Use the real paper title or the working argument.
Step 01
Packet setup
Define the governing object, consequence domain, recursive pressure, and manuscript body before any polished prose starts to harden.
What changed: this route is now the actual packet-builder app, not a brochure page.
Boundary: reviewer output is generated locally from the visible draft, claims, checks,
and controls. No hidden remote model call is required.
TXT, MD, DOCX
No file loaded yet. Upload .txt, .md, or .docx. Older
.doc files should be resaved as .docx first.
This frames the risk surface and the level of precision the packet must satisfy.
Pressure level decides how aggressively the packet should challenge its own wording.
Paste or upload text
Step 02
Claim mapping
Keep direct evidence, supported inference, and unresolved unknowns visible instead of laundering them into one polished lane.
Method rule: if a claim outruns its visible source anchor, narrow it or mark it as
unresolved. Do not let fluent wording impersonate evidence.
Accuracy boundary: seeded claims are heuristic candidates based on visible lexical cues.
Confirm the text, class, and anchor note manually before letting any seeded line travel.
0 claims mapped
0 direct 0 inferred 0 unresolved
Step 03
Recursive checks
Run contradiction, boundary, and overclaim checks before this packet is allowed to circulate.
Progress 0 / 6 passed
Step 04
Henry reviewer stack
Generate supportive, cold-method, and harsh heuristic review pressure locally from the visible packet state.
Improvement: the old one-file version attempted a direct model call from the browser. This
public route now produces bounded local review pressure from visible packet data so the app actually works
after deployment.
Run all three reviewers at once for full pressure coverage.
Supportive focus
Confirms what the manuscript already anchors well and where the argument is behaving responsibly.
Awaiting review trigger.
Cold-method focus
Evaluates inference drift, missing source classes, and unearned causality without praise language.
Awaiting review trigger.
Harsh focus
Names the single sentence most likely to fail under cross-examination and explains why.
Awaiting review trigger.
Step 05
Control assignment
Attach a named owner, evidence requirement, and review interval to each load-bearing claim path.
Gate: export should not travel without at least one complete control row. If nobody owns
the next review, the packet is not ready.
Step 06
Export layer
Export a bounded packet with readiness, controls, disclosure language, and review pressure visible in one markdown file.
Markdown export
Ready for bounded export