Nucleus of the cardiometabolic-exposures cluster’s meal-timing facet. The decision question:
does when you eat — the clock position of the eating window — change weight or cardiometabolic health
independent of what and how much you eat? Two RCTs answer it with opposite designs, and read
together they decompose the popular claim rather than settling it as a single number.
The bottom line, up front
- The weight-loss rationale for TRE is weak — and this is now RCT-evidenced, not inferred. The schedule-isolating test is Liu 2022 (NEJM, n=139, 12 months): both arms were prescribed the same calorie deficit and only the 8-hour window differed, so it isolates the window from intake as far as self-reported intake can verify (both arms had the same prescribed deficit; total energy expenditure was not measured, so a small window-induced intake difference the logs could not see remains possible — corrected 2026-08-08). Adding the window bought no significant extra weight loss («net difference, −1.8 kg; 95% CI, −4.0 to 0.4; P = 0.11»), and every metabolic secondary was null between groups. The earlier free-living RCT (TREAT, n=116) pointed the same way against ad-libitum eating; Liu is the cleaner isocaloric head-to-head. The thesis timing acts on weight mainly via energy intake, the window adds little beyond the deficit is thus evidenced by a 12-month high-tier RCT, no longer transferred by analogy. (Liu et al., 2022)
- The muscle-loss harm may not be intrinsic to the window — protein is the leading candidate, not a proven cause (corrected 2026-08-08). TREAT’s disproportionate lean-mass loss did not reproduce in Liu, which supplied a daily protein shake for 6 months and found no between-group lean-mass difference. Neither trial randomized protein, so the mechanism is a hypothesis (Lowe’s own «it is possible that protein intake was altered by TRE … this clearly warrants future study»), not an established finding (refinement below).
- But there is a real, weight-INDEPENDENT metabolic signal from eating EARLY. In a tightly controlled, weight-matched crossover (eTRF, n=8), shifting the same food earlier in the day improved insulin sensitivity, blood pressure and oxidative stress with no weight change at all.
- These do not conflict. They test different exposures: restrict the window whenever (and eat less if you happen to) vs shift eating earlier, calories held constant. The naive TREAT says TRE does nothing / eTRF says it does everything tension dissolves once the exposures are separated (parameter table below). The lever, where it exists, is circadian timing, not the window per se — and it is small, short-tested, and on surrogates only.
The parameter table — why TREAT and eTRF are NOT the same quantity (op-weave 2a)
| Parameter | TREAT (Lowe 2020) | eTRF (Sutton 2018) | Same quantity? |
|---|---|---|---|
| Exposure | 16:8 window, late (eat 12-8pm, skip breakfast) | 6-h window, early (dinner before 3pm) | NO — late vs early clock position |
| Calories held constant? | NO — ad libitum, deficit permitted | YES — «fed enough food to maintain their weight», meal-by-meal matched | NO — the decisive difference |
| Setting | free-living, remote (app + scale) | supervised controlled feeding | NO — effectiveness vs efficacy |
| Weight changed? | small loss, no advantage vs control («−0.26 kg … P = .63») | held stable by design («D = 0.5 ± 0.3 kg; p = 0.12») | NO — one permits weight change, one forbids it |
| Population | n=116, BMI 27-43, both sexes, non-diabetic | n=8, prediabetic men, BMI ~32 | NO — different size + risk + sex |
| Primary endpoint | weight (null between groups) | glucose tolerance + postprandial insulin + insulin sensitivity (3-h OGTT primaries); BP/oxidative-stress secondary (corrected 2026-08-08) | NO — different primary |
| Design | RCT, parallel, 12 wk | RCT, crossover, 5 wk/arm | partly |
| Insulin / glycaemia | fasting insulin/glucose/HOMA-IR/HbA1c all null | mean insulin −26 mU/L; insulin resistance −36 U/mg (p=.005); glucose null | NO — free-living null vs controlled improvement |
| Blood pressure | between-group null | SBP/DBP −11 / −10 mm Hg (p=.03) | NO — opposite verdicts, but different exposures |
The fourth column is almost entirely NO — and that is the finding. These trials do not disagree about a shared quantity; they measure two different interventions. So there is no tension to file — the honest artifact is a decomposition (type-A emergent synthesis): the popular umbrella term TRE hides at least two distinct exposures that behave differently.
(Lowe et al., 2020) (Sutton et al., 2018)
The decomposition — three separable claims
- Restrict your eating window whenever, change nothing else → little independent effect on weight. TREAT is the clean test: adherence was high, yet the window alone bought no weight advantage and no metabolic movement. This is the same shape as the wiki’s macronutrient findings — energy is the headline; the clever protocol is the small print (Free Sugars Intake). TRE works in uncontrolled reports mostly because a shorter window cuts intake and helps adherence, not because the timing itself does something. (Lowe et al., 2020) The not-because-timing attribution is the wiki’s inference, not Lowe’s data — indeed Lowe found «no differences in estimated energy intake between groups» in its own trial. (corrected 2026-08-08)
- Shift eating EARLY, calories constant → a real but small weight-independent metabolic signal. eTRF is the only trial here that isolates timing from energy, and it found improvements in insulin sensitivity, β-cell responsiveness, BP and oxidative stress with zero weight change. If real, this is a genuine circadian effect — metabolism «is optimized for food intake in the morning», so eating aligned to it helps. (Sutton et al., 2018) The mirror is that late eating tends to be worse (Sutton cites prior late-TRF trials that were null or harmful) — so direction-of-shift, not window-length, carries the signal. But the eTRF ledger is not all wins (added 2026-08-08). eTRF also raised morning fasting lipids: «eTRF did increase morning fasting levels of triglycerides by 57 ± 13 mg/dL (p = 0.0007), which translated into a 13 ± 5 mg/dL relative increase in morning fasting levels of total cholesterol (p = 0.02)». Sutton attributes the triglyceride rise to the unmatched 18-h fasting duration and flags that future trials are needed to confirm the phenomenon is not pathophysiologic — so it is plausibly an artifact, but the harm-direction finding belongs on the ledger, not just the benefits. Symmetric caveat on the headline BP win: Sutton warns that «measuring only morning fasting values may overestimate eTRF’s effects on blood pressure». (Sutton et al., 2018)
- The window can add a specific HARM: lean-mass loss. TREAT’s decision-relevant find is that TRE’s weight loss was «approximately 65% … lean mass» vs a normal 20-30%, with a significant between-group appendicular-lean-mass deficit — plausibly because a short ad-libitum window cuts protein intake and protein is eaten mostly at meals. A late/short window that quietly under-feeds protein trades fat-loss you wanted for muscle you didn’t -> Protein and Resistance Training for Muscle and Strength. (Lowe et al., 2020)
Liu 2022 — the RCT that isolates the schedule from the deficit
The load-bearing gap the page previously named (a calorie-matched TRE-vs-continuous-restriction test — does the null generalize?) is now held. Liu et al. (NEJM 2022) randomized n=139 obese, non-diabetic adults (BMI 28-45, Guangzhou) to an 8-hour window (8:00am-4:00pm) plus calorie restriction vs the same daily calorie restriction without a window, for 12 months. Both arms got an identical prescribed deficit (men 1500-1800, women 1200-1500 kcal/day), high and equal adherence («84.0±16.1% … 83.8±12.6%»), 85% completion. So the intended difference between arms is the clock — the quantity TREAT could not isolate — with the caveat that intake equality rests on prescribed targets plus self-reported logs/photos, not measured energy expenditure (Liu notes «total energy expenditure was not assessed in this trial»). (Liu et al., 2022)
- Weight (primary): null between groups. −8.0 kg (TRE) vs −6.3 kg (CR); «Changes in weight were not significantly different in the two groups at the 12-month assessment (net difference, −1.8 kg; 95% CI, −4.0 to 0.4; P = 0.11)». Both arms lost real weight — the deficit worked; the window added nothing detectable. (Liu et al., 2022)
- Every secondary null between groups too: body fat (−5.9 vs −4.5 kg), visceral fat, waist, BMI, lean mass (−1.7 vs −1.4 kg), systolic/diastolic BP, fasting and 2-h glucose, HOMA-IR, insulin disposition index, and lipids — no significant between-arm difference on any. Adverse events similar; no deaths. (Liu et al., 2022)
- Authors’ own verdict: «caloric intake restriction explained most of the beneficial effects seen with the» time-restricted-eating regimen, which «did not produce greater weight loss than the regimen of daily calorie restriction, with both regimens» reaching similar caloric deficits. (Liu et al., 2022)
Why Liu is a cleaner quantity than TREAT — matched parameters (op-weave 2a)
(inferred from Liu et al., 2022; Lowe et al., 2020) — cells are extracted per source; the same quantity? judgment is the wiki’s.
| Parameter | TREAT (Lowe 2020) | Liu 2022 | Same quantity? |
|---|---|---|---|
| Comparator arm | window vs ad libitum (no deficit prescribed to control) | window vs isocaloric continuous deficit (same target both arms) | NO — Liu matches the deficit out; TREAT does not |
| What is isolated | the window + whatever intake change it causes | the window alone (schedule) | NO — Liu isolates the clock, TREAT confounds it with intake |
| Window clock position | late (12-8pm) | early/midday (8am-4pm) | NO — different circadian placement |
| Endpoint | weight, 12 wk | weight, 12 mo | partly — same measure, longer horizon |
| Between-arm weight verdict | null vs ad libitum | null vs isocaloric CR | consistent — both null, but against different comparators |
The two nulls agree in direction and Liu is the stronger design for the schedule question — but they are not the same comparison: TREAT asks does a window beat eating freely?, Liu asks does a window beat the identical deficit delivered without one? Liu answering no is the decisive one for is the clock itself a lever — it removes the intake difference TREAT leaves in. type-F (Liu bounds TREAT’s reading with the cleaner design).
The lean-mass harm does NOT reproduce — protein is the leading candidate explanation (type-F)
Liu explicitly joins TREAT’s muscle-loss finding: it cites Lowe and reports the opposite result, and its Discussion points to protein/diet balance. This is the source’s own contrast, not a manufactured one — but neither trial randomized protein, so protein is the leading candidate explanation for the non-replication, not a proven cause (corrected 2026-08-08). Liu gave the shake to both arms «to help improve adherence to the permitted calorie intake», so it cannot isolate protein either, and the Liu-vs-TREAT contrast also differs in window clock position, macro-balanced counselling, population, and duration.
| Parameter | TREAT (Lowe 2020) | Liu 2022 | Same quantity? |
|---|---|---|---|
| Lean-mass result | ~65% of weight lost was lean; significant appendicular-lean deficit | «no significant differences between the groups in gains in body lean mass» (−1.7 vs −1.4 kg) | NO — within-arm loss share vs between-arm difference |
| Protein provision | none; ad-libitum short window | daily protein shake, first 6 months + balanced-macro prescription | NO — the decisive difference |
| Liu’s own reading | (contrast cited) | «the importance of a balanced diet and adequate protein consumption in patients who are adhering to a diet regimen of time-restricted eating» | — |
Refinement (corrected 2026-08-08): the lean-mass loss is plausibly a downstream consequence of a window that under-feeds protein rather than an intrinsic effect of restricting the eating window — but this is a candidate explanation for a non-replication across two very different trials, not a tested causal result. The decision-relevant instruction the page gives (eat earlier, keep protein up) is supported by the direction of both trials; it should not be sold as guard protein and the loss cannot appear on a certainty neither source claims.
The pooled IF picture (Semnani-Azad 2025 NMA) — confirms Liu, disambiguates the regimens
The IF network-MA the page awaited is now held: 99 RCTs, 6582 adults, CINeMA/GRADE, EASD-commissioned (gold-tier, surrogate-only). It does three things for this page. (Semnani-Azad et al., 2025)
- (1) It confirms Liu’s single-RCT null at the pooled level (type-F, NOT type-E — shared trials). Across the whole TRE evidence base, TRE adds essentially nothing to a continuous deficit: the TRE-vs-CER difference is a trivial 0.39 kg (the CER:TRE cell, moderate certainty; the tiny edge nominally to CER) — as null as Liu’s −1.8 kg. And vs ad-libitum, «TRE showed a trivial reduction (−1.72 kg (−2.21 to −1.22), moderate certainty of evidence)» — below the 2.0 kg MID. (Semnani-Azad et al., 2025)
- (2) It disambiguates the IF umbrella (type-B). “Intermittent fasting” is three distinct regimens with different effects: TRE (daily window), ADF (24 h fast alternate days), WDF (5:2). Only ADF beats CER: «alternate day fasting was the only form of intermittent fasting diet strategy to show benefit in body weight reduction (mean difference −1.29 kg (95% CI −1.99 to −0.59), moderate certainty of evidence)» — but −1.29 kg is below the 2.0 kg MID, so «particularly ADF versus CER (−1.29 kg) fell below this level, suggesting somewhat limited clinical impact». So even the best IF form buys a clinically trivial edge over plain calorie restriction. (Semnani-Azad et al., 2025)
- (3) It bounds the claim by duration — even the trivial ADF edge vanishes long-term. «Estimates were similar among trials with less than 24 weeks follow-up (n=76); however, moderate-to-long-term trials (≥24 weeks, n=17) only showed benefits in weight reduction in diet strategies compared with ad-libitum» — «with no differences between intermittent fasting strategies and CER in these moderate-to-long term studies». The source’s FIRST-listed explanation for this loss of association is underpowering, not settled equivalence (restored 2026-08-08): «The loss of association in the network assessment of moderate to longer term trials (≥24 weeks) may be due to an insufficient number of studies available» (the ≥24-wk stratum is only 17 trials); it then adds adherence decline (one WDF trial: 74% at 6 wk -> 22% at 52 wk) and metabolic adaptation. So the long-horizon question is underpowered-and-open on the data — though the authors also state in their own voice that «both intermittent fasting and CER appear to provide similar moderate- to-long term improvements over ad-libitum diets» and call this a «critical take-away for clinicians managing chronic metabolic conditions», which is the warrant for reading IF == CER at the horizons that matter. (Semnani-Azad et al., 2025)
Why this is shared-evidence (F), not independent backing (E) — the trial-overlap check
(inferred from Semnani-Azad et al., 2025) The convergence with the held pages is not an independent second route to the same claim — the NMA pools the very trials this page already cites, so a RAG over Semnani-Azad alone reproduces the held finding. Verified against its reference list:
| Held trial | In Semnani-Azad’s pool? | Reference |
|---|---|---|
| Lowe TREAT 2020 | yes — pooled | ref 81 (included-studies range 32-128) |
| Liu 2022 (NEJM) | yes — pooled | ref 79 (included) |
| Sutton eTRF 2018 | no — Discussion citation only | ref 134 (outside 32-128) |
So on the weight question the NMA is type-F (it bounds/quantifies the held single-RCT nulls into a
pooled magnitude the wiki could not compute itself — type-A aggregation), never [E-independent]. On the
early-timing question it brings no new evidence at all: Sutton is not in the pool, and Semnani-Azad
cites the same external proof-of-concept the page already holds. No independence to claim in either place.
Parameter table — the pooled TRE-vs-CER null vs Liu’s single-RCT null (op-weave 2a)
| Parameter | Liu 2022 (single RCT) | Semnani-Azad 2025 (NMA) | Same quantity? |
|---|---|---|---|
| Comparison | TRE+deficit vs isocaloric CER | TRE vs CER (network estimate) | yes — window vs continuous deficit, weight |
| Estimate | net −1.8 kg (95% CI −4.0 to 0.4, NS) | 0.39 kg (CER:TRE cell, moderate, NS) | consistent — both trivial/null (nominal signs differ, both non-meaningful) |
| Evidence | 1 RCT, n=139, 12 mo | pooled TRE arms across 99 RCTs | NO — single trial vs pooled (Liu is inside the pool) |
| Horizon | 12 mo | median 12 wk; ≥24 wk stratum also null | partly |
The two nulls agree and are the same conceptual quantity — but they are not independent (Liu is one of the pooled trials). type-F: the NMA raises the precision/generality of Liu’s reading, not a second witness to it.
Why this is not a filed tension (the not-joined check)
Ran the counter-passage check on both sources’ own design/limits statements end to end. The apparent clash fails the joined-issue test on (ii) different scope/intervention: TREAT permits a calorie deficit and shifts eating late; eTRF holds calories fixed and shifts eating early. They are consistent once matched — indeed TREAT’s authors cite Sutton approvingly and note their own late window «might not be optimal for the metabolic advantages of TRE». Same-question-clashing-backing is absent; this is one exposure term splitting into two. Recorded as a decomposition (distinction), not a tension.
Meal frequency — the same isocaloric logic, and where it stops being held [2026-08-22]
The grazing belief — small, frequent meals raise your metabolism — is the meal-count version of the window question, and the held principle answers the weight half of it directly: meal pattern is not a weight-independent lever once total intake is fixed. Liu’s 12-month isocaloric null and TREAT both put the effect in the deficit, not the schedule (above), so eating more often will not, by itself, raise weight loss beyond whatever it does to total intake.
Where the held fabric stops (the gap). Meal frequency is a different exposure from the eating window, and the trials the wiki holds fixed it as a matched control (3 meals/day) rather than testing it as an exposure. So the narrower believer’s claim — that more frequent meals raise energy expenditure via a more frequent thermic effect of food — is insufficient-evidence here, neither confirmed nor refuted: the fabric holds no frequency-specific EE/TEF or hard-outcome SR. The decision answer is held (meal count is not a weight lever independent of intake); the metabolic-rate mechanism is a named gap (G), worth queuing only if a believer-facing deliverable needs the number.
(inferred from Liu et al., 2022; Lowe et al., 2020)
Decision relevance
- Do not sell TRE as a weight-loss method on its own. For someone whose goal is weight, the honest message is: a shorter/later window helps only if it makes you eat less and you can stick to it — it is a delivery system for a calorie deficit, and adherence, not timing, is what it buys (Does Weight Loss Reduce Cardiovascular Events: weight loss itself is not even a proven hard-CV lever via lifestyle).
- If you do a window, putting it EARLY is the better-supported bet — but the support is weak and contested (corrected 2026-08-08). The only weight-independent benefit signal is from early eating, and it rests on a single n=8 crossover; the page’s own 12-month early-window RCT (Liu, matched calories) showed no between-arm metabolic advantage, so “put it EARLY” is a low-confidence preference, not an evidenced win. The clearer harm signal is muscle loss from an ad-libitum short window. Net practical read: eat earlier if you do a window, and keep protein/meal-frequency up inside it.
- Everything moved is a surrogate — and the ledger is mixed, not all wins. eTRF’s insulin/BP/ oxidative-stress improvements and its morning-fasting triglyceride/total-cholesterol rise are all markers with no hard outcome and no weight change — legitimate as a target only if transmission to a patient-important outcome is evidenced, which it is not here (Surrogate Outcomes).
- Ranks LOW as a lever. Meal timing is heavily discussed relative to a small, surrogate-only, weight-independent effect — the telos’s attention-is-an-anti-signal rule applies with force (Layer 1 - Ranking Interventions for a Stratum).
- Which IF form, if any? ADF > TRE/WDF on weight, but the gap is trivial and short-lived. If someone is set on an IF pattern for weight, the pooled evidence favours alternate-day fasting over a daily window or 5:2 (ADF vs TRE −1.69 kg, ADF vs WDF −1.05 kg) — but every one of these is below the 2.0 kg clinical threshold and gone by ≥24 weeks. The honest ranking: pick the pattern you will adhere to; the regimen label is second-order to the deficit and to sticking with it. (Semnani-Azad et al., 2025)
- The realistic pharmacological alternative dwarfs it (Layer 1). Semnani-Azad’s own contrast: «GLP-1 receptor agonists, such as semaglutide, result in substantial weight reductions of 10-15% body weight (approximately 8-12 kg for an 80 kg individual)». (Semnani-Azad et al., 2025) That is roughly 3-8x any IF effect (wiki arithmetic against the pooled IF magnitudes above), on drugs the source credits with HbA1c and cardiovascular-risk improvement. (corrected 2026-08-08) For a weight-driven decision at meaningful baseline risk, IF is a small lever beside the drug class (Semaglutide for Cardiovascular Risk in Obesity, GLP-1 Non-Cardiometabolic Effects and Safety).
Certainty and gaps
confidence: medium(raised from low on Liu). The weak weight-loss rationale leg is now firm — two RCTs, including a 12-month high-tier isocaloric head-to-head (Liu, n=139) that isolates the window and finds it adds nothing. The real early-timing benefit leg still rests on a single n=8 crossover (eTRF) with the fasting-duration confound and morning-only measurement — genuinely preliminary. And Liu is a disconfirming constraint on it, not out of bearing (corrected 2026-08-08). Liu ran an early-ish window (8am-4pm) at matched calories — the same weight-independent comparison design as eTRF — for 12 months and found no between-arm metabolic advantage on any secondary, including a 12-month systolic-BP between-arm difference of «−0.3 (−3.7 to 3.1)», a CI that excludes eTRF’s −11 mm Hg. It is not a clean refutation (Liu’s window is 8 h not 6 h, its control window was itself early-ish ~10 h, and its population is non-diabetic Chinese adults vs eTRF’s prediabetic US men), so the two do not settle the early-timing question — but they collide, and the n=8 crossover benefit must be read against the 12-month RCT null, not shielded from it. No hard outcomes anywhere; weight/waist are measured endpoints but CV/mortality trajectory is unmeasured; eTRF is men-only, prediabetic; Liu is non-diabetic Chinese adults, no diabetes/CVD.- Gaps (G): the calorie-matched TRE-vs-continuous-restriction RCT is now held (Liu), and the TRE-vs-CR SR/MA is now held too — the Semnani-Azad 2025 network-MA (section below) confirms Liu’s single-RCT null generalizes across the pooled TRE literature. Still open: an early-vs-late head-to-head at the same window length (the eTRF question at scale); a breakfast-skipping RCT; any hard-outcome or trajectory data (Semnani-Azad is surrogate-only, median 12-week follow-up).
- Adherence caveat cuts both ways: eTRF participants found eating within 6h harder than the 18h fast, so the efficacious protocol may be the least feasible one — an 8h+ early window is the realistic target.
Self-critique [run 2026-07-29, before commit]
- Overclaim check. The central claim is a decomposition, not a discovery that early TRE works —
worded as such throughout. The eTRF leg is explicitly held at n=8/surrogate/confounded;
confidence: low. Stripping the page’s most dramatic move (the −11 mmHg BP number) leaves the decomposition standing, so LOW is not resting on drama — consistent with the confidence grade. - Not-joined / fake-tension check. Explicitly ran it and declined to file a tension; the parameter table’s NO column is the evidence, per op-weave 2a.
- Independence. No
[E-independent]claimed — the two trials are not independent backing for one claim; they are two exposures. Correct classification is A (decomposition), not E. - Attribution. Both trials read directly; quotes located via source. TREAT authors’ approving citation of Sutton is the counter-passage that closes the apparent clash — verified in TREAT’s Discussion.
Self-critique [re-run 2026-07-31, Liu append, before commit]
- Overclaim check. Liu is stated precisely as a null for the window’s ADDED effect on an isocaloric
deficit — not a null for weight loss (both arms lost 6-8 kg; the deficit worked). The wording «did
not produce greater weight loss than the regimen of daily calorie restriction» and «caloric intake
restriction explained most of the beneficial effects» are the source’s own; I did not upgrade the
between-arm null (P=0.11, CI crossing 0) into a claim of equivalence — reported as no significant
difference, which is what it is. Confidence raised low->medium survives stripping the −1.8 kg number:
even without it, two RCTs + the field carry the window is mainly a deficit-delivery vehicle; the eTRF
early-timing leg is untouched and stays weak, so
highis not earned. - Parameter-table discipline. Both cross-source claims (Liu vs TREAT on the isolated quantity; Liu vs TREAT on lean mass) carry a matched-parameter table with a same quantity? column, individually quoted. The tables’ NO cells are the point: Liu’s between-arm null and TREAT’s window-vs-adlib null are not the same comparison — flagged, not laundered into one.
- type-E guard (single RCT vs the MA). Liu is a single RCT; Semnani-Azad 2025 is an IF network-MA that almost certainly includes Liu -> shared evidence, F not E. No type-E backing asserted; the AWAITS states the F/shared expectation explicitly so it is not later mis-scored as E.
- Counter-passage. Read Liu’s full Discussion end to end before writing the lean-mass refinement — Liu itself cites Lowe (#11) and states the opposite lean-mass result with the protein explanation, so the contrast is the source’s own joined issue, not a manufactured tension. Verified.
- Attribution. All quotes emitted via
bin/cite.pyin the verified «…» form, slug-pinned to chunk 01.
Self-critique [run 2026-08-04, Semnani-Azad NMA append, before commit]
- The not-E call is the load-bearing one, and it is verified, not assumed. Checked Semnani-Azad’s
reference list directly: Lowe TREAT (ref 81) and Liu (ref 79) are IN the pooled included-studies range;
Sutton eTRF (ref 134) is a Discussion-only citation, not pooled. So the convergence with the held nulls
is shared-trials -> type-F/A, and
[E-independent]is not asserted anywhere in the append. The trial-overlap table and the parameter table are the evidence the check ran. A RAG over Semnani-Azad alone would reproduce the held claim (it re-pools TREAT+Liu) — the textbook laundered-E case, declined. - Overclaim check. The pooled TRE-vs-CER −0.39 kg and ADF-vs-CER −1.29 kg are reported as the source states them, with the MID-relative “trivial”/“limited clinical impact” qualifier attached (not upgraded to “equivalent”). The duration finding is stated as the authors’ own attribution (too-few-trials + adherence + adaptation), not asserted as established mechanism. Surrogate-only + median-12-week caveats kept. [Correction 2026-08-08: this bullet certified a guard the body did not carry — at the time of writing, the body stated only adherence + adaptation and had DROPPED the source’s first-listed too-few-trials explanation. The too-few-trials qualifier has now been restored to the body (Semnani-Azad section, finding (3)), so the certification is accurate going forward. Flagged rather than silently rewritten, per the append-don’t-rewrite discipline.]
- Confidence held at
medium, not raised to high. The weight-loss-rationale leg is now MA-firm (99 RCTs), but the early-timing (eTRF) leg is untouched and still n=8, and everything remains surrogate-only with no hard outcomes or trajectory data — so the page as a whole does not earnhigh. Stripping the −0.39 kg number leaves the decomposition and the deficit-delivery reading standing. - Attribution. All quotes emitted via
bin/cite.pyin the verified «…» form, slug-pinned to chunk 01.
The 65% lean-mass figure - read the denominator (deliverable-critique, 2026-08-01)
A reasonable “hard to believe” (deliverable-critique). Three caveats make the alarming percentage fragile, and two are already on this page:
- Read the absolute correctly — it is 1.10 kg, not 0.65 kg (corrected 2026-08-08). The 65% was computed on the in-person DXA cohort’s within-arm loss of 1.70 kg: «the average weight loss in the TRE group was 1.70 kg. Of this, 1.10 kg (approximately 65% of weight lost) was lean mass; only 0.51 kg of weight loss was fat mass». So the absolute lean loss to read is 1.10 kg, not a fraction of the near-zero total-cohort figure. The earlier “~0.65 kg lean” was 65% applied to the wrong base and understated the observed lean loss. (Lowe et al., 2020)
- The near-zero denominator belongs to a DIFFERENT cohort — do not blend them. The total-cohort between-group net change was tiny and not different from control («-0.26 kg … P = .63»), but that is a different cohort, instrument (at-home scale vs DXA visit), and quantity (between-group difference vs within-arm loss) from the 65% denominator, so it does not license reading the 65% as a near-zero-base ratio. The statistic that is genuinely NOT denominator-fragile is the between-group appendicular-lean- mass deficit — «a significant difference between groups (−0.47 kg; 95% CI, −0.82 kg to −0.12 kg; P = .009)» — a real loss on a non-near-zero base. (Lowe et al., 2020)
- It did not replicate (Liu, with a daily protein shake, showed no between-group lean-mass difference - the section above), and protein is one candidate explanation for the non-replication (see the protein caveat above), not the window itself.
So the honest read: the percentage share is fragile and base-dependent (though the absolute 1.10 kg and the −0.47 kg between-group ALM deficit are not); the durable instruction is keep protein up in any window, and milder schedules (12:12, 14:10, or simply not eating in the ~3 h before sleep) carry even less of this risk.