{"library":"eidos-color-team-roster","note":"Six bounded review contracts. Each finding stays in its own review; the roster never folds findings into a composite score.","teams":[{"accent":"#215fa1","boundaries":["Missing wallet-forensics or market inputs never classify LOW. Work can be withheld as UNSCORED, UNKNOWN, STALE, or UNDER_REVIEW.","The Evidence review qualifies evidence; it never scores a proposal and never fills a gap with a fixture value.","A disclosed term is not proof that delivery is enforceable, a branch is independent, or a conditional price is causal."],"contract":{"handoff":{"detail":"Complete forensic evidence can move to monitored-vector matching; qualified proposal terms can move to priceability stress testing. Partial and unavailable packets keep their uncertainty attached for every downstream team.","title":"Qualified record"},"inputs":{"detail":"Wallet history, funding path, stake accumulation and recovery, trade execution, market data, a source-dated treasury snapshot, and the time-stamped disclosures of the decision's record — its options, selection rule, window, decider and terms — covering material facts, the disclosure cutoff, specification, delivery enforcement, participation access, branch dependencies, and causal assumptions. Where the decision's trades are recorded, the wallet-attributed swap record is qualified too, across every option's pool. For a launch-covenant record, the packet is the disclosed terms at issuance.","title":"Evidence packet"},"outputs":{"detail":"Availability, confidence, source freshness, field-level forensic coverage, disclosure qualification, a wallet-level census of an indexed swap record with the playbook gates it fills and the ones it cannot, and an explicit withheld state when required evidence is incomplete; a vector classification never runs without its own complete evidence gate.","title":"Coverage state"}},"discipline":"evidence qualification","heading":"Qualify the evidence.","id":"evidence","mission":"Before a decision gets a vector classification or priceability review, the Evidence review records what is present, what is missing, and how current it is. It qualifies disclosed specification, delivery terms, corrective access, and causal assumptions without filling gaps or deciding what those terms are worth.","name":"Evidence","role":"Evidence","summary":"Qualifies market, wallet, and decision disclosures before another team treats the packet as priceable.","version":"2.0.0"},{"accent":"#135394","boundaries":["Indicators are never added across vectors into a composite proposal-risk score.","A documented vector never produces a classification; only monitored vectors run.","A vector no-match is never presented as a safety rating.","Routine arbitrage and directional bots are never counted as wash: paired conditional legs are netted out first, and presence in more than one option's pool, machine cadence, frequency and tilt shift count only beside round trips or a near-neutral ending.","A semantic weakness in a decision's disclosed rules may be stress-tested without being asserted as an attack finding."],"contract":{"handoff":{"detail":"A monitored finding travels with its vector id, version, and scope note. A documented threat travels as an unclassified candidate and evidence request. The Claims review checks both claim boundaries before publication.","title":"Scoped result"},"inputs":{"detail":"An Evidence-qualified packet: wallet history, funding path, accumulation, stake recovery, execution series, and the complete time-stamped semantics the market is asked to price. A qualified swap record feeds only the playbooks whose gates it fills; the rest stay withheld with their missing evidence named.","title":"Qualified packet"},"outputs":{"detail":"A semantic stress-test record plus, only for a monitored detector with complete evidence, a count of matched indicators and a classification scoped to that one vector; otherwise the state is explicitly withheld or documented-only.","title":"Vector finding"}},"discipline":"attack testing","heading":"Try to break it.","id":"attack","mission":"The Attack review stress-tests the decision's disclosed selection rule, specification, delivery options, participation constraints, and branch dependencies, then runs versioned monitored playbooks against qualified evidence. Every classified result belongs to one mechanism, and no match is not a claim of safety. Documented threats remain sourced hypotheses with detector requirements: the Attack review may challenge them but never assess or classify them.","name":"Attack","role":"Attack","summary":"Stress-tests a decision's disclosed rules and runs only monitored economic attack playbooks against evidence.","version":"2.0.0"},{"accent":"#003d75","boundaries":["The Market review may not invent forensics: attack indicators stay dark unless an indexed swap record fills a playbook's own gate, and that reading belongs to the Attack review.","Provisional live math is never presented as a final outcome.","A pressure watch is a market observation over retained reads; it never becomes an attack-vector indicator or classification.","A flip-cost figure is a lower bound under stated assumptions; it is never a safety rating.","An order-flow toxicity reading is a measurement of imbalance in volume time. A decision market concludes by trading one-sided, so a high reading is never an attack finding and a low one is never safety.","Microstructure readings ship separately and are never folded into a signal-credibility or market-quality index; a composite hides which input moved.","A metric that presupposes a limit order book is refused rather than approximated. These markets trade in automated pools with no bid, ask, queue, or maker inventory.","Corrective-access and decision-fragility observations are not approval, a causal claim, an attack finding, or a rating."],"contract":{"handoff":{"detail":"Market state, corrective access, and decision fragility travel beside forensic context without standing in for missing wallet history, volume, execution evidence, or causal attribution.","title":"Defense watch"},"inputs":{"detail":"The engine's own accounts and decision record provide current state, the selection rule and threshold, each option's time-weighted price, and, for a measured-outcome market, the outcome measured; retained reads place prices in the window, and venue rules and live depth describe who can supply corrective exposure. Where a decision's swaps are indexed, that public event record supplies the signed order flow, and it states how much of the window it reaches.","title":"Decision record"},"outputs":{"detail":"Under a price rule the best challenger's time-weighted price must strictly beat the default's by the threshold; a pass/fail market is the two-option case. Retained reads report the margin, crossings, closing-stretch pressure, an outlook while the window is open, corrective access, and disclosed-depth flip cost; a closed decision reports its completed print, and a measured-outcome market the outcome it was decided on; an indexed swap record additionally reports order-flow toxicity, signed imbalance, cumulative delta, per-pool price impact, and participation structure, each separately. Live math stays provisional; after settlement, the tape sits beneath the engine’s authoritative outcome.","title":"Threshold read"}},"discipline":"market watch","heading":"Watch the market.","id":"market","mission":"The Market review reads the decision's state, each option's time-weighted price, the engine's selection rule — a default option that stands unless a challenger beats it by a threshold, or a measured outcome read after the window — and observable access to corrective exposure. It describes decision fragility and participation constraints as market structure; its observations stay separate from the Attack review’s forensic result.","name":"Market","role":"Market","summary":"Observes live decision fragility and corrective participation access without turning either into a rating.","version":"2.0.0"},{"accent":"#2e6bad","boundaries":["The incident never becomes the detector’s name.","A playbook ships only with a stable id, a semver version, ordered indicators, methodology sources, and a scope note.","A sourced candidate stays documented until its named evidence gate and detector both ship."],"contract":{"handoff":{"detail":"The Attack review receives runnable monitored playbooks and unclassified documented candidates. Published cases and theory remain provenance beneath them rather than hard-coded conclusions.","title":"Standing check"},"inputs":{"detail":"Cited decision records, market math, wallet forensics, explicit observations, theoretical mechanisms, counterarguments, and falsifiers from published source research.","title":"Published research"},"outputs":{"detail":"A versioned monitored mechanism with ordered indicators and an independent scope boundary, or a documented candidate with its detector evidence gate left visibly pending; covenant checks remain versioned against published launch terms. Against one decision, which monitored playbooks its evidence could run and what each other playbook still waits on.","title":"Runnable definition"}},"discipline":"playbook building","heading":"Codify the research.","id":"playbooks","mission":"The Playbooks review turns published investigations into reusable definitions. A decision-priceability candidate gets a stable name, semver version, source, explicit assumptions, evidence requirements, indicators, and detector-pending status before any implementation exists; the incident or theory never becomes a classification.","name":"Playbooks","role":"Playbooks","summary":"Versions sourced attack candidates, evidence gates, and runnable checks without promoting theory into findings.","version":"2.0.0"},{"accent":"#014787","boundaries":["The check moves forward; the old case’s conclusion does not.","An uncited lesson does not enter the loop — every replay keeps its source.","An unobserved branch is never invented to declare what would have happened under a different decision."],"contract":{"handoff":{"detail":"The Market and Attack reviews receive the updated check and its falsifiers; the old case keeps its citation, observed branches, and uncertainty as a replay.","title":"Refreshed watch"},"inputs":{"detail":"Published incidents and experiments with their evidence, market outcome, delivery record, falsifiers, and vector result preserved together as cited replays.","title":"Closed cases"},"outputs":{"detail":"Reusable, versioned checks and documented-candidate evidence updates extracted from observed outcomes, with unresolved and counterfactual outcomes kept out. Against one decision, what the same decider's earlier decisions on its engine selected, counting only those that closed before it opened, cited to the engine's record.","title":"Standing checks"}},"discipline":"learning loop","heading":"Close the loop.","id":"learning","mission":"The Learning review links what the Attack review learned in published cases with what the Market review should watch next. For decision-priceability candidates it retains disclosed assumptions, observed outcomes, sanctions, delivery, and falsifiers while keeping unobserved branches explicitly unobserved.","name":"Learning","role":"Learning","summary":"Retains observed outcomes and falsifiers so later reviews can learn without inventing counterfactuals.","version":"2.0.0"},{"accent":"#3a77ba","boundaries":["Complete evidence with 0–1 matched indicators is NO-MATCH, never LOW. NO-MATCH means this vector did not match; it does not mean the decision is safe.","A classification never leaves its vector’s scope; market outcome and forensic match stay separate.","A covenant review is never a decision classification: unknown covenant terms stay unknown, and a disclosed covenant is not a safety rating.","No roster output approves a decision, proves causation, combines reviews into a composite score, or supplies a safety rating."],"contract":{"handoff":{"detail":"Only scoped, cited, per-review claims reach the page; approval, causal-proof, composite, and safety language goes back to the team that made it.","title":"Published record"},"inputs":{"detail":"Every team’s finding with its evidence state, vector scope, data vintage, decision terms, causal assumptions, and documented-versus-monitored status attached.","title":"Draft findings"},"outputs":{"detail":"Claims trimmed to what the evidence establishes, with withheld states and detector-pending threats explicit rather than rounded into approval, causal proof, or safety; against one decision, the vectors measured, their scope, and how many said nothing.","title":"Publishable claims"}},"discipline":"claim control","heading":"Keep claims honest.","id":"claims","mission":"The Claims review checks the boundary around every published conclusion: what was measured, which vector it belongs to, whether evidence was complete, which causal assumptions remain unsettled, and what the result cannot establish. It refuses to turn six review findings into approval, causal proof, a composite score, or a safety rating.","name":"Claims","role":"Claims","summary":"Refuses approval, causal-proof, composite, and safety claims; a vector no-match is never a safety rating.","version":"2.0.0"}]}