configured`.
+After installation succeeds, run the six entries in `job_queue`, taking each
+configuration path relative to this directory. Limit concurrent jobs to two:
+each configuration contains one task and one attempt. For example:
+
+```bash
+uv run --project benchmarks/harbor python -m harbor_adapter run \
+  --config benchmarks/harbor/configs/truncation-recovery-repeat3-20261005/scheme.json \
+  --job-name truncation-recovery-repeat3-scheme-1-20261005
+```
+
+Use fresh job names when reproducing the experiment; the runner never overwrites
+existing results. No whole-trial retries or replacements are allowed. Only
+verifier timeouts may retry, at most three attempts with the original per-attempt
+limit and the same agent output. Record truncation, actual recovery, subsequent
+tool execution, terminal outcome, score, and accounting completeness per trial.
+A passing trial without truncation does not demonstrate recovery effectiveness.
+
+The recorded run used checkout `382cb38413d7d5368ad1be97e85142f280172a5b`.
+See the [six-run report](../../reports/truncation-recovery-repeat3-20261005.md)
+and [structured results](../../results/tb21-truncation-recovery-repeat3-20261005.json).