# Fine-Tune Report Card — Qwen3-4B ReST Fused > Shipped: retain only as a routed file-operations specialist; reject as a general successor and do not use as the Pace default planner Canonical: https://posttrainllm.com/report-cards/qwen3-4b-rest-fused PostTrainLLM fine-tune report card · schema v1 ## Subject **Artifact:** **Artifact id:** qwen3-4b-rest-fused; **Kind:** mac-safetensors-hf; **Package dir:** specialists/qwen3-4b-rest-fused; **Path:** hf://models/posttrainllm/qwen3-4b-rest-fused; **Routing constraint:** retain only as a routed file-operations specialist; reject as a general successor and do not use as the Pace default planner. Do not use for: Pace default planner without re-distillation and its product ship gate; unqualified broad general-agent claims. — Evidence state: measured — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#verdict, specialists/registry.json#do_not_use_for; **Shipped:** true **Base model:** Qwen/Qwen3-4B-Instruct-2507 (bf16) — Evidence state: measured — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#base **Candidate model:** qwen3-4b-rest-fused — Evidence state: measured — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#id **Method:** teacher-free ReST over checker-passing interleaved trajectories with a file-ops gold depth anchor — Evidence state: measured — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#training_method **Owner goal:** Not recorded — Evidence state: missing — The specialist package format does not record the owner goal that framed the run. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json **Target:** Qwen3-4B ReST Fused — Evidence state: measured — Sources: specialists/registry.json#name ## Decision **Blocked by:** None recorded **Decision:** ship **Evidence sources:** - specialists/qwen3-4b-rest-fused/eval_report.json **Failure reason:** Not recorded **Failure reason confidence:** not-applicable **Lesson:** Not recorded **Next action:** Not recorded — Evidence state: missing — The specialist package format records no machine-readable next action. The release action lives in the public artifact registry (docs/factory/public-artifacts.md) as prose. — Sources: specialists/registry.json **Outcome label:** routed-ship **Reason:** retain only as a routed file-operations specialist; reject as a general successor and do not use as the Pace default planner **Verification blockers:** - Primary gate `file_ops_hard_gate` baseline is `historical`, not a current measurement. - Primary gate `file_ops_hard_gate` candidate is `historical`, not a current measurement. - Primary gate `file_ops_hard_gate` has no threshold value (state `missing`). - Primary gate `file_ops_hard_gate` has no passed value (state `missing`). - Frozen-eval identity is not recorded as a current measurement. - Train/eval overlap (leakage) was not checked with current evidence. **Verified:** false ## Before and after - **Baseline:** 0.75 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].stock_4b; **Candidate:** 1 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].rest_4b; **Delta:** 0.25 — Evidence state: derived — Derived from at least one non-current value; inherits the weaker provenance of its inputs. — Derived from: - specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].stock_4b - specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].rest_4b — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].stock_4b, specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].rest_4b; **Eval identity:** **Command:** Not recorded — Evidence state: missing — The specialist package format records no eval command, so this gate cannot be replayed from the report card alone. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0]; **Date:** 2026-09-03 — Evidence state: measured — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#evaluation_date; **Frozen:** Not recorded — Evidence state: missing — The package does not record whether this suite is frozen. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0]; **Suite:** file_ops_hard_gate; **Frontier ceiling:** 1 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].frontier; **Metric:** file_ops_hard_gate; **Name:** file_ops_hard_gate; **Passed:** Not recorded — Evidence state: missing — The specialist package records no ship threshold for the primary gate, so a pass/fail result cannot be derived. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0]; **Role:** primary; **Sample size:** 12 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].n; **Threshold:** Not recorded — Evidence state: missing — The specialist package format records no per-gate threshold. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0] - **Baseline:** 0.6666666666666666 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].stock_4b; **Candidate:** 0.5555555555555556 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].rest_4b; **Delta:** -0.111111 — Evidence state: derived — Derived from at least one non-current value; inherits the weaker provenance of its inputs. — Derived from: - specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].stock_4b - specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].rest_4b — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].stock_4b, specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].rest_4b; **Eval identity:** **Command:** Not recorded — Evidence state: missing — The specialist package format records no eval command, so this gate cannot be replayed from the report card alone. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1]; **Date:** 2026-09-03 — Evidence state: measured — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#evaluation_date; **Frozen:** Not recorded — Evidence state: missing — The package does not record whether this suite is frozen. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1]; **Suite:** out_of_domain_breadth; **Frontier ceiling:** 0.9777777777777777 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].frontier; **Metric:** out_of_domain_breadth; **Name:** out_of_domain_breadth; **Passed:** false — Evidence state: derived — No threshold was recorded. Derived as failing because the candidate scored below the baseline on a non-primary gate. — Derived from: - specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].stock_4b - specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].rest_4b — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].stock_4b, specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].rest_4b; **Role:** breadth; **Sample size:** 45 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1].n; **Threshold:** Not recorded — Evidence state: missing — The specialist package format records no per-gate threshold. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[1] ## Per-slice evidence None recorded ## Cost and performance **Eval time seconds:** 2011.194777167053 — Unit: s — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#performance.eval_time_seconds **Latency ms:** Not recorded — Evidence state: missing — Training duration and normalized per-request latency were not measured. Eval time is candidate depth plus breadth wall time; throughput and peak RSS are the candidate depth values. Per-suite values and raw receipt hashes are preserved in the published result. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#performance.latency_ms **Peak rss mb:** 7755.078125 — Unit: MB — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#performance.peak_rss_mb **Tokens per second:** 13.39701217776838 — Unit: tok/s — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#performance.tokens_per_second **Training cost usd:** 0 — Unit: USD — Evidence state: historical — Teacher-free local ReST iteration; no paid model API was used. Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#performance.training_cost_usd **Training time seconds:** Not recorded — Evidence state: missing — Training duration and normalized per-request latency were not measured. Eval time is candidate depth plus breadth wall time; throughput and peak RSS are the candidate depth values. Per-suite values and raw receipt hashes are preserved in the published result. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#performance.training_time_seconds ## Eval validity and leakage **Frontier ceiling:** 1 — Evidence state: historical — Recorded evidence quality: fresh-paired-frontier-qualified-result-with-hashed-raw-receipts. Imported from a committed specialist package rather than a canonical factory-run folder, so it lacks current run provenance (command, hashes, raw predictions). — Sources: specialists/qwen3-4b-rest-fused/eval_report.json#scores[0].frontier **Frozen eval:** Not recorded — Evidence state: missing — The package records no frozen held-out split identity. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json **Known limitations:** - Values come from a committed specialist package, not a canonical factory-run folder: eval commands, dataset hashes, and raw predictions are unavailable for independent replay. **Leakage:** Not recorded — Evidence state: missing — No train/eval overlap check is recorded for this package. — Sources: specialists/qwen3-4b-rest-fused/eval_report.json ## Caveats - The fresh candidate won file-ops depth 12/12 versus 9/12 and removed all eight stock side effects. - The same candidate lost frontier-qualified breadth 25/45 versus stock 30/45, so the general-successor decision is reject. - The tracked result preserves raw receipt hashes; raw traces remain local and gitignored. - Pace requires a different intent envelope and product-specific ship gate. - Do not use for: Pace default planner without re-distillation and its product ship gate - Do not use for: unqualified broad general-agent claims ## Source evidence - **Kind:** committed-package-file; **Label:** model card; **Path:** specialists/qwen3-4b-rest-fused/model_card.md; **Sha256:** 6d5fefe5efcf76f67ae84204136f36955d2dbc4043f9b0704405284867ef1890 - **Kind:** committed-package-file; **Label:** eval report; **Path:** specialists/qwen3-4b-rest-fused/eval_report.json; **Sha256:** e9924e33ad56792a89017a36e012bccf9b3703c5953a17d30e317e93d42be30c - **Kind:** committed-package-file; **Label:** reproducibility lock; **Path:** specialists/qwen3-4b-rest-fused/tinygpt.lock.json; **Sha256:** 9a699a938a99ac6f923958b770b7fcc800f3baeb047e69c65c5290cb80318edc - **Kind:** committed-package-file; **Label:** prompt contract; **Path:** specialists/qwen3-4b-rest-fused/prompt.md; **Sha256:** 1571ecc23b1327131d633dfbfbe65db047a4d09d71597994f2050360891e07b6 - **Kind:** committed-registry; **Label:** specialist registry entry; **Path:** specialists/registry.json; **Sha256:** f051c563b97adab75186d78d0adb6aa656a8cb38181d6970c877d81a572bfc6f - **Kind:** historical-record; **Label:** recorded result source; **Note:** The document the legacy score was recorded in.; **Path:** evals/verified-wins/rest-requalification-result-v1.json; **Sha256:** 368104c72ef06c66fe840fb0e61384b9e6630d3bde6e4a0c7181026ae75cf87a - **Kind:** external-artifact; **Label:** published weights; **Note:** Public weight location; not hashed by this compiler.; **Path:** hf://models/posttrainllm/qwen3-4b-rest-fused ## Compiled from **Compiler:** scripts/factory/build_fine_tune_report_card.py **Compiler version:** 1.0.0 **Dataset hashes:** None recorded **Source id:** specialists/qwen3-4b-rest-fused **Source kind:** specialist-package ## Report card id qwen3-4b-rest-fused