tournament-corpus

1. When to use, and when not

Use this skill when you need to interpret the output of a Co-Scientist tournament. Entry points include:

• Ingesting a raw Co-Scientist tournament export as the co-scientist-specific entry point for hypothesis assessment. For the vendor-neutral entry point covering all strategies, see the hypothesis-entry skill.
• Determining whether a tournament produced a clear leader or whether the ranking is contested.
• Checking which ideas carry contradicted deep-verification claims.
• Reading one idea's prose (summary, description, reviews, verification) for detailed assessment without streaming it into context.

Do not use when:

• You need to evaluate the scientific evidence for or against a target identified by the tournament -- use the capability skills for that evidence type (structure, expression, constraint, regulatory variants, citations).
• You need to search for literature supporting or contradicting an idea -- use literature search tools and citation-resolution.

2. Preconditions

• Export file (for ingest): the raw Co-Scientist tournament export as a JSON file. The export is a serialised JavaScript object whose keys are minifier output -- the idea list may be Ur or gs, not a stable name. The CLI locates each section by its shape (presence of eloRating and ranking fields) and fails loudly when a required section is absent or ambiguous.
• Normalised artifact (for analyze and show): the .tournament.json produced by ingest. This is the stable Layer 0 record -- do not pass the raw export to analyze or show.
• No authentication needed -- the tool reads local files only.
• No network -- ingest, analyze, and show are all offline.
• Run pkde doctor before first use. It ends with a verdict line: STOP means fix or report before running anything; PROCEED means work, and the grouped warnings tell you which commands would refuse, which results need careful reading, and which are the tooling lead's to clear. Do not judge by the warning count; the verdict line grades them for you.

3. Tool invocations

Run ingest before analyze or show. analyze reads from disk and applies the coscientist threshold set. It can be re-run with different thresholds (--elo-decisive-gap, --min-win-rate, --max-contradicted-claims) without re-ingesting.

ingest preserves the raw export verbatim (.export.json) alongside the normalised record (.tournament.json). The normalisation is lossy and a reviewer must be able to reach the original bytes.

show writes one idea's prose to a file and prints the path -- it does not stream content into context. Select an idea by --rank or --gene. A gene name is matched against each member of a multi-gene idea (e.g. "CCNE1, CCNE2"), so a match on CCNE1 selects the idea even if the gene field lists both. Sections: summary, description, reviews-summary, verification-summary, all.

All output options: --json for machine-readable output, --quiet for paths only, --out to override the default output directory.

4. Interpretation contract

Tournament verdicts

analyze applies the coscientist threshold set and emits one of:

• clear-leader -- the top-ranked idea leads by at least the configured elo_decisive_gap and carries no advisories.
• leader-with-advisories -- the ELO gap is decisive but the leader carries advisories (contradicted claims, low win rate, or too few matches).
• no-clear-leader -- the ELO gap between #1 and #2 is below the configured elo_decisive_gap. The ranking does not support a confident selection.
• single-candidate -- only one idea was ranked. No comparative judgment is possible.
• unrankable -- no ideas had ELO ratings. The tournament produced no usable ranking.

Per-idea advisories

Each ranked idea is checked against four conditions. An advisory is not a veto -- it is a fact that must travel with any claim based on the idea's ranking:

• Contradicted claims -- the idea has more deep-verification claims contradicted than the configured max_contradicted_claims threshold.
• Low win rate -- the idea's win rate is below the configured min_win_rate threshold.
• Too few matches -- fewer than the configured min_matches threshold were played. The ranking is not well supported.
• No claims present -- the idea has no deep-verification claims at all. Absence of flagged claims is not evidence the idea was verified.

The denominator caveat

Every .analysis.json carries a metrics.claim_denominator field that states this caveat. The deepVerification section of the export is an exception list, not a census: it contains only the claims the verifier took issue with. In observed exports, every reported claim has carried a negative verdict (INACCURATE or LEANING_INACCURATE). That is not a high error rate; it is the definition of the list.

The analysis carries two fields that look like a denominator but are not:

• n_claims_reported_total -- the length of the exception list. It is the number of claims the verifier flagged, not the number of claims the idea made.
• claim_verdicts -- the same list broken down by verdict category (e.g. {INACCURATE: 12, LEANING_INACCURATE: 4}). Every entry is a disputed claim, because undisputed claims do not appear.

Both denominators are "claims already known to be disputed." Dividing n_contradicted_claims by n_claims_reported_total gives ~1.0 and is meaningless. The per-idea field n_contradicted_claims is an absolute count of flagged claims (checked against the configured max_contradicted_claims threshold). Do not compute a rate.

Mandatory relays

partial_export is the sole mandatory relay -- a qualifier that scopes a result rather than blocking a substitution. It is conditional: it fires when n_ideas < n_ideas_generated, confining conclusions to the ideas present in the export. The denominator caveat, previously carried as an always-true relay (coscientist.claim_denominator), is now a metrics.claim_denominator field on every .analysis.json -- it no longer competes for attention in the relay channel. The skill's own Section 4 ("ELO is relative ranking, not absolute quality") and the leader_gap_is_decisive documentation are load-bearing rather than supplementary: they carry the constraint that a single conditional relay cannot, because a decisive ranking is not evidence that the idea is scientifically correct.

Check mandatory_relays in both the .analysis.json and the .meta.json. Every relay code present must be satisfied in the finding.

ELO is relative ranking, not absolute quality

The ELO gap measures how consistently idea #1 beat idea #2 in pairwise comparisons. A decisive gap means the ranking is stable, not that the idea is correct. A "clear-leader" verdict says the tournament converged on this idea -- it does not say the idea is scientifically sound. A finding must not conflate ranking confidence with hypothesis confidence.

The assessment carries leader_gap_is_decisive -- a tri-state:

• true -- the ELO gap meets elo_decisive_gap. The verdict is either clear-leader or leader-with-advisories depending on whether the leader carries advisories.
• false -- the gap is below the threshold. The verdict is no-clear-leader.
• null -- there was only one idea (single-candidate) or none (unrankable). No gap exists. Do not read null as false.

Consequence rules

• Clear leader: state the gene, the ELO gap, and the win rate. Proceed to downstream assessment -- but the ranking does not substitute for evidence from the tools that evaluate the idea.
• Leader with advisories: state the advisories alongside the ranking. Contradicted claims or low win rate may indicate the idea won despite weaknesses, not that the weaknesses do not matter.
• No clear leader: the tournament did not converge. Do not select an idea on the basis of this ranking.
• Single candidate: no comparative judgment was possible. The idea is not "the winner" -- it is the only entry.
• Partial export: when partial_export fires, the ranking covers only n_ideas of n_ideas_generated. An idea absent from the export was not evaluated by this tool -- it was not eliminated by the tournament.

Cross-disciplinary consequences

• A tournament leader's gene becomes the subject of downstream evidence gathering (structure, expression, constraint, regulatory variants). The ranking is the entry point, not the conclusion.
• Contradicted deep-verification claims in the leader should be checked against the evidence tools -- a fabricated claim about a real gene does not make the gene a bad target, but it does make the tournament's reasoning unreliable at that point.

What this section produces

Following this contract produces a Layer 1 finding in findings/ that cites the .analysis.json, satisfies the mandatory relay, reports contradicted claims as counts, and does not conflate ranking confidence with hypothesis confidence. Everything the tools emitted stays under raw/hypotheses/.

Thresholds are cited by name (e.g. "the configured elo_decisive_gap threshold"), never by value. The values in force are in the artifact: every .analysis.json carries threshold_set (name@version), thresholds_applied, threshold_sources (default | program | flag), and threshold_provenance. A specialist quoting a number should quote it from the analysis they are citing.

5. Failure modes and anti-fabrication guard

Do not compute these values yourself. The value is whatever the tool emitted. If the tool did not run, there is no value. If a required tool is unavailable, report the task as blocked. Do not estimate, and do not proceed on an assumed result.

Named pathologies

• Computing a contradicted-claim rate. The denominator does not exist. deepVerification lists only claims the verifier disputed -- in all exports observed, every listed claim carried an inaccurate verdict. Dividing by the number of listed claims gives ~1.0 and is meaningless. The metrics.claim_denominator field requires reporting the count.
• Treating "no claims" as "no problems." An idea with zero deep-verification claims has not been verified -- it has been unexamined. The advisory exists because this silence looks like a clean bill of health.
• Conflating ELO ranking with scientific merit. A decisive ELO gap means the ranking is stable. It does not mean the idea is correct, safe, or supported by experimental evidence. The ranking is an input to investigation, not a conclusion.
• Selecting an idea from a no-clear-leader tournament. The tournament did not converge. Picking the top-ranked idea anyway imports a false precision -- the gap is below the threshold that makes the ranking meaningful.
• Ignoring advisories on a clear leader. "Clear leader" means the gap is decisive and the leader has no advisories. "Leader with advisories" means the gap is decisive but the idea has problems. Dropping the advisories turns the second verdict into the first.
• Treating a partial export as a complete census. When partial_export fires, the ranking covers only the carried ideas. An idea absent from the export was not evaluated -- it was not eliminated.
• Reporting minified keys as meaningful identifiers. The export's JavaScript keys (Ur, gs, BVa) are minifier output and change across Co-Scientist builds. The normalised .tournament.json uses stable names; cite those.
• Streaming show output into context. show writes prose to a file deliberately. Reading the file when needed keeps long-form tournament prose out of the working context. Pasting it inline defeats the zero-bloat design.