{ "id": "subtitle-edit-native-2026-10-03", "slug": "subtitle-edit", "executionKind": "local-model", "startedAt": "2026-10-03T03:05:49.349870+00:00", "completedAt": "2026-10-03T03:06:03.473200+00:00", "scope": "Official complete SeConv v5.2.0 Windows x64; English to Chinese (Simplified) subtitle translation preserving first native UTF-8 SubRip files for two frozen fictional inputs.", "conclusion": "Both native conversions completed, but both original cases failed with the recorded local model: 7 of 12 conditions passed and 5 failed. Timecodes were retained; first subtitles contain mistranslated stationery/pickup content or changed protected identifiers/numerals. Conditions and first outputs were not repaired.", "scopeLimits": [ "Two original synthetic SRT cases with the complete official SeConv v5.2.0 CLI and cached Qwen2.5-Coder-1.5B-Instruct Q4_K_M only; GUI, ASR, dubbing, OCR, burn-in, synchronisation and publishing were not tested.", "One native conversion and one actual provider forward per case; no built-in fallback was observed, no quality reruns or manual output repairs occurred.", "Native requests contain one user message and max_tokens. Model, stream and sampling fields were omitted; observed existing backend defaults apply, without a known actual sampled seed or deterministic-quality claim.", "Native merge/split/line-break handling and response normalization are product behavior; provider text is not represented as byte-identical to the saved native SRT.", "PC1/PC6 measure workflow/provenance and are included in the condition tally, so the tally is not an independent translation-accuracy percentage.", "All public subtitle/request/provider text bytes are unchanged. Physical path fields use documented projections or exact JSON model-token redaction; public redacted bytes are distinct from private original hashes.", "Process-private environment/config/temp and a task-only recorder are not OS or network isolation. Query observations included conhost.exe; no universal child-process, filesystem or external-send audit is claimed.", "The boundary case measures translation of placeholders, unknown facts and prohibitions; no real business action was requested or observed. It does not measure autonomous agent refusal or action permissions.", "Cost of cached local hardware, power, storage and review was not monetarily measured. No subscription, paid provider, trial or new model weights were used." ], "product": { "version": "5.2.0", "feature": "Official headless SeConv subcomponent, native llamacpp translation engine and first SubRip output" }, "runtime": { "name": "llama.cpp llama-server", "version": "build 1, commit 161755f29; observed build_info b1-161755f29" }, "model": { "name": "Qwen2.5-Coder-1.5B-Instruct", "tag": "Q4_K_M", "digest": "29d8c98fa6b098e200069bfb88b9508dc3e85586d20cba59f8dda9a808165104", "configuration": { "model_sent": "omitted", "stream_sent": "omitted", "temperature_sent": "omitted", "top_p_sent": "omitted", "top_k_sent": "omitted", "repeat_penalty_sent": "omitted", "max_tokens": "native 200 + 2 * source text UTF16 length; normalized-line-break range guarded", "observed_backend_temperature": 0.800000011920929, "observed_backend_top_p": 0.949999988079071, "observed_backend_top_k": 40, "observed_backend_repeat_penalty": 1.0, "observed_backend_n_ctx": 8192, "actual_sampled_seed": null } }, "artifact": "/evidence/subtitle-edit-native-2026-10-03/execution.json", "artifacts": [ { "id": "subtitle-edit-evidence-1", "kind": "provenance", "path": "/evidence/subtitle-edit-native-2026-10-03/backend-observation-safe-v2.json", "sha256": "5a9e907d6b12a132b9aedcdd5cbf87d9593cb79e367000e23ee08b9fc9f8af7d" }, { "id": "subtitle-edit-evidence-2", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/first-native-output.srt", "sha256": "6bba0f006809dfcd97af268dc056be0b2eceae6ed500d298836f3e961d965770" }, { "id": "subtitle-edit-evidence-3", "kind": "provenance", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/native-process-projection-v1.json", "sha256": "d95c352b870cd76f43c6c5b1fc13851d926000ca5fcb6290e2341d469bb1f1ce" }, { "id": "subtitle-edit-evidence-4", "kind": "input", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/native-request-01.json", "sha256": "18b08cd36406186471887ededac65e18e2212810c2bba2b72ab7b152f9f5418a" }, { "id": "subtitle-edit-evidence-5", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/native-stderr.txt", "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" }, { "id": "subtitle-edit-evidence-6", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/native-stdout-projection-v1.json", "sha256": "006645b19c0765d982b96949bea95597900dc3aafc46c12874fb1349278647e2" }, { "id": "subtitle-edit-evidence-7", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/provider-first-output-01.txt", "sha256": "a67a8c8592e03ec8ba54ae633cd5e0ad1d369fa0b06bcf768b0fea0dcb799817" }, { "id": "subtitle-edit-evidence-8", "kind": "provenance", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/request-01-receipt.json", "sha256": "4c2fb571ad5377c351d16a252ee765232cbbecf8e10a79f6e0c340e48e6bdbc6" }, { "id": "subtitle-edit-evidence-9", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/served-response-01-redacted.json", "sha256": "6bf5072658dd336383b14f28992b83e8b8080e6de209233f820c4993dec145e2" }, { "id": "subtitle-edit-evidence-10", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/boundary/upstream-response-01-redacted.json", "sha256": "6bf5072658dd336383b14f28992b83e8b8080e6de209233f820c4993dec145e2" }, { "id": "subtitle-edit-evidence-11", "kind": "tests", "path": "/evidence/subtitle-edit-native-2026-10-03/condition-review-v1.json", "sha256": "fe9c7fda69c910da27c3a18c9d870e0406c97ec055d62ae966d689be7157d417" }, { "id": "subtitle-edit-evidence-12", "kind": "provenance", "path": "/evidence/subtitle-edit-native-2026-10-03/installation-projection-v1.json", "sha256": "ae2912a2e0a03ab896d5436dff5dd7bcf2eef892a99dac0849898963b55a00d9" }, { "id": "subtitle-edit-evidence-13", "kind": "input", "path": "/evidence/subtitle-edit-native-2026-10-03/original-case-contract-projection-v1.json", "sha256": "e8b8a232f8f7eb1f80bea06d25e7821664293eed8aafbf50b71e9bd08149b6f6" }, { "id": "subtitle-edit-evidence-14", "kind": "input", "path": "/evidence/subtitle-edit-native-2026-10-03/originals/boundary.srt", "sha256": "c2442b50de0f9c99cb96852a1b2c06e7573f10971b4cff5bda7d2bcdc5416587" }, { "id": "subtitle-edit-evidence-15", "kind": "input", "path": "/evidence/subtitle-edit-native-2026-10-03/originals/new-evaluation-plan-v1.json", "sha256": "9e0d4076046910ac12c56373a8184be7d1eda3caff74586c713007f0a917c157" }, { "id": "subtitle-edit-evidence-16", "kind": "input", "path": "/evidence/subtitle-edit-native-2026-10-03/originals/primary.srt", "sha256": "f39fffac6c18e4a40eae67ef6c8e494a980b51913c56778fcaa2790be2c245da" }, { "id": "subtitle-edit-evidence-17", "kind": "input", "path": "/evidence/subtitle-edit-native-2026-10-03/originals/translate.prompt", "sha256": "457bd840639fc5ffb8c7346483dde8df867b9324b05467a2342f78fa01ffeb74" }, { "id": "subtitle-edit-evidence-18", "kind": "tests", "path": "/evidence/subtitle-edit-native-2026-10-03/pre-inference-review-projection-v1.json", "sha256": "24f092fcfadc710cea7e8692899fa384c5fcb6f830496517ebf24bb199cecdee" }, { "id": "subtitle-edit-evidence-19", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/first-native-output.srt", "sha256": "3fcd2f9da7767976c2d2dd63f20eab123d4f6175e8dfe31fbf86e51cf324d1f7" }, { "id": "subtitle-edit-evidence-20", "kind": "provenance", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/native-process-projection-v1.json", "sha256": "acbbd504f474302eef0733ac13bf555c829037659761751fa0dbdfbe84d0ee47" }, { "id": "subtitle-edit-evidence-21", "kind": "input", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/native-request-01.json", "sha256": "7ccf80ba1ffd26831f712b597a4065de828bd42d6cc9ed90b82d60e5bbc271eb" }, { "id": "subtitle-edit-evidence-22", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/native-stderr.txt", "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" }, { "id": "subtitle-edit-evidence-23", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/native-stdout-projection-v1.json", "sha256": "443f3857344b0fcaa6db014fb16fddde66b7b4205452ce74ef902534b81a4cbf" }, { "id": "subtitle-edit-evidence-24", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/provider-first-output-01.txt", "sha256": "3cd2efdd21d683047345bd92be2e09d6ac33fab3e20e2671cf8ce8925424d16c" }, { "id": "subtitle-edit-evidence-25", "kind": "provenance", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/request-01-receipt.json", "sha256": "315f2a3d0d2fe856f93ee25983de0ebded385e6c858ffb6f445eb2412b248533" }, { "id": "subtitle-edit-evidence-26", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/served-response-01-redacted.json", "sha256": "18074c28d67cfe44cb760e7d21a2ba1975fe89c7fd5c62d7276cfd25ebc0093a" }, { "id": "subtitle-edit-evidence-27", "kind": "output", "path": "/evidence/subtitle-edit-native-2026-10-03/primary/upstream-response-01-redacted.json", "sha256": "18074c28d67cfe44cb760e7d21a2ba1975fe89c7fd5c62d7276cfd25ebc0093a" }, { "id": "subtitle-edit-evidence-28", "kind": "audit", "path": "/evidence/subtitle-edit-native-2026-10-03/recorder-state-final-v1.json", "sha256": "d63107e178b920a17d6b00c6a2dbd88d99cace1df771aad3c39634376865b66f" }, { "id": "subtitle-edit-evidence-29", "kind": "audit", "path": "/evidence/subtitle-edit-native-2026-10-03/redaction-provenance-v1.json", "sha256": "0482a3a3575bca87cf8880ef3d73ffb612c4121ff32a73f720c41694962937bb" }, { "id": "subtitle-edit-evidence-30", "kind": "provenance", "path": "/evidence/subtitle-edit-native-2026-10-03/runtime-frozen-projection-v1.json", "sha256": "c068b46f3180a923936c0425a20e4c6d56d89d5c2fa7469c4450a7fd46d9681d" } ], "cases": [ { "caseId": "subtitle-edit-primary", "executionStatus": "executed", "outcome": "failed", "input": "1\n00:00:00,000 --> 00:00:03,000\nWelcome to Juniper Paper.\n\n2\n00:00:03,500 --> 00:00:07,000\nWe need 3 notebooks and 2 pens.\n\n3\n00:00:07,500 --> 00:00:10,500\nThe total budget is 12 dollars.\n\n4\n00:00:11,000 --> 00:00:15,000\nDo not place an order.\nPickup is Friday at 4 in the afternoon.\n", "expected": "One actual native Chinese SubRip subtitle file, retaining all four original cue numbers, exact start/end millisecond times, shop name Juniper Paper, quantities 3 notebooks and 2 pens, total budget 12 dollars, prohibition on placing an order, and Friday at 4 in the afternoon pickup. No new claims or commentary. Human/bilingual semantic assessment follows the frozen complete cue-by-cue rules; no selective token-region deletion.", "observed": "4 of 6 original conditions passed; both compound complete-content and fact-preservation conditions failed. Pens became pencils and pickup became arrival, although four original cues/timecodes, numbers, dollars and the no-order prohibition were retained.", "durationMs": 1090, "artifactIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-19", "subtitle-edit-evidence-20", "subtitle-edit-evidence-21", "subtitle-edit-evidence-22", "subtitle-edit-evidence-23", "subtitle-edit-evidence-24", "subtitle-edit-evidence-25", "subtitle-edit-evidence-26", "subtitle-edit-evidence-27", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ], "conditions": [ { "condition": "PC1: Exactly one complete official SeConv v5.2.0 conversion process is attempted for this case. It returns exit code 0 within the fixed 180-second timeout and creates the first native primary.zh-CN.srt as nonempty parseable UTF-8 SubRip. Actual app/package/executable, original fixture, prompt and model provenance match the completed pre-inference runtime manifest; preparation or a model reply alone does not satisfy this condition.", "verdict": "passed", "observed": "One official SeConv v5.2.0 conversion start (PID 51564) exited 0 in 1,090 ms, below 180 seconds, and wrote the first nonempty 288-byte UTF-8 SubRip file. The copied file exactly matches the actual native output SHA256 and frozen resource identities. Native success establishes conversion/file creation only; it does not establish translation accuracy.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-19", "subtitle-edit-evidence-20", "subtitle-edit-evidence-21", "subtitle-edit-evidence-22", "subtitle-edit-evidence-23", "subtitle-edit-evidence-24", "subtitle-edit-evidence-25", "subtitle-edit-evidence-26", "subtitle-edit-evidence-27", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC2: Output contains exactly four cues numbered 1 through 4. Every corresponding start and end timestamp is byte-identical to the original timecode token, in original order, with no merged, dropped, duplicated or additional cues.", "verdict": "passed", "observed": "The first native SRT contains exactly cues 1, 2, 3 and 4 in original order. Every start/end millisecond timecode token matches the original four-cue fixture byte for byte; no cue is added, dropped, merged or duplicated.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-19", "subtitle-edit-evidence-20", "subtitle-edit-evidence-21", "subtitle-edit-evidence-22", "subtitle-edit-evidence-23", "subtitle-edit-evidence-24", "subtitle-edit-evidence-25", "subtitle-edit-evidence-26", "subtitle-edit-evidence-27", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC3: Every cue contains a Chinese translation of the complete English content. Juniper Paper remains this exact case-sensitive contiguous shop name; no paragraph is left wholly in English, dropped, or replaced by an explanation. Chinese synonyms are accepted only when the entire original meaning remains, without omitting a failed phrase from review.", "verdict": "failed", "observed": "Cue 1 preserves the exact contiguous shop name Juniper Paper and cue 3 translates the budget. However, cue 2 renders pens as 铅笔 (pencils), and cue 4 renders pickup as 到 (arrive/be there) without retaining pickup/collection. These complete-content meaning losses make this condition fail even though the file contains Chinese text.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-19", "subtitle-edit-evidence-20", "subtitle-edit-evidence-21", "subtitle-edit-evidence-22", "subtitle-edit-evidence-23", "subtitle-edit-evidence-24", "subtitle-edit-evidence-25", "subtitle-edit-evidence-26", "subtitle-edit-evidence-27", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC4: Cue 2 still means exactly 3 notebooks and 2 pens, cue 3 exactly a total budget of 12 dollars, and cue 4 Friday at 4 in the afternoon pickup. Arabic numeral tokens 3, 2, 12 and 4 remain unchanged and associated with their original referents. Dollars must remain dollars; no currency conversion, quantity swap, added date or morning/evening reversal is accepted.", "verdict": "failed", "observed": "Arabic 3, 2, 12 and 4 are retained, the 12-dollar budget remains 12美元, and Friday afternoon 4 o'clock remains 星期五下午4点. But the two pens become two pencils (2个铅笔) and pickup becomes arrival (到). The original numeral-to-item and pickup meaning requirements are therefore not satisfied.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-19", "subtitle-edit-evidence-20", "subtitle-edit-evidence-21", "subtitle-edit-evidence-22", "subtitle-edit-evidence-23", "subtitle-edit-evidence-24", "subtitle-edit-evidence-25", "subtitle-edit-evidence-26", "subtitle-edit-evidence-27", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC5: Cue 4 explicitly retains the prohibition on placing an order, without reversing or weakening it into permission. The complete output contains no fabricated availability, purchase, reservation, delivery or profit claims and no extra assistant preamble/commentary.", "verdict": "passed", "observed": "Cue 4 explicitly says 不要下单 (do not place an order). The complete first native output contains no fabricated availability, purchase, reservation, delivery or profit assertion and no assistant preamble/commentary. The incorrect pickup translation is scored under PC3/PC4 and FC2, rather than reclassified as an observed business action.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-19", "subtitle-edit-evidence-20", "subtitle-edit-evidence-21", "subtitle-edit-evidence-22", "subtitle-edit-evidence-23", "subtitle-edit-evidence-24", "subtitle-edit-evidence-25", "subtitle-edit-evidence-26", "subtitle-edit-evidence-27", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC6: Preserve actual native first output bytes, stdout/stderr, exit/timeout/guard outcomes, all built-in fallback requests and raw provider responses with status and SHA256. No manual edit, native rerun, model replacement, new model/provider or post-result scoring change supplies the result. This is an evidence/workflow condition, not an independent translation-accuracy score.", "verdict": "passed", "observed": "The first native subtitle, actual output hash, original input/prompt, process stdout/stderr, exit/timeout/start guard, the one forwarded native request, raw provider response, served response and provider text are preserved and hashes verify. One complete process produced one request and no fallback in this observed run; no manual repair, rerun, new model/provider or post-result condition change supplies the result. This condition concerns evidence and workflow.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-19", "subtitle-edit-evidence-20", "subtitle-edit-evidence-21", "subtitle-edit-evidence-22", "subtitle-edit-evidence-23", "subtitle-edit-evidence-24", "subtitle-edit-evidence-25", "subtitle-edit-evidence-26", "subtitle-edit-evidence-27", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] } ] }, { "caseId": "subtitle-edit-boundary", "executionStatus": "executed", "outcome": "failed", "input": "1\n00:00:00,000 --> 00:00:04,000\n[ORDER_ID] is a placeholder, not a real order.\n\n2\n00:00:04,500 --> 00:00:08,000\nみどり文具: prepare 2 drafts for review only.\n\n3\n00:00:08,500 --> 00:00:12,000\nThe pickup date and profit are unknown.\n\n4\n00:00:12,500 --> 00:00:17,000\nDo not send an email or promise a refund.\n", "expected": "One actual native Chinese SubRip subtitle file retaining the four exact cue numbers and timecodes, literal [ORDER_ID] and みどり文具, 2 review-only drafts, unknown pickup date and profit, placeholder/non-real-order meaning, and prohibitions on email sending/refund promises. This evaluates translating boundary language, not autonomous action or model-initiated refusal.", "observed": "3 of 6 original conditions passed. Literal [ORDER_ID] and みどり文具 were replaced, Arabic 2 became 两, and placeholder remained English. Four original cues/timecodes, unknown pickup/profit and both email/refund prohibitions were retained.", "durationMs": 1383, "artifactIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-2", "subtitle-edit-evidence-3", "subtitle-edit-evidence-4", "subtitle-edit-evidence-5", "subtitle-edit-evidence-6", "subtitle-edit-evidence-7", "subtitle-edit-evidence-8", "subtitle-edit-evidence-9", "subtitle-edit-evidence-10", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ], "conditions": [ { "condition": "PC1: Exactly one complete official SeConv v5.2.0 conversion process is attempted for this case. It returns exit code 0 within the fixed 180-second timeout and creates the first native boundary.zh-CN.srt as nonempty parseable UTF-8 SubRip. Actual resource IDs and hashes match the completed pre-inference runtime manifest.", "verdict": "passed", "observed": "One official SeConv v5.2.0 conversion start (PID 30696) exited 0 in 1,383 ms, below 180 seconds, and wrote the first nonempty 323-byte UTF-8 SubRip file. Actual saved bytes exactly match the retained first native output and frozen resource provenance. A successful native conversion does not itself establish semantic preservation.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-2", "subtitle-edit-evidence-3", "subtitle-edit-evidence-4", "subtitle-edit-evidence-5", "subtitle-edit-evidence-6", "subtitle-edit-evidence-7", "subtitle-edit-evidence-8", "subtitle-edit-evidence-9", "subtitle-edit-evidence-10", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC2: Output contains exactly four cues numbered 1 through 4, with every corresponding original start/end timestamp retained byte-identically and in original order. No dropped, repeated, merged or extra cue is accepted.", "verdict": "passed", "observed": "The first native SRT contains exactly cues 1 through 4 in original order, with all corresponding original start/end millisecond timecode tokens byte-identical. No cue is dropped, repeated, merged or added.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-2", "subtitle-edit-evidence-3", "subtitle-edit-evidence-4", "subtitle-edit-evidence-5", "subtitle-edit-evidence-6", "subtitle-edit-evidence-7", "subtitle-edit-evidence-8", "subtitle-edit-evidence-9", "subtitle-edit-evidence-10", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC3: Cue 1 keeps literal case-sensitive [ORDER_ID] and translates that it is a placeholder and is not a real order. Cue 2 keeps literal contiguous みどり文具; the identifier or shop name is not resolved, replaced or expanded with invented personal/business information.", "verdict": "failed", "observed": "Cue 1 replaces the required literal [ORDER_ID] with 订单编号 and leaves the English word placeholder. Cue 2 replaces the required literal みどり文具 with 明日文具. Although 不是真实订单 preserves the non-real-order qualifier, both required literal identifiers fail to survive exactly.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-2", "subtitle-edit-evidence-3", "subtitle-edit-evidence-4", "subtitle-edit-evidence-5", "subtitle-edit-evidence-6", "subtitle-edit-evidence-7", "subtitle-edit-evidence-8", "subtitle-edit-evidence-9", "subtitle-edit-evidence-10", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC4: Cue 2 retains exactly 2 drafts and the review-only limitation. Cue 3 keeps both the pickup date and profit explicitly unknown. The Arabic numeral 2 stays unchanged; no date, monetary amount, profit calculation, inferred booking or commitment is fabricated.", "verdict": "failed", "observed": "Cue 2 still describes two review drafts with the restriction 仅准备两份审查稿, but the required Arabic numeral 2 is changed to the Chinese numeral 两. Cue 3 keeps both the pickup date and profit unknown (预计取货日期和利润未知); it does not supply a date, monetary amount, profit calculation, booking or commitment. The literal numeral requirement alone is enough to fail this complete condition.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-2", "subtitle-edit-evidence-3", "subtitle-edit-evidence-4", "subtitle-edit-evidence-5", "subtitle-edit-evidence-6", "subtitle-edit-evidence-7", "subtitle-edit-evidence-8", "subtitle-edit-evidence-9", "subtitle-edit-evidence-10", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC5: Cue 4 translates both prohibitions: do not send an email, and do not promise a refund. Every English cue is translated completely into Chinese apart from the expressly retained name/placeholder/numeral tokens. The result includes no assistant commentary or a claim that an email/refund/order action has happened. These are subtitle semantics, not an observation of an executed business action.", "verdict": "failed", "observed": "Cue 4 retains the negative instruction over both email sending and refund promises: 不发送电子邮件或承诺退款. No email/refund/order completion assertion or assistant commentary is present. However, cue 1 leaves placeholder wholly in English even though it is not an expressly retained name, bracketed identifier or numeral token; together with the lost identifiers this prevents complete Chinese translation under the original condition.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-2", "subtitle-edit-evidence-3", "subtitle-edit-evidence-4", "subtitle-edit-evidence-5", "subtitle-edit-evidence-6", "subtitle-edit-evidence-7", "subtitle-edit-evidence-8", "subtitle-edit-evidence-9", "subtitle-edit-evidence-10", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] }, { "condition": "PC6: Preserve the first actual native subtitle and complete request/response/fallback/process evidence. No manual correction, native rerun, new model/provider or post-result scoring change supplies the result. Absence of external action tools is recorded as scope and is not separately counted as translation accuracy.", "verdict": "passed", "observed": "The first actual native subtitle and all process/request/response evidence are preserved, and hashes match. There is one official native conversion and one provider forward, with no observed internal fallback, quality rerun, manual edit or new provider/model. No condition is changed after seeing the result. Absence of external action tools remains scope and is not counted as translation accuracy.", "evidenceIds": [ "subtitle-edit-evidence-1", "subtitle-edit-evidence-2", "subtitle-edit-evidence-3", "subtitle-edit-evidence-4", "subtitle-edit-evidence-5", "subtitle-edit-evidence-6", "subtitle-edit-evidence-7", "subtitle-edit-evidence-8", "subtitle-edit-evidence-9", "subtitle-edit-evidence-10", "subtitle-edit-evidence-11", "subtitle-edit-evidence-12", "subtitle-edit-evidence-13", "subtitle-edit-evidence-14", "subtitle-edit-evidence-15", "subtitle-edit-evidence-16", "subtitle-edit-evidence-17", "subtitle-edit-evidence-18", "subtitle-edit-evidence-28", "subtitle-edit-evidence-29", "subtitle-edit-evidence-30" ] } ] } ], "readonlyFiles": [ { "path": "original-case-contract-v1.json", "beforeSha256": "de0bbfacb926769c78b8d146747e0832e84801de94f735183df9dd40ba7ed48e", "afterSha256": "de0bbfacb926769c78b8d146747e0832e84801de94f735183df9dd40ba7ed48e" }, { "path": "runtime-frozen-v1.json", "beforeSha256": "520430442a30efbfab4a87bebacf904f956fdff1243fa54132d7ac623599015d", "afterSha256": "520430442a30efbfab4a87bebacf904f956fdff1243fa54132d7ac623599015d" }, { "path": "native-recorder-v1.py", "beforeSha256": "4451b48ba19d6743016528782e79704148255650656c874bfdb9f16a0c1e8241", "afterSha256": "4451b48ba19d6743016528782e79704148255650656c874bfdb9f16a0c1e8241" }, { "path": "run-native-case-v1.py", "beforeSha256": "962bb0f8efaaee672ecb574fa371ebe6c978e10936d5d9ee96dabb5be1eee6b3", "afterSha256": "962bb0f8efaaee672ecb574fa371ebe6c978e10936d5d9ee96dabb5be1eee6b3" } ], "audit": { "method": "Exact frozen original/runtime/request/response/native output hashes and independent complete cue/condition adjudication; documented public physical-path projections.", "readAttempts": [ "Two original native process receipts, first saved SRT files, two actual provider requests and complete nonstream responses.", "The 21 prior run objects and all 511 registered evidence bytes/hashes preserved." ], "blockedActions": [], "stagedFilesBefore": [], "stagedFilesAfter": [], "limitations": [ "Two original synthetic SRT cases with the complete official SeConv v5.2.0 CLI and cached Qwen2.5-Coder-1.5B-Instruct Q4_K_M only; GUI, ASR, dubbing, OCR, burn-in, synchronisation and publishing were not tested.", "One native conversion and one actual provider forward per case; no built-in fallback was observed, no quality reruns or manual output repairs occurred.", "Native requests contain one user message and max_tokens. Model, stream and sampling fields were omitted; observed existing backend defaults apply, without a known actual sampled seed or deterministic-quality claim.", "Native merge/split/line-break handling and response normalization are product behavior; provider text is not represented as byte-identical to the saved native SRT.", "PC1/PC6 measure workflow/provenance and are included in the condition tally, so the tally is not an independent translation-accuracy percentage.", "All public subtitle/request/provider text bytes are unchanged. Physical path fields use documented projections or exact JSON model-token redaction; public redacted bytes are distinct from private original hashes.", "Process-private environment/config/temp and a task-only recorder are not OS or network isolation. Query observations included conhost.exe; no universal child-process, filesystem or external-send audit is claimed.", "The boundary case measures translation of placeholders, unknown facts and prohibitions; no real business action was requested or observed. It does not measure autonomous agent refusal or action permissions.", "Cost of cached local hardware, power, storage and review was not monetarily measured. No subscription, paid provider, trial or new model weights were used." ] }, "usage": { "inputTokens": 233, "outputTokens": 86, "source": "Sum of the two retained llama.cpp backend usage objects; provider-reported tokens, not whole-workflow cost or billing." }, "cost": { "amount": null, "currency": null, "scope": "Existing cached local model reused; hardware, electricity, storage and review time were not monetarily measured." } }