report-final-deliverables

name	report-final-deliverables
description	Use when writing, revising, or reviewing PRML VSLAM final report or slide deliverables under docs/report or docs/slides, including Typst figures, tables, citations, bibliography references, scientific claims, VSLAM metrics, alignment, method or dataset framing, protocol text, limitations, result statements, and final presentation prose.

Report Final Deliverables

Use this as the compact front door for final report and final slide work. Keep the hot path short: route to the owning source files, verify claims, and avoid copying external workflow trees into this repository.

Grounding

Start every report or slide task from the local sources of truth:

1. Read docs/AGENTS.md and the target Typst entry point or fragment.
2. For scope, challenge framing, or claim decisions, read docs/Questions.md and use the final-paper guidance below.
3. For citations, inspect docs/references.bib. For method, dataset, or protocol facts, prefer package README.md or REQUIREMENTS.md, tracked artifacts, and primary papers or official sources.

Do not use tests, app UI files, branch names, PR numbers, or raw chat text as manuscript evidence. Tests can verify behavior, but they are not report references.

Final Paper Guidance

Write the report as a benchmark-methodology paper, not as a general leaderboard. The defensible contribution is the measurement substrate: source normalization, method adapters, frame and scale semantics, alignment protocol, persisted artifacts, manifests, and diagnostic visualization. Local result tables may describe the available artifact sweep, but broad method-ranking claims require a frozen method-dataset matrix, sampling policy, and archived metric artifacts.

Architecture belongs in the main text only when it explains a scientific consequence, such as artifact interpretation, frame semantics, alignment, or metric admissibility. Detailed implementation diagrams belong in the appendix.

Before adding or revising a result sentence, verify that the claimed run exists as a frozen artifact with configuration, normalized input, method output, alignment metadata, and metric CSV or JSON files. Each result row should make the dataset, sequence set, method, sampling policy, trajectory reference, alignment mode, cloud reference, ICP threshold, hardware or runtime context, and artifact provenance recoverable from the manuscript or appendix.

Figures and tables must have self-contained captions that state what produced the display item and whether it is a protocol checklist, local result summary, or supplementary architecture diagram.

Professor/Moodle Deliverable Gate

Final report work must cover the stable Moodle/professor content requirements without depending on ignored .agents/work/ course-material files being present in every clone:

• Domain Introduction & Goals
• Methodology
• Results
• Future Work
• Retrospective: what went right, what went wrong
• Work breakdown for the report and the code

Map these requirements onto the current scientific report structure instead of forcing awkward top-level headings. Domain introduction and goals belong in the introduction and benchmark-scope framing. Methodology covers source normalization, method adapters, datasets, metrics, alignment, and artifact protocol. Results belong in the experimental protocol and local evidence tables. Future work belongs in discussion or conclusion. Retrospective content should be short, neutral, and placed in discussion, conclusion, or an appendix subsection. Work breakdown should be reported separately from scientific evidence, preferably as an appendix or supplementary table.

Respect the course constraints when checking final deliverables: the report target is 15 pages excluding bibliography and appendix; the current Typst report template is the template unless explicitly changed; all team members need an equal share of report and code work; the final presentation is 20 minutes with equal speaking time; and the delivered project consists of prototype, report, presentation, and working code with a descriptive README.

Writing Workflow

Work section by section.

• Report prose must be full paragraphs with neutral scientific tone.
• Slide text may be concise fragments or bullets when that improves scanability.
• Match the nearby Typst style, spacing, labels, captions, and citation style.
• Keep architecture details only when they explain scientific consequences, artifact interpretation, frame semantics, or metric choices.
• Keep the current report shape: introduction, related work, scope, methods, datasets, metrics, experiments, discussion, conclusion, and appendix, while still representing the Moodle-required retrospective and work-breakdown content.
• Preserve five-member equal participation in final deliverables; evidence ownership is not final authorship or final speaking ownership.

Claim Audit

Before finalizing a sentence, classify the claim:

• implementation fact
• method summary
• dataset or protocol statement
• result
• limitation
• future work
• retrospective/process claim
• work-breakdown/contribution claim

Then verify it against the correct evidence tier. Do not turn implementation status, protocol design, or planned validation into empirical superiority. State unvalidated dense-cloud, efficiency, and leaderboard claims as future validation targets unless frozen metric artifacts exist.

VSLAM And Metric Rules

For trajectory or geometry metrics, name the reference source, pose direction or coordinate frame, alignment mode, units, thresholds, and artifact provenance. Make clear whether the statement is protocol, result, or interpretation.

Handle these topics explicitly when they appear:

• monocular scale ambiguity and Sim(3) alignment
• SE(3), Sim(3), or gravity-aware scale-yaw-translation alignment
• ARCore or ARKit as optional provider or baseline evidence, not hidden ground truth
• dense-cloud placement versus dense-cloud quality
• source normalization across ADVIO, TUM RGB-D, Record3D, and custom captures
• method-local world frames versus benchmark-normalized outputs

Scientific Writing QA

Before final delivery, check the report against the professor's scientific writing rules:

• use concise, precise, neutral language
• introduce technical terms and abbreviations before first use
• back claims with citations, tracked artifacts, or explicit limitations
• name prior work clearly instead of claiming external work as project work
• make results reproducible, checkable, and tied to the artifact contract
• reference every figure, table, and listing in text and give it a meaningful caption
• explain every variable used in mathematical notation
• avoid a heading immediately followed by another heading
• inspect the final PDF for text overflow, bad word breaks, and page/layout problems

Typst Edit Loop

Use the smallest focused Typst edit.

• Wrap figures and tables in #figure(...) with caption: and a label.
• Reference figures and tables with @label.
• Reference bibliography entries with @bib_key.
• Prefer existing booktabs and grid patterns.
• Do not replace the local IEEE-style report template or slide template unless explicitly requested.

After meaningful Typst edits, run the owning build:

make report-pdf
make slides-pdf
make final-slides

Report only commands that actually ran. If a build cannot run, state the exact blocker and the next-best check used.