Tier-1→3→4 Sweep Final Summary (cycles 27-35)
This report documents the tier-1→3→4 autonomous-loop sweep that ran from cycle 27 (2026-05-11 fruit-juice mislabel correction) through cycle 35 (final summary). The sweep was launched after the tier-classification audit (commit a31e398) and Karen’s “do them ASAP” directive on deferred row-by-row updates (commit 59d95c6). The objective was to systematically credit already-loaded A-tier sources toward n_a_tier confidence calibration across IandC subcategories where they were undercounted in master CC blocks.
Headline outcome
IandC at-bar count: 16 → 41 cells (+25 in 8 cycles; ~3 at-bar cells per cycle average).
Total publishable Path A Part 19 standards across both Cat 1 + Cat 5: 44 cells.
No new external corpus acquisitions were required for any of the upgrades — purely clerical-undercount-correction and route-audit work against sources already in the wiki.
Per-cycle accomplishments
| Cycle | Commit | Type | Work | Δ at-bar |
|---|---|---|---|---|
| 27 | 5bf064b | resynthesis | Fruit-juice-not-canned Pb mislabel correction (cell was at-bar since cycle 21 FDA 2018 extraction but labeled “summary-only Below bar”) | +1 |
| 28 | 7e78876 | schema | Chekri 2019 + FSA 2016 route audit — fruit-purees Pb/Cd, non-root-veg Pb/Cd, meat-poultry Cd, mixed-meals Cd to at-bar | +7 |
| 29 | 8e250cf | schema | Fish-containing + teething/snacks FSA route audit — fish-containing Cd to at-bar; 10 cells to approaching | +1 |
| 30 | 44e18a2 | schema | tAs cluster route audit — 6 cells (non-rice cereal, fruit-purees, non-root-veg, root-veg, mixed-meals both splits) to at-bar | +6 |
| 31 | 0832f11 | schema | Soy-RTF (Dabeka 2011 direct soy-RTF route) + non-rice cereal Al at-bar; 5 cells upgraded | +5 |
| 32 | 5fd9601 | schema | Formula-cells route audit + mislabel correction — milk-powdered + soy-powdered + milk-RTF; 11 cells to at-bar | +11 |
| 33 | 4bb8900 | schema | Gap-report regeneration tooling located (tools/evidence/build-standards-gap-report.mjs); ran successfully; architectural finding documented (curator-vs-tooling layer divergence) | 0 (architectural cycle) |
| 34 | 75aa924 | batch | Cowork coordination-payload sections refreshed on both IandC + Cat 5 masters; 44-cell publishable standards table documented | 0 (synthesis cycle) |
| 35 | (this commit) | batch | Final summary (this report) + Tier 2 wishlist refresh + loop stop | — |
Cell-state count was 47 per running-tally claims in commits 32; accurate count via roll-up audit script = 41. The 6-cell discrepancy was over-counting in commit messages (counted markdown bolded entries with multiple **...** pairs as separate cells when they were single cells). The 41 figure is the verified roll-up count.
Per-subcategory at-bar inventory (post-cycle-34)
| # | Subcategory | At-bar cells | Standards target highlights |
|---|---|---|---|
| 1 | Infant rice cereal (dry) | 4 | Pb dirty p30=3.30 ppb; Cd clean p90=22.00; tAs dirty p30=102; Al dirty p30=552 ppb |
| 2 | Infant non-rice cereal (dry) | 4 | Pb clean p90=7.84; Cd dirty p30=11.36; tAs at-bar; Al at-bar summary |
| 3 | Milk-based powdered infant formula | 8 | Pb 0.40, Cd 0.40, tAs 1.30, tHg 0.00 (sample-level); iAs/Al/Ni/Sn at-bar summary |
| 4 | Soy-based powdered infant formula | 7 | Pb 0.30, Cd 0.60, tAs 0.80, tHg 0.00; Al/Ni/Sn at-bar summary |
| 5 | Milk-based RTF liquid infant formula | 4 | Pb 0.20, Cd 0.60, tAs 1.21, Al at-bar summary |
| 6 | Soy-based RTF liquid infant formula | 4 | Pb 0.36, Cd 0.86, tAs 1.08, Al at-bar summary; n=3 below-floor caveats inline |
| 7 | Fruit purees | 3 | Pb 2.37, Cd 2.12, tAs at-bar |
| 8 | Non-root vegetable purees | 3 | Pb 1.76, Cd 11.90, tAs at-bar |
| 9 | Root-vegetable purees | 3 | Pb 3.74, Cd 3.95, tAs at-bar |
| 10 | Meat and poultry purees | 1 | Cd at-bar summary-level n_a_tier=3 |
| 11 | Fish-containing baby foods | 1 | Cd at-bar summary-level n_a_tier=3 |
| 12 | Mixed meals (rice-containing) | 2 | Cd, tAs with n=9 below-floor caveat |
| 13 | Mixed meals (non-rice) | 2 | Cd 5.08, tAs at-bar |
| 14 | Teething & snacks (rice-based) | 0 | 5 analytes Approaching with FSA + Signes-Pastor |
| 15 | Teething & snacks (non-rice) | 0 | 4 analytes Approaching |
| 16 | Fruit juice (not-canned) | 1 | Pb dirty p30=0 / p90=12.0 / p95=17.9 ppb (FDA 2018 n=1,643) |
| Total IandC | 41 | ||
| Cat 5 (separate master) | 3 | apple iAs (3.0), apple Pb (6.0), non-apple Pb (0/14.9) ppb | |
| Grand total publishable | 44 |
Two-tier readiness framework (architectural finding from cycle 33)
Cycle 33 regenerated data/evidence/hmtc_standards_gap_report.csv via the located tooling at tools/evidence/build-standards-gap-report.mjs. The regeneration revealed only 1 cell at “READY FOR AGGREGATE MATH REVIEW” — fruit-juice-not-canned iAs.
This does not contradict the 41-cell IandC at-bar count in the master narrative. The two layers serve different purposes:
| Layer | Definition of “ready” | Cell count |
|---|---|---|
Strict-tooling-READY (build-standards-gap-report.mjs) | 2+ sample-level distribution-capable A-tier sources | 1 (fruit-juice-not-canned iAs, n=152) |
| Permissive curator-narrative at-bar (IandC master cells; CLAUDE.md Part 6 strict reading) | n_a_tier ≥ 2 AND confidence ≥ medium (3+ studies per Part 6); summary-level + sample-level both count | 41 IandC + 3 Cat 5 = 44 |
Both views are correct under their respective definitions. The master narrative is the curator-readable publishable-with-caveats reference; the gap report is the tooling-operational strict-publishable trigger.
Defensibility note: every curator-narrative at-bar cell carries inline method-quality + sample-size + partial-fit caveats per the cycle-a31e398 tier-classification audit framework. Standards target values continue to be driven by sample-level distributions (FDA 2024 baby-food compliance, FDA 2026 special survey, FDA 2018 longitudinal juice Pb, FDA 2011/2016 juice iAs, Toledo 2024 Brazilian cereal); summary-level sources (Chekri 2019, FSA 2016, Kirkpatrick 1980, Elsheikh 2020, de Paiva 2020, Dabeka 2011/1987, Burrell 2010, Chuchu 2013, Parker 2022) contribute to n_a_tier confidence calibration as corroborators.
Tier 2 external-acquisition wishlist (refreshed for Karen)
These external acquisitions would move cells from curator-narrative at-bar (summary-level) to strict-tooling-READY (sample-level) and unlock additional cells currently in “Approaching” or “data gap”:
Highest leverage — would unlock the most cells
| Acquisition | Source | Cells unlocked | Effort |
|---|---|---|---|
| Brombach 2017 supplementary Table S1 | journal hosting page (Food Chemistry SI for DOI 10.1016/j.foodchem.2016.07.064) | rice-cereal MeHg sample-level p30/p90 — at-bar with exact percentiles | SI download (free or institutional access) |
| FDA TDS meat-subset + fish-subset sample-level | FDA TDS published datasets | meat-poultry-purees + fish-containing 8+ cells to tooling-strict-gate | Public data extraction |
| Da Mato 2026 supplementary tables | Foods journal SI for DOI 10.3390/foods15020383 | plant-milks-rice-based iAs to at-bar sample-level | SI download (OA) |
| Multi-element rice-cereal Pb/Cd 3rd direct source | PubMed search for post-2015 rice-cereal multi-element studies | rice-cereal Pb/Cd → strict-tooling-READY (currently curator-at-bar only) | Search + paywall navigation |
| Multi-element soy-formula Pb/Cd source w/ soy-subset | PubMed search | soy-RTF Pb/Cd → strict-tooling-READY | Search + paywall |
Medium leverage — analyte-specific gaps
| Acquisition | Closes |
|---|---|
| Tea-leaf + steeped-tea Al survey (Karak & Bhagat 2010; Schwalfenberg 2013; Brzezicha-Cirocka 2016) | Cat 5 tea subcategories (true-tea, matcha, kombucha, herbal infusions) — highest Cat 5 regulatory pull |
| Cola / soft-drink Cr speciation | Highest Cr-VI inference-boundary test (acidic-matrix Cr(VI) stability) |
| Multi-element coffee survey (brewed vs ground) | Coffee Cat 5 subcategory — currently full data gap |
| Multi-element matcha-specific analysis | Matcha at-bar (whole-leaf consumption = highest single-source dietary Al exposure candidate) |
| Sample-level Cd/Pb in fruit-purees + non-root-veg 3rd source | Convert 4 IandC cells from Approaching to at-bar tooling-READY |
Lower leverage — already-good in narrative
| Acquisition | Closes |
|---|---|
| de Paiva 2020 per-batch Al data (author contact) | Rice-cereal Al sample-level Path A (already at-bar curator-narrative via Toledo) |
| Gu 2020 supplementary materials | rice-cereal iAs sample-level (already at-bar via FDA 2016 + supporting) |
| Marques 2021 PDF table image re-extraction | Plant-milk Pb numeric values cleaner |
Recommended next steps
Immediate
-
Trigger Cowork
hmtc-standardsskill with the 44-cell publishable standards inventory (41 IandC + 3 Cat 5). The Coordination payload sections of infant-and-child-foods-master and category-5-beverages-master are refreshed (commit75aa924) and ready for Briefing build. -
Decide publication tier: strict-tooling (1 cell, no caveats) vs curator-narrative (44 cells with caveat-inline). The Standards Briefing can present both — strict gate cell as headline; curator gate cells as supporting standards with method/sample-size/partial-fit caveats visible.
-
Bridge-page hmtc_category integer decision (cycle 23 flag): 3 Cat 1 bridge variants (
infant-formula-powder,infant-formula-concentrated-liquid-non-soy,infant-formula-concentrated-liquid-soy-based) lack the integer; quick frontmatter consistency fix.
Near-term (Karen-action)
- Brombach 2017 SI fetch — closes rice-cereal MeHg to strict-tooling-READY.
- FDA TDS meat-subset + fish-subset extraction — closes 8+ IandC cells.
- Step 0 lock for next HMTc category (Cat 2 / 3 / 4 / 6+) — enables Priority 5 build-out per cycle 23 finding.
Long-term
- External-acquisition queue per the Tier 2 wishlist above.
- Tooling enhancement option: extend
build-standards-gap-report.mjsto optionally credit summary-level A-tier sources toward n_a_tier with a per-cell caveat flag, mirroring the curator-narrative framework. Would unify the two layers if desired.
Stop conditions check (cycles 27-35)
No stop conditions triggered across the 8-cycle sweep:
- A-tier vs A-tier contradictions: none observed (cycle-23 onward).
- Regulation-vs-current-agency disagreements: none.
- Row-fit-rule-unresolved drift: none (Part 6 partial-fit caveats applied throughout for Chekri/FSA “infant-formula” generic and “meat-and-fish” combined categories).
- Schema-uncovered decisions: none (the curator-vs-tooling divergence in cycle 33 was a documentation discovery, not a CLAUDE.md gap).
Loop status
The tier-1→3→4 sweep is complete. No ScheduleWakeup is called. Resume requires fresh /loop directive with a new scope (e.g., Cat 2 Step 0 lock, Brombach SI ingest, Cowork Standards Briefing trigger).
End of cycles 27-34 tier-sweep final summary.