{"path":"research/corpus-graph/cards/what-the-judges-wrote-for-each-pair.md","content":"# What two outside judges wrote for each pair\n\n*Generated by `card_match_test.py collect --write` from the two reports kept word for word beside this\nfile, their keys, and the readers' reasons in the index. Change nothing here by hand. Why it is kept,\nand how to read the pairs a judge refused: [../likeness-seen-from-outside.md](../likeness-seen-from-outside.md).*\n\n34 pairs that were not decoys: 15 same move, same reason; 3 same move, another reason; 16 not the same move. 20 decoys were mixed in unmarked, and 0 of them were accepted.\n\n**How to read the last group.** The judges were asked *do the two passages make the same move*, and\nwere asked what carries over only after a yes to that and to *for the same reason*. So for a pair in\nthe last group nothing was asked about what one side could take from the other. Where a reader had\ntied the pair, the reader's own reason says what it thought carries over, and no outside party has\njudged that yet.\n\n## Same move, same reason (15)\n\n### `docs/research/the-load-bearing-unsaid.md:41` and `docs/research/kulturkapital-and-the-unscalable.md:33`\n\n- tied by a reader; the first judge, pair J5\n- **The reader's own reason:** Pattern: putting a price or a weight on something held outside exchange is itself felt as a violation, whatever the number, and changes the thing. Tetlock's contemplation effect and the practitioners' remark that being paid makes wanted work feel dirty are one effect in two settings. For the sacredness brake this widens the scope: any intrinsically held contribution, not only sacred values, is damaged by a weighing or a payment. The brake then supports the no-reputation-currency rule, and contributor-incentive design can use the taboo-trade-off findings.\n- **The first passage, as the judge read it:** Says that a value which lives by not being traded cannot be put into a weighing without that act itself being the violation, so the weighing must not be done to it.\n- **The second:** Says that an act which lives by inner motive cannot be put on payment without the payment destroying the motive, so payment must not be applied to it.\n- **What carries over:** The first passage states that merely considering the trade-off is already the harm, with outrage extending to those who only considered it, while the second states the harm comes from payment actually offered and only where the motive is truly inner.\n- **Where the likeness stops:** In the first the thing damaged is the standing of a value in a represented text, judged by observers; in the second it is an actor's motivation to contribute.\n\n### `docs/depth.md:118` and `docs/research/how-a-map-is-cut-and-what-it-shows.md:230`\n\n- tied by a reader; the first judge, pair J8\n- **The reader's own reason:** Pattern: ranking by the attention an item already has hides exactly the items that are valuable because nobody has looked yet. The depth page warns that gating decomposition on contestation misses uncontested premises. The map note measures the same trap on display: the first view of a densely linked source is almost all head and shows 1 of 3 unspoken premises, the crux items. Whoever sets the decomposition budget could borrow its remedy: a reserved, marked share for the tail, plus a line saying what was left out.\n- **The first passage, as the judge read it:** Rejects gating work by what has already been contested, because the most valuable items are the ones nobody has contested since nobody could see them yet, and that gate selects on the rarest behaviour.\n- **The second:** Shows by measurement that ranking a first view by connectedness fills it with the well-linked head and leaves out the unspoken premises that are the crux, which are thinly linked or absent.\n- **What carries over:** The first passage limits the rule to its proper use (fine for prioritising machine spend, wrong as a rule about what the graph should contain), while the second shows the failure depends on the source's link density (eleven-twelfths head on a densely linked source, all tail on a uniformly thin one).\n- **Where the likeness stops:** The first is a gate on what gets decomposed and fails through lack of contestation; the second is a ranking for display and fails through thin links.\n\n### `docs/research/the-load-bearing-unsaid.md:136` and `docs/research/what-a-record-linkage-project-teaches-deliberus.md:344`\n\n- tied by a reader; the first judge, pair J10\n- **The reader's own reason:** Pattern: what cannot be rebuilt from what survives is the ground, so it is kept untouched and stamped with when it was captured, while everything derived from it may be recomputed or dropped. The unsaid note found this by accident: the extractor kept only what it judged to be argument, the frame was gone, and eight sources could not be recovered. The record-linkage note states it as a standing sort (ground, judgment, cache, rules). Whoever extends the archive would ask that question of every other store and add a third stamp: when the answer was derived, under which rules.\n- **The first passage, as the judge read it:** Makes the raw original write-once and stamps it with the capture time, because a later backfill is a different object and an archive that cannot tell them apart misleads quietly.\n- **The second:** Sorts stored things by whether they could be restated from what survives, never touching what cannot be rederived, keeping the origin of judgments, and changing the rules only in the open.\n- **What carries over:** The first passage states that a reconstruction made later is a different object from an observation made at the time and needs its own timestamp, and that some originals (eight sources) are simply lost; the second states a graded policy (ground, judgment, cache, rules) and asks who may change the rules and on whose word.\n- **Where the likeness stops:** The first protects one kind of record (a raw source snapshot), while the second grades several layers of derived data and extends to governance of the rules.\n\n### `docs/research/the-load-bearing-unsaid.md:27` and `docs/research/writing-out-the-unwritten.md:180`\n\n- tied by a reader; the first judge, pair J15\n- **The reader's own reason:** Pattern: the premises that hold an argument up are the ones the writing process leaves out, and a reader or model trained on the written record inherits the gap. The unsaid note explains omission by presumed sharedness and says it is not laziness. This note adds a second cause from another field, the cost of writing: textbooks keep conclusions and drop derivations, and a model repeats the compressed form it was trained on. Whoever works on the extraction bias would add cost-of-writing as a cause and treat training on compressed text as a source of the same gradient.\n- **The first passage, as the judge read it:** Says that a system reading only what is written is structurally biased against the most important content, because that content is what goes unwritten.\n- **The second:** Says written knowledge is a compression that keeps conclusions and drops the derivations beneath them, so anything trained on or reading the written form inherits the gap.\n- **What carries over:** The second passage states what the gap costs (a claim cannot be checked without trusting an authority, and links between fields are cut), while the first states that the bias is structural and cannot be tuned away.\n- **Where the likeness stops:** The second concerns derivations that could in principle be written down, whereas the first extends to content that cannot be said in that register at all.\n\n### `docs/worldview-lenses.md:82` and `docs/research/does-descending-change-the-descender.md:117`\n\n- tied by a reader; the first judge, pair J16\n- **The reader's own reason:** Pattern: receiving a finished rendering without producing it leaves the feeling of having understood and none of the change; only generating it does the work. Interpassivity is the effect the explanatory-depth literature measures: the illusion is punctured by generating an explanation, not by reading one. Whoever designs lens switching would stop treating 'wear their glasses' as a viewing mode and make it a generating one, for example stating what this lens-holder would conclude before seeing it, then comparing.\n- **The first passage, as the judge read it:** Warns that a tool performing the perspective-taking for the user yields the feeling of having transcended one's standpoint without the substance, and so requires discomfort the user must undergo.\n- **The second:** Says the sense of understanding is punctured only by generating an explanation oneself, and that reading a finished explanation can raise felt understanding without raising real understanding.\n- **What carries over:** The second passage states a concrete boundary of the effect, that the benefit belongs only to the person who generates and so not to a reader of a finished graph or a read-only tour, while the first states a design remedy of marking structural disagreement so the tool does not reassure.\n- **Where the likeness stops:** The first concerns an ideological illusion of having moved past one's own premises, remedied by discomfort, and the second an illusion of explanatory depth, remedied by self-generation.\n\n### `docs/depth.md:169` and `docs/research/forecasting-room-at-the-top-and-deliberus.md:15`\n\n- tied by a reader; the first judge, pair J17\n- **The reader's own reason:** Pattern: a limit on what more effort can extract is stated only as the remainder after independent attempts to get past it, bounded by anchors, with a stated chance that the answer disappoints. The depth page says the end of decomposition is earned by many honest failed attempts. Alexander's forecasting ceiling is bounded by three independent anchors. Whoever sharpens the depth criterion would specify what counts as an independent attempt (other decomposers, a stronger hand-run as the ceiling) and report the residue as a measured fraction with its anchors.\n- **The first passage, as the judge read it:** Defines the bottom of a descent by the record of repeated honest failed attempts to decompose a claim, not by declaration, and says this same record is what would falsify the thesis.\n- **The second:** Bounds an irreducible remainder by independent empirical anchors rather than asserting it, keeping open the result that disappoints the thesis, and notes the remainder can be typed.\n- **What carries over:** The first passage gives an operational stopping criterion for a single claim (repeated community attempts, each child withdrawn), while the second states that triangulating several independent anchors bounds a ceiling and that typed residues say what kind of wall was hit, which a bare score cannot.\n- **Where the likeness stops:** The first fixes a fallible terminus claim by claim, while the second estimates an aggregate ceiling from outside data.\n\n### `docs/worldview-lenses.md:74` and `docs/research/slicing-stability-first-run.md:26`\n\n- tied by a reader; the first judge, pair J19\n- **The reader's own reason:** Pattern: vary the contestable input across its reasonable range and read the result off the spread: what survives every setting is robust, what flips marks the one choice it hangs on. The worldview page proposes this for priors; the slicing study already ran it on filings (91% of verdicts stable, both sensitive claims resting on one unrecorded choice). Whoever builds the lens readout can reuse that read-only recomputation, report stability and spread, name the choice that flips, and treat a setting that yields no verdict as silence, not a rival verdict.\n- **The first passage, as the judge read it:** Reads a conclusion as robust if it holds under every reasonable set of priors, and a flip between lenses as showing exactly where two worldviews part.\n- **The second:** Recomputes each decomposed claim under every reasonable filing choice, finds the score survives for nearly all, and traces the few that move to one unrecorded choice.\n- **What carries over:** The second passage states the sensitivity was one unrecorded choice with a known remedy (record the interpretation), so it is removable rather than informative, while the first treats a flip as real information about where worldviews part.\n- **Where the likeness stops:** The first varies a holder's base rates, a substantive disagreement to be displayed, while the second varies bookkeeping conventions that should be eliminated by recording them.\n\n### `docs/research/the-load-bearing-unsaid.md:136` and `docs/research/supersession-and-bitemporal-lifecycles.md:12`\n\n- tied by a reader; the first judge, pair J23\n- **The reader's own reason:** Pattern: a record needs two clocks, when the thing was so and when the system learned it, and a snapshot taken later is a different object from one taken then. The unsaid note stamps the fetch time and warns about backfill. The bitemporal note names the axes: valid time and transaction time. Whoever fixes the archive would label each snapshot with the time it claims to reflect as well as the time it was fetched, so a backfilled copy can never stand in for the page as extracted.\n- **The first passage, as the judge read it:** Stamps the raw snapshot with the time the system captured it, apart from what it is about, because a copy backfilled later is a different object from one taken as the pipeline ran.\n- **The second:** Keeps a separate record of when the system learned a fact and retired the record, apart from when the fact held in the world, and never deletes the record.\n- **What carries over:** The second passage states that the system-time axis is independent of the world-time axis and pairs the learned stamp with a retirement stamp so that history can still be queried, while the first states that a late backfill is a different kind of object and that the original must be write-once.\n- **Where the likeness stops:** The first stamps one acquisition time on an unchanging raw archive, while the second runs two axes plus retirement over facts that change.\n\n### `docs/depth.md:114` and `docs/research/the-harness-conviction.md:256`\n\n- tied by a reader; the first judge, pair J25\n- **The reader's own reason:** Pattern: a check built from the internal consistency of what was written cannot see what was never written; omission is caught only by tracing back to the source. The harness note's ledger passage: books balance with a transaction missing, which is why audit vouches against outside documents. The depth passage: coverage fails silently, and the recomposition check works because it is compared with the parent. Whoever adds decomposition checks would see that child-only checks (coherence among parts, sum rules) are the balance, never the audit, and would test completeness from the parent toward the parts.\n- **The first passage, as the judge read it:** Adds a check that rebuilds a parent from its children alone and reports what went missing, because omission fails silently and nothing downstream can see what was never written.\n- **The second:** Notes that balancing the books catches inconsistency but never omission, and so adds an audit against external documents and direct confirmation from the counterparty.\n- **What carries over:** The second passage states that the reference must come from outside the books, with confirmation sought directly from the one party who cannot be outvoted, while the first states that coverage, coherence and atomicity are separate dimensions and that the check turns a loose aspiration into a number.\n- **Where the likeness stops:** The first compares against the parent claim, which is itself text inside the same graph, so it recovers only what the parent asserted, while the second compares against evidence outside the system.\n\n### `docs/worldview-lenses.md:26` and `docs/research/fractal-priors-for-convergence.md:31`\n\n- tied by a reader; the first judge, pair J26\n- **The reader's own reason:** Pattern: parties share a short list of things that matter and differ in how much each counts, so a lens is a weighting over a shared list. The worldview page states it for two politics; the gallery finds it from cuisine to chess. Whoever works on the lens dials would take the shared list as what to discover bottom-up, and note the gallery's limit: it gives the shape of disagreement, not its size, which qualifies this page's claim that a located disagreement is almost always much smaller than it felt.\n- **The first passage, as the judge read it:** Locates disagreement in the weights placed on a shared set of premises, so the same premises produce different conclusions.\n- **The second:** Says that across many domains people work with the same short list of things that matter and differ in how much each counts, so the list is the structure and the weighting is the difference.\n- **What carries over:** The second passage states the pattern as a regularity across many domains, with an example where no one has a stake (taste dimensions in cuisine), while the first states the use made of it, that the system can say \"given your weights this follows\".\n- **Where the likeness stops:** The first also lets definitions and evidence weights differ between lenses, so the list itself is not fully shared there, while the second holds the list fixed.\n\n### `docs/research/reflection-and-the-graph.md:71` and `docs/research/proving-a-pass-reliable.md:304`\n\n- proposed by code from two cards written apart; the second judge, pair J3; the two cards shared the name stem(s) option, reject\n- **The first passage, as the judge read it:** Scores a machine pass so that a confessed does-not-fit strictly beats a wrong nearest-fit, so that a decline option actually gets used instead of a forced guess.\n- **The second:** Lets the machine decline below a confidence threshold and proves the agreement bound only on what it does not decline, at the price of coverage.\n- **What carries over:** The abstention passage states how the threshold is fixed and what it costs (set on about 500 human-labelled items with an exact binomial bound, the judge answering only 56 to 63 percent, the declined items being the contested ones), while the reflection passage states the failure condition that the decline option stays silent unless declining pays strictly better than a wrong fit, and a costless shrug beside it takes the traffic.\n- **Where the likeness stops:** The first is about making a categorical does-not-fit value attractive to the pass, while the second assumes a confidence score and threshold and gives a statistical guarantee on the answered items.\n\n### `docs/bridging.md:5` and `docs/research/simplenote-archive-analysis.md:56`\n\n- proposed by code from two cards written apart; the second judge, pair J4; the two cards shared the name stem(s) agreem\n- **The first passage, as the judge read it:** Says agreement on a statement or conclusion does not show why people agree, and looks instead at whether the reasoning structure is compelling across the divide.\n- **The second:** Says two people can share a conclusion through different reasoning paths, so conclusion-level agreement can mask real disagreement, and asks whether the paths match.\n- **What carries over:** The bridging passage names what conclusion-level agreement might actually be (shared values, ambiguity each side reads differently, or lowest-common-denominator vagueness), while the simplenote passage states the cost of not looking, that the masked disagreement surfaces later under different conditions, with the Swedish child-poverty debate as the case.\n- **Where the likeness stops:** Bridging hunts for agreement on structure despite disagreement on conclusions, while the simplenote passage warns of disagreement on structure despite agreement on conclusions.\n\n### `docs/research/taxonomy-gaps-and-the-closed-enum.md:90` and `docs/research/what-counts-as-evidence/testimony-experience-and-normative-evidence.md:306`\n\n- proposed by code from two cards written apart; the second judge, pair J7; the two cards shared the name stem(s) compat, incent\n- **The first passage, as the judge read it:** Says a does-not-fit value works as a confession channel only to the degree that confessing beats a forced nearest-fit for whoever presses it, and proposes scoring every pass so it does.\n- **The second:** Says a strictly proper scoring rule makes honest reporting of belief the optimal strategy, the formal version of confession must pay.\n- **What carries over:** The scoring-rule passage gives the exact condition under which honesty is optimal (strict properness, which Brier and log scores meet and linear scoring does not), while the taxonomy passage names the failure mode that the channel goes unused when confessing pays less than a forced fit, and leaves the rule as a candidate, unruled.\n- **Where the likeness stops:** Proper scoring rules score probability reports against known outcomes, while does-not-fit is a categorical abstention with no ground truth to score it against.\n\n### `docs/research/weighing-detection-semantic-tier.md:7` and `docs/research/the-remaining-case-for-the-stance-guard.md:39`\n\n- proposed by code from two cards written apart; the second judge, pair J17; the two cards shared the name stem(s) review\n- **The first passage, as the judge read it:** Argues that a lexical detector misses implicit cases by construction and should be backed by an LLM semantic tier, keeping what determinism buys in a two-tier shape.\n- **The second:** Argues that cases a pattern detector missed cannot be minted downstream, so a downstream symptom is used to send the pair to an LLM stance re-read that repairs the edge.\n- **What carries over:** The stance passage states a trigger and repair rule (the re-read runs only on pairs flagged by an independent symptom, and when confident it supersedes the edge while keeping the original visible), while the weighing passage states what determinism still buys and that routing LLM-detected output needs a fresh decision because the earlier ratification assumed a deterministic detector.\n- **Where the likeness stops:** The first adds a semantic tier at detection time across the board, while the second is an after-the-fact repair aimed at one residual miss class.\n\n### `docs/research/session39-what-the-graph-stopped-claiming.md:105` and `docs/research/derivability-of-missing-considerations.md:263`\n\n- proposed by code from two cards written apart; the second judge, pair J25; the two cards shared the name stem(s) explic\n- **The first passage, as the judge read it:** Says that when nothing has checked a recorded fact the system should store a blank that says so, never a plausible stand-in.\n- **The second:** Says a does-not-fit or missing-premise slot must be an explicit output field, because models suspect a missing premise yet do not abstain on their own initiative.\n- **What carries over:** The derivability passage gives the evidence that the blank cannot be optional (reasoning models detect a missing premise early and keep generating anyway, while non-reasoning models abstain better), while the session passage gives the scope and the companion rule that a machine verdict of not-X licenses retiring X and not writing a Y.\n- **Where the likeness stops:** The first concerns what is stored in the graph when a recorded fact is unchecked, while the second concerns the output schema of a generation pass.\n\n## Same move, another reason (3)\n\n### `docs/worldview-lenses.md:70` and `docs/research/store-the-uses-read-the-meaning.md:30`\n\n- tied by a reader; the first judge, pair J20\n- **The reader's own reason:** Pattern: store the observations of use and read the category off them later; never store the category as if it were a datum, because the label carries a stance. The worldview page resolves its dials problem this way: groupings emerge from evaluations, not declared dials. The storage note states the same law for word senses and horizon kinds. Whoever builds lenses would store only the evaluations, treat any profile as a dated reading under a named rule, and recompute it, rather than keep the lens as a field on the person.\n- **The first passage, as the judge read it:** Declines to declare worldview categories in advance and computes them from the pattern of what people actually evaluate.\n- **The second:** Declines to store a word's meaning, stores the recorded uses, and reads the meaning (or kind) off those uses, never setting it.\n- **Where the likeness stops:** The first holds back because a preset taxonomy is itself a stance, while the second holds back because meaning moves with use and a stored label goes stale.\n\n### `docs/research/the-motte-and-the-periphery.md:223` and `docs/research/the-group-premise-evidence.md:33`\n\n- proposed by code from two cards written apart; the second judge, pair J8; the two cards shared the name stem(s) delphi\n- **The first passage, as the judge read it:** Collects each person's view privately before the group hears anyone's, so the pre-exposure view serves as a baseline and the later shift measures the audience effect.\n- **The second:** Requires independent estimates before pooling because showing people others' estimates narrows the spread of answers without improving accuracy.\n- **Where the likeness stops:** The first uses private-first elicitation as an instrument to measure the influence of exposure, while the second uses it as a condition to protect a crowd's diversity and accuracy.\n\n### `docs/research/the-load-bearing-unsaid.md:27` and `docs/research/how-a-map-is-cut-and-what-it-shows.md:230`\n\n- proposed by code from two cards written apart; the second judge, pair J28; the two cards shared the name stem(s) select\n- **The first passage, as the judge read it:** Concludes that an extraction which reads only what is written is structurally biased against the most load-bearing content, because that content is least likely to be written.\n- **The second:** Shows that a first-view map of the best-connected items is mostly head and leaves out the unspoken premises that are the crux until later zoom steps.\n- **Where the likeness stops:** The first is about content the authors never wrote, while the second is about content already minted in the graph that the display ranking leaves out.\n\n## Not the same move (16)\n\n### `docs/depth.md:191` and `docs/research/major-transitions-heredity-and-evolvability.md:177`\n\n- tied by a reader; the first judge, pair J3\n- **The reader's own reason:** Szathmáry's replicator classes show the same mechanism in biology: heredity moves from holistic to modular, copying discrete parts so one can change without redoing the rest. The reading attaches a condition the page lacks. Modular heredity needs units whose identity is cheap to decide, and here identity is judged, so its verdict is \"Conditional on claim identity\". Near-decomposability helps a stranger only if the system can tell when a premise recurs under two parents. That is where the page places its own boundary.\n- **The first passage, as the judge read it:** Argues that a layered structure, whose parts interact mostly within themselves, is what lets one part be revised without recomputing everything above it.\n- **The second:** Tests a borrowed biological mechanism against four checks (recurrence, mechanism identity, adversary, substrate condition) and gives a conditional verdict on whether the analogy holds.\n\n### `docs/bridging.md:91` and `docs/research/legibility-under-power-red-team.md:54`\n\n- tied by a reader; the first judge, pair J4\n- **The reader's own reason:** Hole 2 argues some agreements survive only because they stay vague. UN Resolution 242 dropped \"the\" on purpose. The Good Friday text let each side read it differently. Sunstein's incompletely theorized agreements are a source of stability. It adds that showing why something bridges, this page's stated advantage at line 85, is what makes divergence common knowledge and can break the agreement. The Vagueness Trap treats vagueness only as a risk. The corpus also holds that it can be the achievement.\n- **The first passage, as the judge read it:** Warns that rewarding cross-group agreement selects for statements that commit to nothing, and counters with a structure that forces specific claims.\n- **The second:** Shows that deliberately imprecise wording lets parties with incompatible readings sign one text, and that making the divergence explicit would destroy that agreement.\n\n### `docs/bridging.md:91` and `docs/research/prime-real-estate-words.md:307`\n\n- tied by a reader; the first judge, pair J9\n- **The reader's own reason:** Pattern: assent can be universal because each assenter reads their own meaning into the words, so universal assent is no evidence of shared content. The bridging page's trap is a slogan that bridges every group; the wanted-words note shows a word many people want passes the stranger test falsely, each stranger understanding their own sense. Bridging's countermeasure, preferring empirical claims, misses wanted words inside empirical-looking claims. Someone fixing it would flag key words readers already own a sense of, and show each group's reading beside the score.\n- **The first passage, as the judge read it:** Warns that rewarding cross-group agreement selects for vacuous statements, and counters with specific claims that narrow the disagreement.\n- **The second:** Warns that a key word can pass a comprehension test falsely because each reader fills it with a sense of their own, and adds a check question asking whether readers already bring their own meaning to it.\n\n### `docs/bridging.md:5` and `docs/research/legibility-under-power-red-team.md:56`\n\n- tied by a reader; the first judge, pair J12\n- **The reader's own reason:** Pattern: agreement on a text does not show agreement on what it means, and sometimes the gap is what makes the agreement possible. The bridging page treats differently-read agreement as a risk to filter out. The red team, from diplomacy and law, calls the same thing a fudge and asks when it is a bug and when civic technology. Whoever builds the bridging feed should not read 'each side reads it differently' as always a defect, nor surface the why in a way that detonates a working agreement.\n- **The first passage, as the judge read it:** Separates agreement on a statement from agreement on the reasoning beneath it, since statement-level agreement cannot tell shared values from differently read ambiguity or vagueness.\n- **The second:** Presents a signed text that holds incompatible constructions as the means by which agreement was reached, with defenders calling this necessary skill and critics saying the parties did not really agree.\n\n### `docs/bridging.md:75` and `docs/research/what-counts-as-evidence.md:369`\n\n- tied by a reader; the first judge, pair J13\n- **The reader's own reason:** Pattern: a verdict from a party whose position predicts the opposite verdict cannot be explained by that position, so it is worth more than the same verdict from an ally. Bridging surfaces arguments that even the opposition rates well-reasoned. The evidence note files this as the hearsay exception for statements against interest, and adds that corroboration counts most from people who otherwise disagree, provided it is independent. Whoever builds the bridging score would treat cross-group endorsement as a fact that raises a starting weight, test that the groups are truly adverse, and count shared-source endorsers once.\n- **The first passage, as the judge read it:** Surfaces the opposing group's strongest arguments, those rated well-reasoned even by the opposition, as the starting point for engagement, in place of a physical deliberative poll.\n- **The second:** Treats a fact that fits a recognised hearsay exception as itself a claim that argues for a higher starting weight, with answering critical questions playing the role of cross-examination.\n\n### `docs/research/the-residual-error-taxonomy.md:28` and `docs/research/dogfood-run-4-agent-surface.md:103`\n\n- proposed by code from two cards written apart; the second judge, pair J1; the two cards shared the name stem(s) failur\n- **The first passage, as the judge read it:** Holds that many locally plausible steps from one generator are one sample repeated, so a consistent tilt passes every per-step check and same-source agreement is a weak signal.\n- **The second:** Concludes that a system with no way to confess failure reports success, because the surfaces reporting on the instruments had no instrument of their own.\n\n### `docs/worldview-lenses.md:26` and `docs/research/frontier-extraction-experiment.md:40`\n\n- proposed by code from two cards written apart; the second judge, pair J2; the two cards shared the name stem(s) double\n- **The first passage, as the judge read it:** Shows that the same shared premises, weighted differently, yield different conclusions, so the disagreement sits in the weights and not in the premises.\n- **The second:** Shows that the decisive clash of a debate lies in premises that neither side wrote down.\n\n### `docs/research/active-inference-context-acquisition-and-deliberus.md:17` and `docs/inductive-reasoning-research.md:275`\n\n- proposed by code from two cards written apart; the second judge, pair J5; the two cards shared the name stem(s) inform\n- **The first passage, as the judge read it:** Stops offering clarifying questions once the expected swing from a decisive answer falls below the price of a human's attention.\n- **The second:** Poses the open question of what starting value a brand-new claim should have, listing candidate sources for a prior.\n\n### `docs/research/the-load-bearing-unsaid.md:3` and `docs/research/weakest-link-arithmetic-and-the-merge-hunch.md:26`\n\n- proposed by code from two cards written apart; the second judge, pair J16; the two cards shared the name stem(s) consil\n- **The first passage, as the judge read it:** Frames the research question of whether two findings that looked separate share one mechanism.\n- **The second:** Notes that every field which lets part contributions add found the same attack and built the same two defence families, and treats that recurrence as the document's spine.\n\n### `docs/research/structuring-gradient-lineage.md:13` and `docs/research/decomposing-value.md:156`\n\n- proposed by code from two cards written apart; the second judge, pair J18; the two cards shared the name stem(s) critic, source\n- **The first passage, as the judge read it:** Reports that an intention is documented across seven surfaces, each time as a gap, and that the older argument had been the position the founder was argued into.\n- **The second:** Reports that a claim about the Max-Neef needs quotation was checked against the source and held almost verbatim.\n\n### `docs/research/steadying-an-outside-judge.md:35` and `docs/research/science-reasoning-infrastructure.md:115`\n\n- proposed by code from two cards written apart; the second judge, pair J19; the two cards shared the name stem(s) prereg\n- **The first passage, as the judge read it:** Proposes a with-and-without comparison of a judge's flip rate under fabricated consensus, with the prediction registered before the run.\n- **The second:** Reports that registered reports show 44 percent positive results against 96 percent in standard reports, revealing result-contingent publication.\n\n### `docs/research/what-a-link-rests-on.md:316` and `docs/research/forecasting-room-at-the-top-and-deliberus.md:15`\n\n- proposed by code from two cards written apart; the second judge, pair J21; the two cards shared the name stem(s) triang\n- **The first passage, as the judge read it:** Trusts no single machine naming of a link but treats a claim that recurs under several separate readings as a finding.\n- **The second:** Bounds the irreducible remainder of forecasting empirically, with independent anchors, and maps that remainder onto the residue map for normative conflict.\n\n### `docs/research/scheme-set-exhaustiveness.md:297` and `docs/research/what-a-link-strength-means.md:51`\n\n- proposed by code from two cards written apart; the second judge, pair J22; the two cards shared the name stem(s) validi\n- **The first passage, as the judge read it:** States the limits of a measurement (one model, low effort, few readings, most no-fit links from one source).\n- **The second:** Notes that a link-strength number was requested with no rubric, so nothing exists for it to be calibrated against.\n\n### `docs/research/ai-risk-for-mathematicians-and-the-coordination-game.md:91` and `docs/research/soft-canonical-clustering-and-reversible-merge-semantics.md:391`\n\n- proposed by code from two cards written apart; the second judge, pair J23; the two cards shared the name stem(s) counti, double\n- **The first passage, as the judge read it:** Notes that the crux of a debate is routinely left unsaid by at least one side, and that this essay states outright the premises an earlier run had to write in.\n- **The second:** Gives the test for double-counting as more than one path from an evidence node to a single aggregation point, so merge covers only the duplicate-node case.\n\n### `docs/research/what-the-walkthroughs-and-the-reviews-are-for.md:113` and `docs/research/simplenote-archive-analysis.md:56`\n\n- proposed by code from two cards written apart; the second judge, pair J24; the two cards shared the name stem(s) agreem, consen, incomp, overla, theori\n- **The first passage, as the judge read it:** Proposes a map feature that shows two opponents where they already agree, using a commission that agreed on cases and split on principles.\n- **The second:** Warns that agreement on a conclusion can be illusory because the reasoning paths differ, masking disagreement that surfaces later.\n\n### `docs/worldview-lenses.md:26` and `docs/research/frontier-extraction-experiment.md:60`\n\n- proposed by code from two cards written apart; the second judge, pair J26; the two cards shared the name stem(s) double\n- **The first passage, as the judge read it:** Shows that the same shared premises, weighted differently, yield different conclusions, so the disagreement sits in the weights and not in the premises.\n- **The second:** Shows across two debates that the load-bearing node is routinely the one that at least one side never says out loud.\n"}