# Round five: faster delivery, focused review

This addendum applies to grok-round-005 and supersedes conflicting operational instructions in older packs. The assignment is fifty series. Earlier wording about ten or twenty does not change this roster. Existing Claude, Kimi and Grok assignments remain reserved.

## Produce finished essays, not research reports

Write directly about the work, its development, construction, mint, rights and individual outputs. Explain the actual mechanism and what it makes visible. Keep queries, hashes, access checks and repeated cautions in evidence.json and research-notes.md. The prose must stand on its own as an art essay. Keep the seven ordered sections and paragraph citations. There is no upper word limit; the existing minimum is not a reason to add filler or repeat paragraphs. Do not write labels such as "Sustained reading" in place of a sustained reading.

Read the shared instructions once, then use the individual identity and source leads for each series. Fetch independent records in batches where the service supports it. Reuse a single genuine artist interview or wallet response across relevant essays, retaining the actual evidence and retrieval scope. Do not repeat the same research solely because it appears in several output files. Generate Markdown, citations and reports from the canonical JSON to avoid divergent copies.

The contributor still performs the source, full-code and original-artwork research required by the methodology. The coordinator will audit this evidence rather than automatically repeat it all. For code, give a concise path from token input to parameter choices to drawing and time/interaction behavior, with exact function or line locators. A byte count and a list of function names are insufficient. For visuals, record actual concrete observations for four originals; for animation, record multiple states and honestly disclose timing. Do not describe two rows if the actual output has four columns. Never infer a reserved mint from an early timestamp. Missing historical prices and rights are explicit unknowns, not invitations to guess.

## Return in useful increments

Work through the fifty in roster order, save each complete package immediately, and keep batch-progress.md current. Group the results in five increments of ten for convenient delivery. Each return ZIP uses submission/CANONICAL-SLUG/ with the same seven required files. You may deliver each ten as it is finished while continuing the remainder. Name them grok-round-005-part-01.zip through part-05.zip. Alternatively return one grok-round-005-packages.zip containing all fifty. Do not duplicate an essay across parts. Include the exact supplied validator output for each essay. Mark any incomplete work precisely and continue with the next series.

## Coordinator review policy

On receipt, run package, identity, citation, source-link, duplicated-prose, chronology and presentation checks across every essay. Read every essay for unsupported claims and generic filler. Check the collection example and rights caption for every page. Use the contributor's attributable observations and reproducible research records when they are coherent and adequately supported; a different reviewer need not claim to have repeated them.

For each ten-essay increment, deeply audit at least two essays, including code-to-image claims and original media. Prioritize unusual code, sound or interaction, source contradictions, thin construction explanations, and questionable history. Expand the audit to affected essays if a sample reveals a recurring issue. Correct isolated defects without rewriting sound prose. Escalate only the affected essays for deeper research. Publish accepted essays together after site checks, preserving original credit and honest usage. A structural pass alone never proves factual accuracy, but a full independent rewrite is not the default acceptance step.