--- title: Receipts — method and pre-registration as_of: 2026-08-06 source: receipts.html on datadawgs216.com canonical_url: https://datadawgs216.com/data/receipts-method.md data: https://datadawgs216.com/data/receipts.json --- # Receipts Every call gets a date, a number and a benchmark — written down **before** the games, so neither the result nor the story can be edited afterwards. Locked **2026-08-06**, 272 calls, SHA-256 `4b87fe0e0790a1e6196fbbcad9b444e00735d5c7035ad1c7f9b476a55fc2716a`. **Nothing is graded yet.** The 2026 season has not started. Any claim about how these calls performed is wrong on its face. ## What the calls are Predictions come from nfelo, snapshot `0d3f8418`, model `v4.3.0`. For the 16 games nfelo published directly, its own number is used. For the rest, the probability is derived from its power ratings through a margin model fitted on 4,108 historical games: 23.58 Elo per point, 2.10 points of home-field advantage, residual SD 13.18. The benchmark is the **devigged closing moneyline**, taken at lock. It exists for 51 of 272 games; the rest had no line on 2026-08-06 and therefore have **no benchmark and never will**. Those games test calibration only. ## What is being tested **Primary, resolvable this season — calibration.** Are the stated probabilities calibrated: do the games called at 70% land about 70% of the time? 272 games is enough to catch a real miscalibration. **Secondary, NOT resolvable this season — does nfelo beat the closing line?** One season cannot settle it, and that is being said now rather than in January. Measured on 4,096 historical games, the paired per-game spread between nfelo and the closing line is so wide relative to the edge that a 272-game season gives: - a minimum detectable straight-up difference of **±1.60 pp** — the advertised +0.14 pp is **9%** of that - a minimum detectable Brier gap of **0.0034** — the observed edge is **19%** of that - only about **five games a season where the two even disagree** At the observed effect sizes that takes roughly **26 seasons** on Brier and **130** straight up. **Any 2026 result on this axis is noise, whichever way it falls.** It is logged because season one of twenty-six still has to be season one. ## What would change our mind On calibration — the resolvable question — a miss of more than about **8 points** in any bin holding 30+ resolved games, in the same direction across the 70–90% range, is a real failure and not sampling noise. That is the finding this page is built to catch. A single wrong-looking week is not evidence of anything. On the benchmark comparison, nothing that happens in 2026 changes our mind, by construction. Stated in advance so it cannot be reinterpreted later. ## Scoring Brier score (lower is better) as the headline, because it grades the *number* rather than the pick. Straight-up accuracy alongside it, paired against the benchmark on the identical game set. Calibration by bin. Every figure recomputed from resolved games only — nothing is projected, nothing is back-filled. Grading is idempotent: a game already scored is never re-scored, a game not yet final is skipped rather than guessed, and a tie stores no straight-up result. Finals come from ESPN's public scoreboard. ## Integrity — verify it yourself The call sheet is hashed. If a prediction is ever quietly edited, the hash changes and the receipt is void. Canonical string: take the rows of `data` in `receipts.json` **in the order stored** (the locked order — do not re-sort). For each row emit ``` ${id}|${p.toFixed(4)}|${mk == null ? "" : mk.toFixed(4)} ``` and join the rows with `\n` (LF, no trailing newline). SHA-256 the UTF-8 bytes, lowercase hex. The string is 6,697 bytes over 272 rows and begins: ``` 2026_01_ARI_LAC|0.8060|0.8271 ``` That must equal `4b87fe0e0790a1e6196fbbcad9b444e00735d5c7035ad1c7f9b476a55fc2716a`. The reason the spec is published is so you don't have to trust this site. Recompute it. ## Tiers and the collar Three tiers, one gate: **does it work, and can it be trusted?** - **Pup** — live and useful, not yet validated. It may compute real answers and still have open questions about calibration, assumptions, data quality or edge. Everything starts here. A Pup means use it with your eyes open. - **Working Dawg** — earned its collar. A tool becomes a Dawg only after its evidence survives validation: forecasts need receipts against a benchmark chosen in advance, measurement tools need named sources and reproducible maths. Looking finished is not evidence. - **The DawgHouse** (formerly The Pound) — shelved, with the reason attached. Nothing is here because it is hard; each entry keeps its exact blocker and its minimum path back. The collar is a mark, not decoration, so what it stands for is written here rather than left to the picture. Where a tier is rendered **from `/data/surfaces.json`** instead of typed into a page, `tier: "dawg"` draws a collar glyph carrying the accessible name *Working Dawg — earned its collar*; `labs` and `pound` stay text chips, because only the earned tier gets a mark. The criteria for drawing one are the three paragraphs above and nothing else. Which tools currently hold a collar, what was checked and what was explicitly **not** checked, is the tier audit: [/data/tier-audit.md](/data/tier-audit.md), dated 2026-08-07. A collar certifies what its audit entry says it certifies and no more. nfelo Power Ratings is the standing example: its collar is for faithfully mirroring a named upstream model, not for that model forecasting well, and the evidence on file says it does not beat the market. ## Provenance Forecasts from [nfelo](https://github.com/greerreNFL/nfelo) (MIT), snapshot `0d3f8418`. Schedule and closing lines from [nflverse/nfldata](https://github.com/nflverse/nfldata). Finals from ESPN. Call sheet locked 2026-08-06. ## Survivor inputs **2026-09-04 amendment — live survivor market derivation changed.** The 51 market rows captured on 2026-08-06 used devigged moneyline probabilities. Beginning 2026-09-04, the mutable survivor snapshot reads spread-implied home win probabilities from the daily nfelo mirror, preferring its close and then its open. This changes the derivation, not the quantity: `mk` remains a market-implied home win probability. The mirror records the upstream observation time but upstream does not identify the book, so `mk_book` is null rather than an invented attribution. The original 51 rows are retained as dated carry-forwards where a fresh row is unavailable and are not back-converted. Published receipts and graded rows are unchanged. ## Material changes Changes to how this site names or frames its claims, recorded rather than applied silently. The forecast rows are covered by the hash above; this list covers what the hash cannot see. - **2026-08-10 — Labs is renamed Pup, and stops being a tier name.** Human-facing labels only. Stored machine values did not move: `tier` is still the string `labs`, and /data/surfaces.json still keys the policy on it. The ladder now reads Pup (unvalidated) → Working Dawg (validated) → The DawgHouse (the evidence said stop), which names what is known about a tool rather than where it lives. *Data Dawgs Labs* survives as the name of a PLACE — the lab pages — not a grade. No tool's tier changed, and no forecast, probability or benchmark moved. `tier_meaning` is now asserted byte-identical across all 24 published payloads and the five sources that write them; before this commit only its existence was checked, so two definitions could ship at once. - **2026-08-09 — the collar becomes a symbol, and its criteria are published.** `tier: "dawg"` now renders as a collar glyph with an accessible name wherever a tier is read from /data/surfaces.json, starting with the cards on /arena.html. The criteria are written into this file above rather than implied by the picture; without them the mark would be a claim with nothing behind it. No tool's tier changed, no forecast, probability or benchmark moved, and stored machine values are untouched — `tier` is still the string `dawg`. - **2026-08-09 — The Pound is renamed The DawgHouse.** Human-facing labels only: /pound.html forwards to /dawghouse.html. Stored machine values did not move — `tier` stays `pound` and /data/pound-tools.json keeps its name. No forecast, probability or benchmark changed. The same day the shelf was reduced to genuinely blocked work; the complete NFL tools left it. Full inventory: /data/pound-tools.json.