The refusal register · design-v4 6f56541
Every withhold carries a reason class, a substitute and a sentence written by the thing that produced the number. This is the census of all of them.
Anyone can publish a number. To publish a refusal you must have measured the thing, and have written the bar down before you measured it.The sixty-second version →
How many figures does the estate refuse to publish, with a written reason?83of 83 every one carries a reason class, a substitute and a sentence
How many trained model artifacts were built and never wired to anything?21of 147 DARK — held, measured, not shipped
Of the data the estate holds under a custody receipt, how much may it serve?0of 95 receipts over 65 distinct sources
The estate's own definition of every reason it refuses a figure, verbatim, opened one at a time.
The eleventh class, LEGAL_OR_LICENCE, has zero members and is kept anyway — defined so that the first refusal that needs it has somewhere to land rather than being filed under something close. An empty class held open is itself a statement about what the estate has not yet had to refuse.
The metric is arithmetically sound but scores a different quantity from the one its name promises. The reader's inference, not the computation, is what fails. Canonical shape: a win rate whose denominator is mostly matters the opponent never answered measures how often somebody walked away.
There is no defensible number, because a precondition of measuring one was never established: no as-of feature horizon, no persisted score/label custody, no bound baseline. Distinct from BELOW_BAR, which HAS a number.
Measured honestly, reported at the right grain, and did not clear a bar that was registered BEFORE the measurement. The number exists and is defensible as evidence; it is the promotion to a product claim that fails.
This figure is not itself the held claim, but rendering it lets a reader recompute one. Quantized tiers, orderings, per-bin counts, derived bands. A guard here protects a DIFFERENT key, so its sentence must name that key.
The quantity exists but not at the grain the surface asks for. Either the cell's n is too small to distinguish it from its own population rate, or the pooled figure averages strata that differ so much that the average measures which stratum you are in.
A later, corrected value replaced this one for the same claim. The old number is not wrong-in-a-vacuum, it is stale, and its danger is that it is quotable-looking. Every member of this class has a live successor that must be shown in its place.
An operator or policy decision removed the claim from the product. The evidence may be internally sound; the estate has decided not to sell it. The withhold is a business posture, not a measurement verdict, and must be described as such.
The producing model or pipeline is broken, so its figure measures the defect rather than the claim. Inverted calibrators, leakage-inflated features, join multiplication. Withholding here is not caution; publishing would be a false statement.
The live layer withholds or has corrected the value, but a frozen capture under /static/_snapshots still serves the old bytes off a mount that bypasses every query-layer and template-layer control.
The statistics are fine and the model works; the INPUTS cannot reach decision time. A feature contract that needs T-118 days of coverage against a source that reaches T-846 days cannot score a live matter at any bar.
Publication is constrained by an acquisition licence, terms of use, or a third-party right, independent of whether the number is correct. ZERO members at this commit; retained so the first such key has a class.
Stated here and nowhere else on the page.
A published refusal is a competence claim, and it is the only competence claim in this market that cannot be faked cheaply.
Anyone can publish a number. To publish a refusal you must first have measured the thing, then have a bar you registered before you measured it, then have decided in public that you did not clear it. The register below shows fourteen figures refused as BELOW_BAR: measured honestly, reported at the right grain, and short of a threshold written down in advance. Seventeen are refused as TARGET_SUBSTITUTION — arithmetically sound, but scoring a different quantity from the one the name promises.
That distinction is the product. A win rate whose denominator is mostly matters the opponent never answered measures how often somebody walked away. It is not wrong; it is not the thing you were about to buy. The estate can name that failure because it went looking for it in its own numbers.
Which is why the count that matters most on this page is 0 of 95. Ninety-five custody receipts record where data came from and under what terms it was taken. Not one of them records a right to republish it. That is not a gap in the paperwork — it is the paperwork working. A vendor who cannot tell you which of their sources they are allowed to serve has not read their sources.
The bet this page makes is that a buyer reads a published refusal as competence rather than as incapacity. It has not been tested on a buyer. The measurement is how many priced-offer conversations cite this page unprompted; “they said it was interesting” is not the measurement.
A peer lane is finding that most withholds rendered on the specimen are vacuous — the withhold renders with no real content behind it. So this atlas was audited against itself before it was published.
frozenset[str] that carries names and nothing else.Read the top two rows and the bottom three together. The grammar is not vacuous: every substitute it names resolves, every detector it declares has patterns to match. The specimen is a different mechanism entirely, and it is where the peer lane's finding lives.
The reasons exist — as # comments above each key in specimen_registry.py. A person reading the source can read them. No renderer, template or API can.
This distribution is an inference and is labelled as one. The estate assigns these keys no class. The bars above are this lane reading 46 source comments — “superseded by predictive_interval_95”, “n_cohort=21 < declared min_band_n=30”, “per-molecule n=1; the KM quantity is pooled over 1,359 at-risk NDAs”, “pinned 53-feature contract satisfied for 6/689 live matters (coverage 0.008708)” — and posting each against the class it plainly belongs to. Every one of those reasons is good. None of them can reach a reader. That is the whole defect, and the repair is a data structure, not a decision.
The same register, read in three trees within the same hour.
All three files stamp generated_at_commit: 218909b69fca. The stamp did not move when the keys did: five venues-* keys were added to the working copy at 02:57 today and inherited a provenance pointer to a commit that predates them. A stamp that does not move is a claim, not a receipt. This page pins design-v4 6f56541 and says so, because a census that quietly reads whichever copy is on disk reports a number nobody can reproduce.
Sortable by class, by how many surfaces it governs, and by what is published in its place. Each key opens the first surface it governs.
The key, the reason it is refused, what is published in its place, and the surface it would have appeared on.
| Refused key | Reason class | Published in its place | Surfaces | First surface | |
|---|---|---|---|---|---|
| appellate-ccad-strict-auc | BELOW BAR | The full released unit, not a bare scalar: 0.6902 stated as BELOW its 0.70 pre-registered bar, the within-circuit correction… | 5 | /appellate/ccad | |
| brands-memo-ext-opp-second-witness-absent synthetic | BELOW BAR | The descriptive record the rate was computed over, named rather than quoted: the extension-of-time filings and the oppositions… | 0 | no live route | |
| brands-memo-ext-opp-second-witness-checks-failed synthetic | BELOW BAR | The descriptive record the rate was computed over, named rather than quoted: the extension-of-time filings and the oppositions… | 0 | no live route | |
| brands-memo-ext-opp-second-witness-differ synthetic | BELOW BAR | The descriptive record the rate was computed over, named rather than quoted: the extension-of-time filings and the oppositions… | 0 | no live route | |
| brands-memo-ext-opp-second-witness-unreadable synthetic | BELOW BAR | The descriptive record the rate was computed over, named rather than quoted: the extension-of-time filings and the oppositions… | 0 | no live route | |
| brands-memo-ext-opp-second-witness-verdict-unrecognised synthetic | BELOW BAR | The descriptive record the rate was computed over, named rather than quoted: the extension-of-time filings and the oppositions… | 0 | no live route | |
| brands-memo-ext-opp-unadjudicated synthetic | BELOW BAR | The descriptive record the rate was computed over, named rather than quoted: the extension-of-time filings and the oppositions… | 0 | no live route | |
| brands-score-family synthetic | BELOW BAR | The withheld-score notice text from litan.ui.brands_score_withholding.WITHHELD_NOTICE, plus the descriptive TTAB record the… | 2 | /brands | |
| patent-chain-entity-mismatch-patents synthetic | BELOW BAR | None as a bare number. Member of patent-chain-of-title-per-class-defect-counts; inherits the head's coverage-pair substitute. | 2 | / | |
| patent-chain-npt-after-complaint-patents synthetic | BELOW BAR | None as a bare number. This is a member of the per-class family head patent-chain-of-title-per-class-defect-counts and… | 2 | / | |
| patent-chain-of-title-per-class-defect-counts | BELOW BAR | The assignment COVERAGE pair, which §7 item 5 of the source study independently requires on the surface above any flag:… | 2 | / | |
| patent-chain-title-after-complaint-patents synthetic | BELOW BAR | None as a bare number. Member of patent-chain-of-title-per-class-defect-counts; inherits the head's coverage-pair substitute. | 2 | / | |
| ptab-institutional-auc | BELOW BAR | The observed institution base rate on the same cohort with its denominator, plus the explicit statement that the model's… | 4 | /validation-record | |
| ttab-m0-exante-auc | BELOW BAR | The string WITHDRAWN in the auc field, with a structured auc_withheld object carrying metric_id, state, mechanism, reason_code… | 3 | /api/predict-market/track-record/v1 | |
| ttab-answer-forecast-r1-probability synthetic | COVERAGE UNSERVABLE | Nothing per-matter. The observed answer base rate 0.3502 and the published held-out decile table (10 deciles with n, observed… | 0 | no live route | |
| any_fwd_unpatentable | DEFECTIVE ARTIFACT | The proceeding rows and their recorded dispositions; the derived adverse-FWD label is field-gated out. | 2 | /companies | |
| ccad-reversal-composite-v2-auc | DEFECTIVE ARTIFACT | The evidence that the model failed, kept in full: n_test_rows 3,277 against a 5,000 floor, the 105.083% bridge match, the… | 2 | /api/ccad/reversal-composite/v2/bar | |
| companies-ptab-by-owner-cohort synthetic | DEFECTIVE ARTIFACT | The PTAB proceeding rows for the owner, which are fine to show; the route fails closed to NO_COHORT for the derived label… | 3 | /companies | |
| patent-litigation-outcome-v1-auc | DEFECTIVE ARTIFACT | The cohort sizes, which are disclosure and must render: n_test 1,298 and n_labeled 7,535. Neither reconstructs an AUC without… | 3 | /hub/ip | |
| ptab-inadmissible-predictive-scores | DEFECTIVE ARTIFACT | Observed PTAB outcomes only -- institution and final-written-decision counts with their denominators -- and the retraction… | 3 | /hub/ip | |
| anda-ob-rows | NOT YET MEASURED | The Orange Book rows we do hold, with their as-of date, and an explicit count of what is not covered. | 1 | /verticals/anda | |
| bankruptcy-v2-honest-macro-auc synthetic | NOT YET MEASURED | The placard '[AUC withheld: as-of horizon unestablished]' -- which is baked into 61 substitute PNGs so the withhold survives… | 3 | /bankruptcy | |
| insurance-pricing | NOT YET MEASURED | Inherits the dollar family; this is the render-attribute handle a byte guard can assert on. | 1 | /insurance/pricing | |
| insurance-pricing-dollar-family | NOT YET MEASURED | Inherits insurance-pricing-dollar-premiums. This is the GOVERNANCE spelling of the same claim, covering the severity bands,… | 3 | /insurance/pricing | |
| insurance-pricing-dollar-premiums | NOT YET MEASURED | The observed severity distribution as a range with its case count, and the statement that no indicative premium is produced.… | 4 | /insurance/pricing | |
| ipr_risk_tier | NOT YET MEASURED | The observed petition count and the observed disposition mix for the company; no tier, no band, no ordering. | 1 | /companies | |
| massarb-cost-per-case synthetic | NOT YET MEASURED | The aggregate exposure total; no per-case cost, since the denominator does not exist. | 2 | /exposure | |
| massarb-n-cases synthetic | NOT YET MEASURED | The aggregate figures the source does support, with an explicit statement that no per-case identifier was ever captured so no… | 2 | /exposure | |
| massarb-settle-pct synthetic | NOT YET MEASURED | The aggregate exposure figures only; no settlement percentage. | 2 | /exposure | |
| patents-counsel-etsi-declaration-count | NOT YET MEASURED | The counterparty and linkage counts on this patent's own asserter network, which stay VALUE figures regardless of the… | 1 | /patents/counsel/asserter-exposure | |
| portfolio-exposure-composite-auc synthetic | NOT YET MEASURED | The directional composite's inputs and their direction; the wall cell renders ABSENT because the artifact carries no AUC at all. | 2 | /hub/exposure | |
| precedent-negative-treatment synthetic | NOT YET MEASURED | The citation relationships themselves with their counts, and the statement that no automated negative-treatment signal is… | 3 | /precedent | |
| scotus-rf-v1-test-auc synthetic | NOT YET MEASURED | The walk-forward Brier and vote-direction accuracy SCOTUS is actually served on; no AUC arm exists, so the wall cell is ABSENT… | 2 | /api/predict-market/track-record/v1 | |
| screen-opp-rate-given-extension | NOT YET MEASURED | The raw counts behind the conditional rate -- applications published, applications opposed, and applications that drew an… | 2 | /screen | |
| state-appellate-verified-coverage | NOT YET MEASURED | The raw verified and unverified counts by state and court, with the states that have no coverage named rather than omitted. | 2 | /appellate/state | |
| any_instituted | RECONSTRUCTION RISK | The underlying PTAB proceeding rows themselves, from which a reader can see what happened in each; the derived boolean is not… | 2 | /companies | |
| appellate-ccad-case-reserve-band synthetic | RECONSTRUCTION RISK | The observed reversal base rate for the circuit with its count; no dollar reserve low/mid/high. | 2 | /appellate/ccad | |
| ccad-hold-gallery-withheld-evidence-fields synthetic | RECONSTRUCTION RISK | The gallery shot renders with its evidence, signature, token and values fields withheld; the shot itself and its route label… | 13 | /gallery/** | |
| ccad-pooled-segment-bands-auc | RECONSTRUCTION RISK | The per-segment bands themselves, which are what the drill table is for; the pooled reference line renders as the HOLD… | 1 | /score | |
| courts-duration-ranking synthetic | RECONSTRUCTION RISK | Venues listed in a neutral order (alphabetical or by case count) with the disposition mix; no speed ordering, no… | 3 | /circuits | |
| n_ipr_petitions | RECONSTRUCTION RISK | The petition count is itself disclosure and should render; what is gated is its use as a denominator for a withheld per-owner… | 1 | /companies | |
| ptab-slice-family-reconstruction | RECONSTRUCTION RISK | A coarsened slice family whose cells cannot be differenced back to the withheld per-party or aggregate rate; the disclosure… | 1 | /hub/ip | |
| ttab-x4-raw-score-tier-facet synthetic | RECONSTRUCTION RISK | The descriptive TTAB browse rows with no score-derived ordering; the tier control and its five-label taxonomy are removed, not… | 2 | /marks-v2-ttab | |
| venues-disposition-mix synthetic | RECONSTRUCTION RISK | The settled / dismissed / judgment mix survives as its own fields; only the duration legs and the duration-blended composite… | 3 | /circuits | |
| stale-producer-portfolio-patent-counts synthetic | STALE FROZEN BYTES | A 410 with the reason body, which states exactly what was wrong: per-company patent counts were multiplied by each company's… | 2 | /static/_snapshots/patents_pharma_case_explorer.html | |
| appellate-ccad-v2-internal-auc | SUPERSEDED | Nothing numeric on any external surface; in its place, the sentence that the serialized appellate_v2 bundle is a post-decision… | 0 | no live route | |
| bankruptcy-retired-macro-auc-lift synthetic | SUPERSEDED | The current lift figure against the current featureless baseline, with the baseline named; the retired lift renders only… | 3 | /bankruptcy | |
| ptab-pre-repair-institution-rate-snapshot | SUPERSEDED | The repaired institution rate from the live query layer with its corrected denominator, and a dated note that the earlier… | 1 | /static/_snapshots/** | |
| scotus-ot2025-decided-brier synthetic | SUPERSEDED | A unit-explicit receipt showing the walk-forward figure 0.1969 labelled per_justice_vote over a 35-term cohort, and the… | 3 | /scotus | |
| scotus-perjustice-tierc-brier | SUPERSEDED | The current per-justice figures; the retired GBT Brier 0.2009 / accuracy 0.698 / ECE 0.054 appear only inside a retirement… | 0 | no live route | |
| state-appellate-federal-v0-bar | SUPERSEDED | Renders today at 0.783 through the NAMED fail-open residual, not through a binding. The correct substitute is the same bar… | 1 | /appellate/state | |
| ttab-mature-base-rate-write-gate | SUPERSEDED | The right-censoring-corrected base rate from the live layer with its cohort window stated, replacing the frozen 67.4% / 0.674. | 1 | /static/_snapshots/** | |
| courts-duration-median-district synthetic | TARGET SUBSTITUTION | The non-duration legs of the venue card -- the settled / dismissed / judgment disposition mix -- which survive the duration… | 3 | /circuits | |
| patent-pooled-plaintiff-win-rate synthetic | TARGET SUBSTITUTION | The disposition mix as counts -- default judgments, consent judgments, and contested merits adjudications separately -- so the… | 3 | /static/_snapshots/patents.html | |
| ptab-aggregate-institution-rate | TARGET SUBSTITUTION | The per-slice institution counts with their denominators; the aggregate is replaced by the count of slices and the range… | 2 | /hub/ip | |
| ptab-corpus-kill-rate | TARGET SUBSTITUTION | The count of claims cancelled and the count of claims challenged, shown as two integers rather than their quotient. | 1 | /hub/ip | |
| ttab-oppose-card-r2-can-plaintiff-win | TARGET SUBSTITUTION | The CAN three-arm decomposition with its own n values (43,467 / 22,920 / 1,067) and per-cell contested share. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r2-can-plaintiff-win-ci-hi | TARGET SUBSTITUTION | None separately -- atomic with its point estimate. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r2-can-plaintiff-win-ci-lo | TARGET SUBSTITUTION | None separately -- atomic with its point estimate. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r2-opp-plaintiff-win | TARGET SUBSTITUTION | The three-arm decomposition: walkover, settlement-exit and adjudicated arms with their own n values (OPP: 123,757 / 83,657 /… | 1 | /brands/settle-posture | |
| ttab-oppose-card-r2-opp-plaintiff-win-ci-hi | TARGET SUBSTITUTION | None separately -- atomic with its point estimate. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r2-opp-plaintiff-win-ci-lo | TARGET SUBSTITUTION | None separately -- a rendered band is the same retired claim with error bars. Inherits the three-arm decomposition. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r3-can-defended-win | TARGET SUBSTITUTION | The CAN adjudicated arm (n=1,067) with its contested share. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r3-can-defended-win-ci-hi | TARGET SUBSTITUTION | None separately -- atomic with its point estimate. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r3-can-defended-win-ci-lo | TARGET SUBSTITUTION | None separately -- atomic with its point estimate. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r3-opp-defended-win | TARGET SUBSTITUTION | The adjudicated arm of the three-arm decomposition (n=3,332 for OPP), which is the population this rate was trying to… | 1 | /brands/settle-posture | |
| ttab-oppose-card-r3-opp-defended-win-ci-hi | TARGET SUBSTITUTION | None separately -- atomic with its point estimate. | 1 | /brands/settle-posture | |
| ttab-oppose-card-r3-opp-defended-win-ci-lo | TARGET SUBSTITUTION | None separately -- atomic with its point estimate. | 1 | /brands/settle-posture | |
| ttab-opposer-win-rate | TARGET SUBSTITUTION | Contested share and merits-decided count per party, both rendering today: on this league's decided proceedings 22.4% were… | 2 | /hub/ip | |
| ptab-anda-join-fwd-unpatentable-rate | UNRELIABLE STRATUM | The count of joined matters reaching a final written decision, and the count found unpatentable, as two integers. | 2 | /hub/ip | |
| ptab-anda-join-institution-rate | UNRELIABLE STRATUM | The number of ANDA matters that join to a PTAB proceeding at all, and the institution rate of the full PTAB population, kept… | 2 | /hub/ip | |
| ptab-named-party-institution-rates | UNRELIABLE STRATUM | Per-party petition COUNTS and the population institution rate beside them; no per-party rate at any n. | 1 | /hub/ip | |
| ttab-settle-propensity-can-plaintiff | UNRELIABLE STRATUM | The CAN population negotiated-exit rate (25.36%, a different and quotable claim) with its denominator 18,441 and its contested… | 1 | /brands/settle-posture | |
| ttab-settle-propensity-opp-plaintiff | UNRELIABLE STRATUM | Governed but currently QUOTABLE at 27.86%, and it must never render as a bare percentage: it is the SHARE of 779 profiled… | 1 | /brands/settle-posture | |
| ttab-settle-propensity-population-can | UNRELIABLE STRATUM | QUOTABLE at 25.36% with denominator 18,441, contested share 35.99%, and contested-arm rate 31.36%; never the pooled figure… | 1 | /brands/settle-posture | |
| ttab-settle-propensity-population-opp | UNRELIABLE STRATUM | Governed and QUOTABLE at 30.51%, but never alone: the contested share of its own denominator (38.65% of 57,190) and the… | 1 | /brands/settle-posture | |
| anda-wf-oof-auc-ci-hi | WITHDRAWN CLAIM | None separately -- atomic with anda-wf-oof-auc-v2-full. | 1 | /verticals/anda | |
| anda-wf-oof-auc-ci-lo | WITHDRAWN CLAIM | None separately -- this key is atomic with anda-wf-oof-auc-v2-full and inherits its substitute. A band without its point is… | 1 | /verticals/anda | |
| anda-wf-oof-auc-v2 | WITHDRAWN CLAIM | Inherits anda-wf-oof-auc-v2-full. This is the middleware's VALUE-keyed net over the same three numbers, existing because seven… | 1 | ALL | |
| anda-wf-oof-auc-v2-full | WITHDRAWN CLAIM | Everything descriptive the ANDA bundle still supports: Paragraph-IV filings, docket events, sponsor identity, Orange Book… | 4 | /hub/ip | |
| marks-v2-status-ablation-auc | WITHDRAWN CLAIM | The ablation's Brier, ECE, baseline-AUC and lift keys, which are NOT withdrawn and stay quotable; the withheld claim is 'how… | 2 | /marks-v2-ttab | |
| marks-v2-status-ablation-auc-high | WITHDRAWN CLAIM | None separately -- atomic with marks-v2-status-ablation-auc. | 2 | /marks-v2-ttab | |
| marks-v2-status-ablation-auc-low | WITHDRAWN CLAIM | None separately -- atomic with marks-v2-status-ablation-auc. | 2 | /marks-v2-ttab | |
| 83 refusals | 10 classes | 83 substitutes | 150 | 74 with a live surface |
The register prints a count of governed spellings and never a governed spelling. What you can read here is what class of thing is refused and what stands in its place — never the refused value. That constraint is why this page can exist at all.
Four gaps found while building this page. None is repaired here.
frozenset[str]. Its reasons live in comments. The grammar and the specimen do not share a single key.DARK rows span four ledger sections with no total, so the true share of built-and-unwired models is unstated.Sources, verbatim and pinned: litan/reports/N0813-CLA-UI-WITHHOLD-GRAMMAR-R1/withhold_grammar.json · litan/src/litan/served_truth/legacy_unbound_registry.py · litan/src/litan/served_truth/specimen_registry.py · litan/src/litan/ui/withheld_response.py, all read at git show HEAD: on design-v4 6f56541. Ledger figures measured in the canonical repo and cited by file and line in this folder's RECEIPT.md. The byte-level middleware declares 18 policy objects over rendered output — the outer net, after every template.