These examples show how Engineering Paper Skills turns rough engineering evidence into bounded manuscript prose, then keeps safety checks and provenance separate.
Illustrative examples are hand-written for readability and are marked as illustrative.
Recorded local Codex outputs are raw final messages generated from prompt fixtures and stored in the repository. They are strong public-beta evidence and top-tier candidate evidence, not top-tier-ready proof.
Manuscript Writing Examples should read like manuscript prose first. The strongest public cards are Results, Ablation, Chinese-to-English, and Methods.
Input skeleton:
84% over 180 trials; fixed-view and open-loop
baselines; overhead final occlusion, side-view depth
uncertainty, and open-loop contact error.
Short excerpt from recorded output:
In contact-rich peg insertion, the best-performing
policy is the one that prevents visual ambiguity
from becoming unrecovered contact error.Input skeleton:
17%, 23%, 44%, 76%, 88% additive rows.
Identify the main bottleneck and contribution roles.
Short excerpt from recorded output:
The largest gain appears when overlay self-
verification is added, lifting success from 44% to
76%; this delta attributes a central contribution
role to rejecting poorly grounded action hints.Input skeleton:
遮挡下 peg-hole 插入; fixed cameras lose cues;
open-loop carries pose error; full system 84% vs
69%/72%/61%; bounded evaluation.
Short excerpt from recorded output:
Occlusion during the final approach, rather than
initial visual recognition, is the central failure
mode in peg-hole insertion under fixed camera views.Input skeleton:
overhead RGB-D -> pose/confidence -> optional
additional view -> refined pose -> guarded
insertion with force and pose-deviation stops.
Short excerpt from recorded output:
Because pose uncertainty that survives into contact
is difficult to correct safely, the method first
resolves the insertion pose in perception.Six-sentence Abstract with supplied evidence only and a scope-aware final sentence.
Open abstract rerunTechnical-axis positioning with citation placeholders and the verification note kept after the draft.
Open Related Work rerunTwo paragraphs: method and evidence first, then a concrete failure boundary before platform scope.
Open Conclusion rerun2Main result, ablation, and failure-boundary paragraphs as one mini Results section.
Open full-section outputThree Methods paragraphs: overview, intermediate representation, and guarded execution path.
Open Methods section outputMinimal rough notes transformed into bounded Results prose with scope retained.
Open rough-user Results outputDe-identified author-style notes converted into bounded Results prose and missing-evidence notes.
Open realistic-note outputChinese notes reconstructed as section-specific English prose without sentence-by-sentence translation.
Open Methods outputSafety And Audit Examples are for checking claim strength, figure/table support, response truthfulness, and readiness. They are separate from the writing gallery.
Input:
Evidence: 84% success over 180 trials, one robot,
one tabletop fixture, one object family.
Draft:
The system is deployment-ready and proven robust
for industrial insertion.Illustrative audit:
Diagnosis: not ready as written.
Safe rewrite:
In the evaluated tabletop insertion setup, the
system achieved 84% success over 180 trials on one
robot arm and one object family. Broader industrial
deployment remains untested.Input:
Reviewer asks for stress-test evidence.
Draft response says:
We added Table 3, new stress tests, and final line
numbers.Illustrative audit:
Package status: revision plan only.
The draft response cannot claim added experiments,
tables, or final line numbers unless those changes
are present in the revised manuscript package.Recorded demo output for a practical notes-to-prose request remains available as a general usage example.
Open notes-to-manuscript demoRough evidence:
bad notes:
multi-view RGB-D peg insert, numbers 84 over 180,
fixed views 69/72, open loop 61. maybe view update
helps. one robot/fixture/peg family only.
Safe manuscript sentence:
In the evaluated tabletop setup, the system achieved
84% success over 180 trials and improved over the
fixed-view and open-loop baselines.
Missing evidence:
confidence intervals, statistical tests, trial split,
fixture diversity, and cross-robot evaluation.This section contains manifests, eval JSONL, human reviews, rejected outputs, and previous behavior evidence. It is for reviewers and maintainers.
Structured JSON Audit, Response Diff Verification, Inline Audit Snippets, source_span, and remaining_gap records remain available for maintainers.