A recommendation is not an effect estimate, and was never meant to be. It is a composite of the evidence plus considerations that have nothing to do with whether the exposure works. This page is the diagnostic for separating them — and the central finding is that on the guidance the wiki holds, the separation is already published, so this is a reading skill rather than an accusation.
The structure — GRADE names four determinants, only one of which is evidence
Certainty is one input to strength, alongside the balance of desirable and undesirable consequences, values and preferences, and resource use. (Schünemann et al., n.d.) -> Certainty of Evidence vs Strength of Recommendation
So a recommendation can legitimately be strong on weak evidence, or weak on strong evidence, and neither is an error. The question is never did non-evidence considerations enter? — they are supposed to. It is which one moved this, and by how much?
The instrument — the Evidence-to-Decision table
WHO SFA 2023 publishes an EtD table with named domains, each carrying a judgement on a fixed scale
(No / Probably no / Probably yes / Yes / Varies / Don't know), the research evidence behind it, and
additional considerations. Two domains are pure non-evidence:
Acceptability — “Is the intervention acceptable to key stakeholders?” WHO’s listed influences include “level of awareness of the health problem”, “potential impact on national economies”, and “compatibility with existing policies”.
Feasibility — “Is the intervention feasible to implement?” Influenced by “the existing relevant infrastructure… and the available resources”, and by the fact that “widespread use and availability of certain food items high in SFA and/or TFA may pose challenges in decreasing consumption to meet the recommended intake.”
(World Health Organization, 2023)
A guidance body naming national economies among its considerations is the strongest available evidence that the multi-objective concern is real — and it is the body’s own disclosure.
The worked case, and it runs AGAINST the intuitive prediction
The obvious hypothesis is that feasibility and acceptability soften recommendations. On the one case the wiki holds, they did not.
| Domain | WHO’s judgement | Effect on the recommendation |
|---|---|---|
| Certainty (SFA -> CVD) | moderate | carried it |
| Acceptability | ”Varies” — unfavourable | none |
| Feasibility | ”Probably yes” | none |
| Resulting strength | STRONG | — |
WHO recorded that “there are many for whom the recommendation may not be acceptable, based on the current, popular perception that diets high in SFA do not pose a health risk” — and issued a strong recommendation regardless, justified on evidence alone: “Recommendation 1 was assessed as strong because evidence of moderate certainty overall from different study types assessing both risk factors and incidence of CVDs suggested reduced risk of CVDs with lower SFA intake.” (World Health Organization, 2023)
So the claim that multiple objectives make guidance less stringent is not entailed, and here is falsified. The direction is a per-case question.
Direction is indeterminate — the reason the general claim fails
Non-evidence objectives can push either way, and which way depends on the exposure:
- Toward laxity where an exposure is a dietary staple, cheap, or economically entrenched — the case the multi-objective worry has in mind.
- Toward stringency where an exposure carries environmental or ethical load. Environmental objectives would push red-meat guidance harder than health evidence alone warrants, not softer.
A claim that multi-objective reasoning weakens guidance is therefore a hypothesis about which master dominates for this exposure, and needs the EtD table to settle. It is not a general property of guideline-making. (inferred from Schünemann et al., n.d.; World Health Organization, 2023)
Status-quo anchoring — the best-evidenced non-evidence objective, and it is SYMMETRIC
Feasibility and acceptability do not push randomly. They push toward what is currently eaten — and that is a more predictable, more testable bias than “toward industry”.
NASEM states it almost directly, in the one commodity-adjacent passage of its whole review, on whether the dairy group is necessary or merely sufficient to meet the modelling goals:
“It seems the idea of a food group’s traditional role in the American diet may serve as an underlying constraint on the modeling, and a greater array of food group combinations and amounts may be possible that would meet energy and nutrient goals if that constraint were lifted.”
And the food supply is an explicit input to the writing step:
“Considerations are also made regarding how a proposed change might affect the food supply, because changes to better align with a proposed recommendation might affect the overall nutritional profile of a food product.”
WHO’s EtD feasibility row says the same from the other side: “widespread use and availability of certain food items high in SFA and/or TFA may pose challenges in decreasing consumption to meet the recommended intake.”
(National Academies of Sciences Engineering and Medicine, n.d.)
The purest observable case: a ceiling anchored to current consumption
NNR on white meat is status-quo anchoring with nothing else mixed in, and it is worth holding as the worked example.
- Health is neutral — “no currently available evidence for beneficial or detrimental effects of white meat consumption for cardiovascular diseases and type 2 diabetes.”
- The constraint is environmental — “To minimize environmental impact, consumption of white meat should not be increased from current levels, and may be lower.”
- Current levels are stated — “approximately 20 to 50 g/d”.
So a bound exists without a number being written. Do not increase from X plus a known X is a ceiling, and its location is set by consumption history rather than by evidence. NNR notes the intake “has increased the last decades and is the main driver of increased total meat intake” — so had consumption risen faster, the implied ceiling would sit higher.
This is the sharpest available test of the anchoring thesis because nothing else is doing work: no health signal, no quantified environmental magnitude, no stated threshold. What remains is the status quo, functioning as the recommendation’s location.
And it is not auditable at layer 3. The anchor is a range of national averages, so “current levels” has no individual referent — a person eating 80 g/d exceeds every country’s average, one eating 10 g/d is under all of them. A reader who accepts the constraint still cannot apply it.
(Nordic Council of Ministers, 2023)
Why the symmetry matters — it makes the thesis testable
Status-quo anchoring predicts conservatism in both directions, and both preserve the current diet:
| Direction | Predicted effect | Example |
|---|---|---|
| Novelty-aversion | guidance will not recommend an unfamiliar food even where the nutrition case is decent | insect protein; heavy fermented soy in a Western guideline |
| Removal-aversion | guidance will not recommend abstaining from an entrenched staple | bread, pasta, baked goods |
So the same mechanism produces two opposite-looking omissions — and a capture hypothesis explains only the second. This is the better-evidenced rival: it accounts for the same observations without requiring bad faith, and NASEM reaches for it (“traditional role”) precisely where an industry explanation would have been available. (inferred from National Academies of Sciences Engineering and Medicine, n.d.)
The internalization claim is DIFFERENT, and the held evidence runs against it
The further claim — that long-standing recommendations become internalized and acquire momentum as “ground truth” — is separable, and two held data points cut against it:
- Behavioural adherence is low. “less than 10 percent of Americans consume a diet fully consistent with the DGA.” Whatever has been internalized, it is not the eating.
- On the specific case, popular belief runs AGAINST the guidance. WHO records that many find its SFA recommendation unacceptable “based on the current, popular perception that diets high in SFA do not pose a health risk.” If the low-fat message had hardened into unquestioned ground truth, WHO would not be recording popular perception as an obstacle to it.
So the momentum claim needs a distinction it does not currently make: institutional momentum —
guidance bodies reluctant to contradict their predecessors, which NASEM does evidence in the form of
revision inertia — is a different thing from population internalization, which the held evidence
contradicts. AWAITS a source on public nutrition beliefs before the population limb is carried.
(World Health Organization, 2023)
A two-institution standpoint split — a body that produces the estimate and REFUSES the recommendation [2026-08-06, EFSA]
This page’s cases so far are bodies that make a recommendation and disclose (some of) what moved it. EFSA 2022 on dietary sugars is the opposite specimen — a clean instance of divergence reason #1 (different standpoint) split across two institutions: a body that runs the evidence appraisal and then declines to set the target by mandate.
«Establishing dietary goals or recommendations for dietary sugars (e.g. a limit of intake) and FBDGs on sugar-containing foods is part of national nutrition policies and thus in the remit of individual EU Member States, not under EFSA’s remit.» (European Food Safety Authority, 2022)
EFSA delivers a UL risk assessment (hazard characterisation) and states «a UL is not a recommended
level of intake» (European Food Safety Authority, 2022). Its output — no threshold identifiable, «as low as possible»
(European Food Safety Authority, 2022) — is deliberately number-free, and the number-setting (WHO’s <10%, national FBDGs) is
handed to a different actor operating under a different standpoint.
Why it sharpens the diagnostic: the whole difficulty this page addresses — separating the effect estimate from the non-evidence considerations that move a recommendation — is here done for you, institutionally. EFSA holds the estimate/hazard side; Member States hold the values, feasibility and communicability side. So WHO-vs-EFSA on sugars is a textbook divergence-class-1 (different standpoint, no error by either): not two bodies disagreeing on evidence, but a hazard assessment and a public-health target answering different questions -> Free Sugars Intake. The contrast with WHO SFA 2023 is instructive: WHO folds standpoint into one document and publishes an EtD table so the seam is visible; EFSA puts the seam between two institutions, so no EtD table is needed on its side — there is no recommendation to weight. Neither hides anything; they cut the estimate-from-recommendation join in different places. (inferred from European Food Safety Authority, 2022)
The negative case — two families differ, and NO non-evidence objective moved it (ACC-AHA 2026 vs ESC-EAS 2025) [2026-08-06]
This diagnostic’s failure mode is reading any guidance difference as objective-driven. The two current lipid-guidance families are the clean counter-example: where they converge, and where they differ, the account is entirely on the evidence side — no EtD non-evidence domain is doing work.
- They converge on the target. ACC-AHA 2026 restored numeric LDL-C goals (the 2018 US guideline had dropped them) and landed on ESC’s long-standing very-high-risk goal — LDL-C <55 mg/dL (1.4 mmol/L) + ≥50% reduction. Both rest on the same trial base (CTT per-mmol; FOURIER/ODYSSEY/CLEAR), so the agreement is shared-warrant, not two objectives happening to align -> LDL Lowering and Cardiovascular Events.
- Where they differ, the difference is INSTRUMENT, not objective. ACC-AHA stratifies with PREVENT-ASCVD (10-y ASCVD, ages 30-79); ESC with SCORE2/SCORE2-OP (fatal + non-fatal CVD, region- calibrated, ages 40-89). Different endpoint, calculator, calibration — divergence classes 1-2 (standpoint / evidence base), not a non-evidence domain moving the number -> SCORE2 Baseline Risk and the ESC Treatment Thresholds.
So the honest verdict is unprobed for objectives, explained by instruments — there is no published
EtD table to read here, and the visible differences are fully accounted for by calculator and endpoint
choice, so charging a non-evidence objective would be exactly the over-read the guard below forbids. The
one place an objective could enter — feasibility/cost of universal apoB or Lp(a) testing, or of PCSK9
agents at population scale — is not disclosed by either body, so it stays the corpus’s standing
finding: considerations undisclosed, weights certainly undisclosed.
(inferred from Blumenthal et al., 2026; Mach et al., 2025)
A second guidance family with a published EtD — and cost named in the balance (Demay 2024) [2026-09-02, Demay]
The Limits below awaited a second guidance family with a published EtD. Demay 2024 (Endocrine Society, vitamin D for disease prevention) supplies it — a GRADE guideline that ran EtD tables (GRADEpro, published online) and names its non-evidence domains in the document itself: «considered patient values, costs and resources required, acceptability and feasibility, and impact on health equity of the proposed recommendations» (Demay et al., 2024). So the corpus now holds two guidance families (WHO SFA 2023, Endocrine Society 2024) that disclose EtD considerations — the pattern is no longer one body.
Weights still absent — the standing observation holds at a second body, under a different framework. Demay names the domains and gives per-recommendation prose justifications, but publishes no combination rule or trade-off weighting, and for most questions «The panel did not identify robust evidence pertinent to these EtD factors for most clinical questions» (Demay et al., 2024). Considerations disclosed, weights not — a second instance, at a body using GRADEpro EtD rather than WHO’s in-document table (inferred from Demay et al., 2024).
The closest instance yet of a non-evidence domain entering the DIRECTION of a recommendation. For adults >=75 the mortality evidence is borderline (RR 0.96, CI upper bound 1.00), and the panel reaches its FOR by explicitly folding in cost — «given minimal costs of vitamin D supplementation, the panel reasoned that vitamin D is likely to be cost-effective with regard to its (likely) mortality benefit» (Demay et al., 2024), concluding «the panel judged that the balance between desirable and undesirable effects probably favors empiric vitamin D supplementation» (Demay et al., 2024). A non-evidence domain (low cost / minimal harm) is thus disclosed as part of the balance that tipped a borderline case toward FOR — visible, per this page’s method, in the panel’s own words -> Vitamin and Mineral Supplements for Disease Prevention.
But the over-read guard still binds — this tips DIRECTION, not STRENGTH, and the counterfactual is unavailable (inferred from Demay et al., 2024). Two limits keep it honest: (1) all 14 Demay recommendations stay conditional (GRADE “2”) — cost strengthened nothing to STRONG, so this is not the case the Limits bullet awaited (a non-evidence domain changing a recommendation’s strength); it changes the direction of a weak rec. (2) The panel frames the mortality point-estimate as the primary driver and cost as reinforcing, so whether the rec would have been AGAINST absent the cost reasoning is not stated — the ESC counterfactual caution applies. What Demay adds is a body reasoning from cost to a FOR direction in the open, on borderline evidence: the mechanism made visible, its magnitude still unmeasured.
Standpoint note (divergence class 1). Like EFSA on sugars, Demay marks its remit boundary — «The process to develop this clinical guideline did not use a risk assessment framework and was not designed to replace current DRI for vitamin D» (Demay et al., 2024) — a guideline answering the supplement-decision question, not the DRI-setting question EFSA/DRI bodies hold (inferred from Demay et al., 2024).
Tests / indicators
- Is there an EtD table at all? If yes, the analysis is a reading task. If no, this diagnostic cannot run -> see the guard below.
- Read the evidence domains and the non-evidence domains separately, then ask what changed between them. A recommendation whose strength exceeds what certainty supports has been moved by something; the table says by what.
- Check the direction before assuming it. An unfavourable acceptability judgement that did not soften the recommendation is evidence the body prioritised evidence — a finding worth recording, and the opposite of the usual suspicion.
- Ask whether the exposure is a staple. Where it is, feasibility has the most room to bite, and the EtD table is where that shows.
- Separate the recommendation from the estimate when transporting it. The effect estimate travels to an individual decision; the feasibility and acceptability judgements are population-scale and mostly do not. -> Framing a Decision Question
The guard — this diagnostic is easy to abuse
This wiki’s own telos names process defect the most easily abused of the five reasons guidance may differ, precisely because it licenses dismissing a conclusion without engaging its evidence. Applied loosely, they were just being practical discounts any inconvenient recommendation at zero evidential cost.
- Unstated motives are not evidenceable here. The published EtD table is the only legitimate instrument this diagnostic has.
- Where a body publishes no EtD table, the verdict is
unprobed, nevercompromised. Silence about feasibility is not evidence that feasibility drove the result. - A non-evidence consideration entering is not a defect. GRADE’s design puts it there deliberately. The defect would be a body concealing which consideration moved it — which is the opposite of what a published EtD table does.
- The process-defect (class-5) charge now has an admissible referent for the review layer. This page handles non-evidence objectives moving a recommendation; the separate question — was the systematic review underneath the recommendation trustworthy — is auditable against What a Trustworthy Systematic Review Requires (IOM 2011), the gold-tier institutional standard the telos’s class-5 rule demands (“documented by a source that meets the same appraisal bar”). Before it, the wiki could name a suspected process defect but had no external standard to cite; now a class-5 charge can point to a failed IOM standard number (unregistered protocol, single-reviewer screening, unreported funding) rather than resting on an opinion. Scope it to the failed standard, and check the deviation was not justified — both guards live on that page. (inferred from Institute of Medicine, 2011)
Limits
- Two bodies now, one full table. WHO SFA 2023 is the only EtD table the wiki holds in full; the
AWAITSfor a second guidance family with a published EtD is cashed by Demay 2024 (Endocrine Society — EtD domains named in-document, full GRADEpro tables online), which confirms the disclose-considerations / withhold-weights pattern at a second body under a different framework. What is still unheld is a full second EtD table to read domain-by-domain (Demay’s are online, not in the held PDF). See the Demay section above. - No non-evidence domain has yet changed a recommendation’s STRENGTH — Demay is the closest. Cost visibly enters Demay’s ≥75 balance judgement and tips the direction toward FOR on borderline evidence, but every Demay recommendation stays conditional (GRADE “2”) — so a domain changing the strength grade is still unheld. The mechanism (a non-evidence domain moving a recommendation) is now documented in the open; its magnitude remains unmeasured, and this page should not be read as saying non-evidence domains rarely matter.
- GRADE’s own EtD chapter is only lightly represented in the held handbook chunks; the fuller EtD framework literature is not held.
A second body blends health and environment — and concedes the health effect may be null [2026-07-28, ESC]
NNR is this page’s worked case for environmental reasoning entering a dietary recommendation. ESC does it too, in one sentence, and in a sharper configuration — a cardiology guideline, not a body with an environmental mandate.
«Meat. From both a health and an environmental point of view, a lower consumption of meat, especially processed meat, is recommended. A restriction of red meat may have little or no effect on major cardiometabolic outcomes. However, substituting red meat with high-quality plant foods (i.e. nuts, soy, and legumes) does improve LDL-C concentrations. A recent analysis showed that higher intake of processed meat and unprocessed red meat is associated with a 7% and 3%, respectively, increased risk of ASCVD.» (European Society of Cardiology, 2021)
Quoted whole, because the four sentences pull in different directions and any subset misrepresents it. ESC states, in order: a joint health-and-environment recommendation; a concession that restriction «may have little or no effect»; a substitution benefit on a surrogate (LDL-C); and an associational 3% / 7% risk increase.
Parameter table — the two health claims are not the same quantity:
| Claim | Design | Exposure contrast | Outcome | Same quantity? |
|---|---|---|---|---|
| «little or no effect» | restriction studies (ref 416) | reducing red meat | «major cardiometabolic outcomes» | — |
| «7% and 3%… increased risk» | association analysis (ref 417) | higher intake, observational | ASCVD | NO — trial-style restriction vs observed association, different outcome definitions |
So this is not ESC contradicting itself, and filing it as an internal tension would be the error. It is a body reporting that the interventional evidence is weak and the observational evidence is positive-but-small, then recommending reduction anyway.
Where the diagnostic bites: the recommendation is explicitly labelled «From both a health and an environmental point of view» — so the environmental objective is disclosed, and the health limb is disclosed as weak, in adjacent sentences. Per this page’s classification that is disclosure of considerations without disclosure of weight: the reader cannot tell how much of «lower consumption is recommended» rests on the 3% association and how much on the environmental argument, and ESC never says.
Two things this strengthens, and one it does not.
- It generalises the NNR finding beyond an environmentally-mandated body. NNR was instructed to integrate sustainability; ESC was not, and did it anyway in a cardiology guideline. That the behaviour appears without the mandate is the more interesting datum.
- It is a second instance of the corpus’s standing observation that considerations get disclosed and weights do not.
- It does NOT establish that the environmental objective changed the recommendation. ESC also has a positive (if small) associational health signal and an LDL-C substitution benefit, either of which could carry «lower consumption» on its own. The counterfactual is unavailable, and claiming environment moved this one would be exactly the over-read this page exists to prevent. (inferred from European Society of Cardiology, 2021)
Divergence class 4 (lag), self-documented [2026-07-28, ESC chunk 05]
The telos lists five reasons guidance can differ from a recommendation here, and lag is the hardest to evidence — it usually has to be inferred from publication dates. ESC states it about itself:
«Evidence now suggests that the BP targets in the previous iteration of this guideline were too conservative, especially for older patients.» (European Society of Cardiology, 2021)
A guidance body recording that its own prior edition was wrong, in the direction of under-treatment, with the revision reasoned. ESC gives the grounds (targets «are safely achieved in many older patients and are associated with significant reductions in the risk of major stroke, HF, and CV death») and bounds the new position itself: «the evidence sup- porting more strict targets is less strong for very old people (>80 years) and those who are frail».
Three things this is useful for, and one it is not.
- It is direct evidence that the guidance set moves, so “current guidance says X” has a shelf life. The guidance null is a moving benchmark, not a fixed one.
- The direction is the interesting part. The revision made guidance more aggressive. A fabric that expected guidance to be systematically over-cautious or systematically over-reaching would mis-predict here; this is one instance against both.
- It is a body disclosing its own error rather than a critic alleging one, which is the highest evidential standing a class-4 claim can have — no attribution dispute is available.
- It is NOT a process defect (class 5). ESC followed its method and updated on new evidence. Reading self-correction as a defect would penalise exactly the behaviour a sound process produces, and this page should say so plainly.
A body that draws the single-axis line explicitly — and puts the seam in a decision-relevant place [2026-07-31, USPSTF]
The telos’s one axis, named — never blended rule asks a guideline body to keep health separate from cost/economics. USPSTF states exactly this cut, in one sentence:
«The USPSTF does not consider the financial costs of providing a service in its assessment of the balance of benefits and harms, but may provide contextual information regarding costs for use by providers, including cost-effectiveness studies.» (US Preventive Services Task Force, 2022)
But the seam is not where “no economics” would put it — and that is the interesting part. USPSTF excludes financial cost yet includes opportunity cost inside the harm axis: «Harms may also be considered in the form of opportunity costs for both patients and providers. The Task Force may consider the time and effort required by both patients and the health care system». So the line is not health-vs-everything-non-health; it is dollars out, patient/provider time-and-effort in — the latter reclassified as a harm rather than a cost. (US Preventive Services Task Force, 2022)
Why this sharpens the diagnostic rather than just corroborating it. This page’s finding is that bodies disclose considerations but not weights. USPSTF does something cleaner: it excludes a whole axis by rule and says which one, so there is no weighting question to hide for financial cost — cost-effectiveness is provided as contextual information, outside the graded balance. This is the telos’s discipline enacted, not evaded: name the axis and stop. The one judgement it does fold in (opportunity cost) it re-labels as a harm and grades on the same net-benefit scale, which keeps it inside the health/burden axis rather than importing a second maximand. -> Net Benefit and the USPSTF Recommendation Grid (inferred from US Preventive Services Task Force, 2022)
Self-critique [run 2026-09-02, before commit — Demay 2024 woven]
- No [PRIOR] verdict scored. The Demay section appraises the weighting observation (considerations
disclosed, weights not) as a reading of the guideline; it does not score the weighting [PRIOR]
in
telos-review.md(ingest builds fabric, does not adjudicate) — the scoring is owed to a separate pass. - Over-read guard respected. The cost-tips-direction claim is explicitly bounded: it changes the direction of a conditional rec, not its strength; the counterfactual (would ≥75 have been AGAINST absent cost?) is named as unavailable, matching the ESC-case discipline this page already holds. No claim that cost proved decisive.
- Not laundered independence. Demay is a second disclosing body, but it is not claimed to
independently corroborate WHO SFA — it is a same-pattern second instance (both publish EtD
considerations, neither publishes weights), which is what the standing observation predicts, not an
[E-independent]lift. - Coherence, not validity (R1): the loop is open; this documents what the guideline discloses, not whether its ≥75 FOR is correct.