CLAUDE Decoder — the whole field, one truth, saved forever.
The GROK Decoder studies one pro. The CLAUDE Decoder decodes the entire contest: upload the full DraftKings standings and get every DFS user ranked with fair, field-normalized percentiles, the true field ownership and duplication census reconstructed from every lineup in the file, a complete pre-lock forensic analysis of any user you select — then the deep forensics layer: field-proven win research, a DPC# lab graded against every finish in the contest, red-flag pro-mistake detection, and a CLAUDE BUILD loop that turns the analysis into new lineups on any slate, analyzes the build itself, and saves BOTH analyses forever, linked.
wiring/claude-decoder.wiring.json.
The laws: pre-lock signals only (hindsight is denied at the engine level) · the census is post-lock ground truth for research, never a build input · every save is an immutable version — reopening never recomputes · missing data stays null with a reason.
MATCHUP LAW: hitters face the opposing SP only ·
pitchers (SP/RP/P) never carry opp_pitcher* ·
zero pitcher-vs-pitcher matchups in pool / card / RNK / DPC.
1 Bind the slate
The decoder measures picks against the pool they were chosen from — the date must match the contest's slate.
2 Upload the FULL contest standings
DraftKings contest standings CSV (Rank, EntryId, EntryName, Points, Lineup).
Unlike the GROK decoder, upload the whole file — every entry becomes part of the field catalog
and the ownership census. The file is checksummed (SHA-256) as an immutable source artifact.
…or paste the CSV
3 The field catalog — every DFS user, ranked fairly
Every entry grouped by user. Finish percentile = 100 × (field − rank) / (field − 1) — higher is better, comparable across contests. ROI stays n/a until a verified payout table exists (no invented values).
CLAUDE Decoder progress
Preparing…
The page stays scrollable and interactive while this runs. A failure shows here with a Retry button — the page never remains blank.
The sample & the source artifact
Every number below rests on this — and on an immutable, checksummed source file.
The logic breakdown
What this book does on purpose, stated in our datapoints.
What their picks load on (pre-lock only)
Registry-column lift vs the slate pool. Thin columns (<50% pool coverage) are excluded and reported. Hindsight columns are denied inside the engine — they cannot appear here. Build scoring reads the recipe’s evidence columns + full platform scan on the production deep path.
How they build
Construction, not selection — stacks, pitcher pairs, salary discipline, uniqueness.
Exposure pyramid — LOCK · ANCHOR · CORE · SPICE
Tier thresholds: LOCK ≥80%, ANCHOR 60–79.9%, CORE 25–59.9%, SPICE <25% of the book.
Skill vs variance — luck decomposition
Deterministic bootstrap on the user's finish percentiles: how much of this result is repeatable process?
CLAUDE skill doctrine — the book through the chain
This book judged by the CLAUDE skill chain (theory → foundation → leverage → portfolio → learning overlay): ownership tiers, the duplication stop rule, negative-correlation veto, win-path audit, and portfolio discipline. Descriptive evidence of the pro's process — never an instruction to copy it blindly.
The portable CLAUDE recipe
A typed, provider-agnostic build recipe compiled from pre-lock evidence only. Hard gates (eligibility · salary · exposure · stack · no-hindsight · live-status) always run after any engine.
FIELD CENSUS — true ownership & duplication (post-lock lane)
Reconstructed from every lineup in the standings file: actual field ownership, exact dupes, chalk concentration, and the field's stack census. Ground truth the industry only estimates — kept strictly out of signals and recipes.
4 Save an immutable version to the CLAUDE Library
One save = one permanent version: source checksum, field summary, census, full report, and recipe. Re-saving the same title creates a new version that supersedes the old one — the old version stays recoverable. Saved books are also listed for CLAUDE builds on Mastery.
What this report will not claim
CLAUDE Analysis Library — immutable versions
Every saved analysis in this browser (localStorage + IndexedDB mirror; export JSON for backup). Open renders the stored record — it never recomputes. Superseded versions stay visible.