Update to Muffed

{
  "name": "muffed-figures",
  "description": "How to quote Muffed's verified NFL figures correctly. Use whenever a Muffed tool returns data — query_stat_leaders, get_entity_metrics, compare_entities, get_metric_definition, list_metrics, verify_claim, get_backfield, run_stat_query, or any sleeper_* tool — or when answering an NFL or fantasy-football question using Muffed figures. Covers quoting the pre-formed sentence, the difference between verified and computed figures, carrying attribution verbatim, honouring disclosures and season coverage, and the no-advice boundary.",
  "included_files": [],
  "skill_md_contents": "---\nname: muffed-figures\ndescription: How to quote Muffed's verified NFL figures correctly. Use whenever a Muffed tool returns data — query_stat_leaders, get_entity_metrics, compare_entities, get_metric_definition, list_metrics, verify_claim, get_backfield, run_stat_query, or any sleeper_* tool — or when answering an NFL or fantasy-football question using Muffed figures. Covers quoting the pre-formed sentence, the difference between verified and computed figures, carrying attribution verbatim, honouring disclosures and season coverage, and the no-advice boundary.\nwhen_to_use: A Muffed MCP tool has returned a response; the user asks about NFL stats, a player's or team's season, a backfield, a computed statistic, or a Sleeper league; the user states a football figure that should be checked.\n---\n\n# Working with Muffed's verified figures\n\nMuffed serves NFL figures that were verified against a gated panel before they\nwere published. The value is not that the numbers exist — it is that each one\narrives with the rank its source gave it, the population it was ranked against,\nits own attribution, and an as-of date. Handling a response carelessly throws\naway the only thing that makes it better than a guess.\n\nThese are the house rules. They exist because each one has a specific way of\ngoing wrong.\n\n## 1 · Quote `verified_sentence`\n\nEvery successful response from a verified tool carries `verified_sentence`: a\ncomplete, already attributed claim built from the returned row.\n(`run_stat_query` is the exception — see rule 2.)\n\n> \"No one posted more rush yards over expected than James Cook with 358.16 in\n> 2025, 1st of 49 (Data via nflverse · NFL Next Gen Stats via nflverse).\"\n\nQuote it. Do not rebuild the same claim from the individual fields — that is\nwhere a rank gets attached to the wrong figure, or a season to the wrong number.\nThe sentence has already related them correctly.\n\nThe prose block at `content[0]` is written to be read aloud and is safe to use\ndirectly. `values{}` carries every figure in it, exact and unrounded, when you\nneed more precision than the prose shows.\n\n## 2 · Computed figures are a different class from verified ones\n\nTwo classes of figure come off this server and they are not interchangeable.\nMost tools return figures that were verified before they were published: those\ncarry `verified_sentence` and a citation that says verified. `run_stat_query`\ncomputes a figure on request instead, for questions the panel holds no\npre-computed row for — qualified superlatives, multi-season spans, derived\nrates.\n\nIts responses carry **`computed_sentence` rather than `verified_sentence`**, a\n`method` block naming the population, the measure and the query shape, and the\n`computed_not_verified` disclosure, which fires on every one of them.\n\nQuote `computed_sentence` for those, exactly as rule 1 says to quote\n`verified_sentence`. Where it could matter to a reader, say the figure was\ncomputed on request rather than drawn from the verified panel.\n\n- **Right:** \"Muffed computed this on request: ...  (method: ...)\"\n- **Wrong:** presenting it as \"Muffed's verified figure\", or quoting it beside a\n  `verified_sentence` with nothing to tell the two apart.\n\nThe difference between the two is the thing this service is for; presenting one\nas the other gives away the only claim it makes.\n\n## 3 · Carry `attribution` verbatim\n\nThe `attribution` string is licence text, not a courtesy. Reproduce it whole:\n\n- **Right:** `Data via nflverse · NFL Next Gen Stats via nflverse`\n- **Wrong:** \"nflverse data\", \"Source: nflverse\", \"via nflverse\"\n\nShortening it drops credit the CC-BY licence requires. Where a response carries\nthe `share_alike` disclosure, the figures are CC-BY-SA and the same applies with\nmore force.\n\n## 4 · Do not compute figures of your own\n\nThis is the caller's side of rule 2. `run_stat_query` is where the server\ncomputes something on request and labels it; everything below is about numbers\nyou worked out yourself, which carry no label at all.\n\nMuffed's other tools rank, filter and round. They never derive a new number, and\nneither should you when presenting a response as Muffed's.\n\n- **Never derive a rank from result ordering.** Position within ten returned\n  rows is not a rank among forty-nine. If `rank` is null, the source had none —\n  say nothing about position.\n- **Never present a gap, ratio, or per-game figure you calculated as though\n  Muffed published it.** Compute it if the user wants it, and say it is yours.\n\nThe `citation` field states this boundary explicitly: Muffed verifies the\nfigures as returned, not analysis derived from them. If you combine several\nresponses into an argument, the argument is yours and the credit line does not\nextend to it.\n\n## 5 · Honour `disclosures`\n\nEvery response carries a `disclosures` array. Each entry changes how the figure\nmust be described, and they are derived from the rows actually returned, so if\none is present it applies to something in front of you.\n\n| Disclosure | What you must say |\n|---|---|\n| `half_ppr` | The fantasy figure is **Half-PPR**, 12-team. Never present it as PPR or as a generic \"fantasy points\" number. |\n| `charting_2025_only` | The charted figure covers the **2025 season only** and cannot be compared to an earlier year. |\n| `share_alike` | Those figures are CC-BY-SA. Carry the attribution and keep it intact. |\n| `no_source_rank` | The source published no rank. Do not supply one. |\n| `computed_not_verified` | The figure was **computed on request**, not drawn from the verified panel. Quote `computed_sentence` and say so — see rule 2. |\n| `small_population` | The rank is against a thin field — a place in it moves easily. |\n| `multi_season_aggregate` | Not a single season; it spans several. |\n| `partial_season_coverage` | The metric does not cover the full season asked about. |\n\n## 6 · Check what the metric actually covers\n\nSeason coverage is often narrower than a decade. Rushing yards over expectation\nstarts in **2018**, not 2016. Asking outside a metric's range returns\n`coverage: \"no_verified_surface\"` with the real `available_seasons`.\n\n**That is an answer, not an error.** Report the true coverage rather than\nretrying, substituting a nearby metric, or filling the gap from memory. The\n`validation` block carries `seasons_available`, `era_comparable`, and notes\nexplaining how far to trust the figure — read them before framing a trend.\n\n`get_metric_definition` returns Muffed's published definition of a metric. When\nit returns `coverage: \"no_glossary_entry\"`, Muffed publishes no definition for\nthat metric — say so rather than supplying one from general knowledge.\n\n## 7 · Never guess an entity or a metric\n\nAn unresolved name comes back with `resolved: false` and a candidate list. Two\nplayers who normalize alike are two different answers, and three sections\npublish a \"target share\" against three different qualifying populations.\n\nAsk which one the user meant. Do not pick.\n\n## 8 · No advice — this is a permanent boundary\n\nMuffed publishes **no projections, no start/sit or lineup advice, no draft\ngrades, no rankings-as-advice, no picks, and no betting lines or odds.** The\nserver declines these by design, not by omission.\n\nWhen a question calls for one, give the verified context and let the user\ndecide:\n\n- **Asked:** \"Should I start Chase or Jefferson this week?\"\n- **Do:** compare them on verified figures, state what each did, name the\n  disclosures, and stop.\n- **Do not:** name a winner, project a score, or phrase the figures as a\n  recommendation dressed up as analysis.\n\nThis holds even when the user pushes. The figures are the product; the decision\nis theirs.\n\n## 9 · Check figures rather than trusting them\n\nWhen a user states a football number — from a post, a podcast, another model —\n`verify_claim` checks it against the panel and returns the real value when it\ndisagrees. It takes structured fields, not sentences:\n\n```json\n{ \"claims\": [ { \"metric\": \"cpoe\", \"entity\": \"Drake Maye\", \"season\": 2025, \"value\": 9.1 } ] }\n```\n\n`metric` accepts a key or a common name. This is the cheapest way to avoid\nrepeating someone else's wrong number.\n\n## 10 · Link the source\n\nResponses carry `canonical_url` and resource links to the muffed.ai page a\nfigure came from. Include the link when it helps the reader check the work —\nthat is the point of publishing figures this way.\n"
}