Channel A vs Channel B (design slant vs fabrication) — pilot landed; documented levers under-explain the +7 pp

🟡 A 244-pair curated-pair LLM-extracted pilot plus universe-scale tests of specific levers establish that no individual registered design choice approaches the +7 pp headline magnitude; the residual two to six pp lives in dimensions the TSE registration text does not surface. The substantive answer is documented in paper/paper.tex §5 "Mechanism: declared design and the unobserved layer", Table~5 (design-inventory of ten within-pair structural tests). The universe-scale poll_methodology LLM extractor would tighten SEs on the remaining quantified contrasts but is no longer load-bearing for the paper's conclusion.

Draws on:

The 244-pair curated-pair pilot (poll-methodology-paired): three candidate sample-shaping levers surface (population reference frame, geographic coverage class, IBGE census-setor cluster), but the single quantified interaction (population × sponsored, +4.6 pp, SE 3.7, p = 0.21) is underpowered and does not survive BH multiple-testing correction in Table 5.
The within-pair structural-test inventory at pilot scale (AN-019/020/021/022/041/042/043/051/052/053): every other directional Channel A prediction returns reversed (sponsors describe interviewer training more, use phone less, defer coverage less) or null (audit floor, methodology completeness).
Universe-scale lever tests (already done): AN-024 ruled out coverage-deferral asymmetry (n=14,876, OR=0.89, sponsors defer less); AN-033 ruled out deferral × bias interaction; AN-107 re-ran the coverage × sponsor interaction at universe scale on 14,876 polls.
The days-to-election-decay finding (consistent with verifiable-disclosure pressure, but doesn't discriminate cleanly between Channel A and Channel B).

Size-mismatch is the substantive finding. The generous upper bound on documented Channel A levers (per docs/thinking.md § "Residual decomposition of the +7 pp") puts scenario rotation at 1–2 pp (and walked back to firm-tier composition by AN-052), population-reference mismatch at 0–1 pp, and ponderação at 0–2 pp — together at most 1–5 pp out of the +7 pp. Two to six pp of the headline is unattributed. The paper reports this directly in §5 ("no individual registered design choice, summed or alone, approaches the +7 pp headline magnitude") and concludes by elimination: the carrier mechanism is not captured by what TSE pre-registration makes public.

Selective disclosure is the secondary loud pattern. Sponsored polls under-document sample-shaping dimensions (coverage, audit %, methodology completeness) and over-document visible-rigor dimensions (interviewer training, supervisor role, post- stratification language). The bias contrast does not co-vary with which side documents, so opacity is the absence of information rather than a mechanism by itself (AN-042 within-pair contrast on the differing-training subset: MW p = 0.58).

Status: 🟡 because the substantive question — whether the unattributed 2–6 pp lives in Channel A's wide upper bounds, in sophisticated Channel B variants the digit forensics don't reach, or in design dimensions registration doesn't surface — remains genuinely open at the point-estimate level. The paper's published reading is that Channel A is one candidate, not the leading one, and that the residual is by-elimination from registered dimensions. A universe-scale poll_methodology LLM extractor (queued in pipelines/politica/docs/todo.md) would tighten the small-same- direction levers (population frame, ponderação) but the size- mismatch reading is robust to plausible refinements at universe scale: even with tight SEs on every quantified contrast in Table~5, no row in the inventory crosses +3 pp.

Sources.

Own analysis: poll-methodology-paired finding; AN-024; AN-033; AN-041; AN-042; AN-043; AN-051; AN-052; AN-053; AN-107
Reports: none direct
News anchors: none direct
Cross-refs: poll-methodology-paired (the pilot); days-to-election-decay; paper/paper.tex §5 "Mechanism: declared design and the unobserved layer" (the published version of this finding); docs/source-of-bias.md (canonical living synthesis); H3 channel-a-shrinkage; H4 methodology-flexibility-a; H5 verifiability-decay-b; theory.md § Polls as Bayesian persuasion

Cited analyses

AN-019 yellow descriptive

On the 200-poll methodology subset, slant-permissive coverage classes (specific_neighborhoods + urban_only) appear in 12% of candidate-touched polls vs 10% of independent polls — direction matches Channel A but n_candidate=25 makes the gap noisy. Opaque coverage (deferred + not_specified) is 72% vs 80%, weakly contradicting the "candidates hide scope" prediction.

AN-024 green descriptive

Universe-scale n=14,876. Candidate-touched polls defer at 35.5% vs independent at 38.3% — candidate-touched are LESS likely to defer (odds ratio 0.89, 95% CI [0.80, 0.98], chi-square p = 0.021). The "candidates hide coverage via deferral" hypothesis is statistically refuted at universe scale.

AN-033 green causal

Within-candidate FE on the AN-001 sample (n=27,907 candidate-poll rows from 8,917 protocols): the sponsored_by × deferred interaction γ is null and bouncing-signed across specs — +1.08 (SE 2.08, p=0.60) under candidate FE; +2.05 (SE 2.18, p=0.35) with institute FE + controls; −5.63 (SE 5.46, p=0.30) under race × week FE. Combined with AN-024 (sponsors do NOT disproportionately defer), deferral is ruled out as a load-bearing Channel-A lever in both directions.

AN-107 pending causal

AN-052 green robustness

AN-051's −20.8 pp sponsored-vs-independent rotation gap vanishes within firm — LPM with firm FE gives sponsored coefficient +0.025 pp (cluster-SE 0.027, p = 0.37). 19 of 20 firms with both sponsored and independent pair-sides show identical rotation rates between sides. The rotation contrast is a between-firm composition effect, not a within-firm sponsor-choice effect.

AN-042 green descriptive

Sponsored polls describe interviewer training MORE than matched independent polls (84.4 % vs 72.5 %, McNemar p = 0.002) and supervisor role slightly more (92.6 % vs 87.3 %, p = 0.08). Bias contrast does not track which side describes (MW p ≈ 0.57 for both fields). The opacity gap is field-specific — sponsored polls under-document coverage but over-document interviewer-side rigor.

AN-041 green descriptive

Sponsored polls never use phone (0/244) and use in-person at 95%; independent uses phone at 10%. χ² on the joint mode table p = 0.0003 in the *opposite* direction of cheap-mode-slant. The mode-substitution lever is refuted on this sample.

AN-043 yellow descriptive

Structured `nonresponse_handling` is 100 % `not_specified` on every pair. Diagnostic regex grep confirms the substantive null — undecided/refusal vocabulary appears in only 5 sponsored and 2 independent pair-sides out of 488. The probe is null-by-data-design: registration PDFs are pre-fielding planning documents and do not describe a post-fielding analytical choice. Testing the nonresponse-handling × sponsor lever requires the post-fielding *relatório* PDFs (currently outside the methodology extraction pipeline).

AN-051 green descriptive

Sponsored polls dramatically under-document candidate-name rotation in the questionnaire (5.4 % vs 26.1 % when rotation+random+rotated combined; McNemar p < 10⁻⁷). Sponsored polls file FEWER vote-intention scenarios on average (3.73 vs 4.49, Wilcoxon p = 0.002) — refutes the within-poll scenario-selection / cherry-pick hypothesis. **[2026-06-02 update — see AN-052/AN-053:]** The rotation contrast is a between-firm composition effect, not a within-firm sponsor-choice effect (AN-052 firm-FE LPM p = 0.37). The direct candidate-position priming test refutes the sponsor-exploitation reading (AN-053: sponsored polls list sponsor's candidate LATER, McNemar p = 0.001 in WRONG direction). The AN-051 finding is recontextualized as a firm-tier composition signal — sponsors choose low-discipline firms that don't document rotation — rather than a within-firm Channel-A lever.

AN-053 green descriptive

The direct name-order priming test refutes the sponsor-exploitation reading. On 196 of 244 pairs (after fuzzy name-matching), the sponsor's own candidate appears LATER on the sponsored side than on the matched independent side — sp first-position rate 20.4 % vs ind 30.1 % (McNemar p = 0.001 in the WRONG direction). Sponsored polls do not strategically order their candidate to exploit first-position priming. The within-pair bias-carrier test IS significant (MW p = 0.019) but on the small 17-pair subset where sp lists the candidate earlier — when priming is used, bias is higher; sponsors generally do not use it.