Source-verified transparency reports for Connecticut cannabis — built so every number can be traced back to the product's own Certificate of Analysis (COA).
🔗 Live web app: cannascope-ct.streamlit.app · 💻 Desktop download: see Releases · Current version: V17.4
Connecticut does cap Total THC for adult-use sale, and the cap has changed over time — but the program previously declared “no cap” for every year. V17.4 bakes in CT's real, dated, triple-verified caps and shows them in the standards tables:
| Product type | 2023 → Sep 30 2025 | Oct 1 2025 → (PA 25-166) |
|---|---|---|
| Flower / plant material | 30% | 35% |
| Concentrate / other non-vape | 60% | 70% |
| Vape cartridges | exempt | exempt |
| Edibles | 5 mg/serving · 100 mg/package | same |
| Medical-only / Legacy | no cap | no cap |
- 🧪 Triple-verified against CGS Chapter 420h, CT PA 21-1 (RERACA), CT PA 25-166 (raised the caps eff. 2025-10-01), and CT PA 26-8 (keeps 35% flower / drops the concentrate cap, eff. 2026-10-01 — noted, not yet applied). Shown as VERIFIED with citations + an as-of date.
- ⚖️ Correctly qualified — caps apply to adult-use retail sale; medical-only, “legacy,” and all vape cartridges are EXEMPT; edibles use a mg dose limit, not a %. Presented as reference standards (not auto-flagged violations — the market flag is brand-level, so high-potency items may be lawfully medical-only).
- 🧾 Conflicting-COA tables now carry product + producer — each Multiple / Conflicting COA comparison shows a tinted Product and Producer band, not just lab names and dates.
Detection logic is unchanged (
ANALYSIS_VERSIONstays 17.3.0); this release fixes regulatory-reference accuracy + report identity detail. The microbial / aerobic / pathogen / heavy-metal standards were already verified and are unchanged.
V17.3 adds a scientific classification layer on top of the existing fail→pass / COA-conflict detection. The core principle: mold, yeast, and microbes can be reduced or killed by validated remediation — but chromium, lead, arsenic, cadmium, mercury cannot (they are elemental), and neither can mycotoxins (killing the mold doesn't remove the toxin). So a heavy-metal fail→pass is no longer treated like ordinary mold remediation — it's surfaced as the critical scientific, regulatory, and consumer-disclosure finding it is. This changes classification, severity, and section routing — NOT data values (published findings are byte-identical to V17.2).
- 🛑 New "Non-Remediable Elemental Contamination — SERIOUS CONCERN" section — every elemental heavy-metal and mycotoxin over-limit / fail→pass event with full forensics (failing value · limit on its COA · %/fold · both labs + dates · later value or "generic below action limits" · itemized? · product-type elevation · both clickable COAs · reversal), the scientific framing, and the legitimate explanations. Never alleges fraud.
- 🧭 Class-aware everywhere — a microbial reversal reads "potential remediation/retest event"; a heavy-metal/mycotoxin reversal reads "NON-REMEDIABLE — requires documented explanation." The Labs-With-Altered-Pass/Fail table gains a Class / Interpretation column.
- 🔎 Transparency-failure logic (generic "below action limits" with no itemized values), ⚗️ Severe Chromium Exceedance (≥ 2×) with a side-by-side view, 🍞 mycotoxin-clearance check after a mold remediation.
- 🧾 Coverage-Gap Diagnosis — every COA held out of findings is now classified by its exact reason (OCR / broken link / multi-product routing / source mismatch / parser error / extraction), in the Coverage Integrity Summary +
coverage_gap_diagnosis.csv.
Every flag is a lead to verify against the official COA — never a conclusion, and never an allegation of fraud.
This was the biggest upgrade since V17. It makes the report harder to argue with (every record is accounted for and the math is checked), more complete (over-limit results can no longer hide), and much nicer to read.
Some lab reports say "PASS / Below Action Limits" at the top, yet a result further down the same document is over its printed limit (or marked FAIL). Before, CannaScope trusted the top-line PASS and those results were quarantined. Now every analyte is judged against the limit printed on that same COA — so an over-limit line always surfaces in its contaminant section, flagged "COA Marked Pass But Contains Over-Limit Result," shown as a documented internal contradiction (the header says pass, a body line is over). It is never presented as an outside-confirmed failure — the COA's own PASS is shown right next to it.
The cover now shows Report Validation (how complete the data/parsing is) and Findings Validation (were the published numbers each re-verified against their own COA) side by side. A coverage gap no longer makes verified findings look "failed," and vice-versa.
A new Record Accounting table places every product in the window into exactly one bucket (reviewed · reused from prior verified runs · excluded out-of-window · date-failure · unknown) and the totals must add up to the window — the report refuses to publish if they don't. No more "where did the other records go?"
A dedicated Heavy Metal Coverage page shows, per metal, how many COAs reported it vs how many actually parsed — and says plainly "no findings can currently be claimed" for any metal with zero extraction. The report never implies coverage it doesn't have.
A new Top 100 Highest-Cannabinoid Flower table lists the strongest flower statewide, highest first, with no artificial display cap — biologically implausible values (>45%) are shown and clearly flagged rather than silently dropped.
A clearly-named section surfaces COAs whose own numbers are internally impossible — Total THC greater than Total Cannabinoids, a total below the sum of its parts, negatives, or impossible percentages. These are data-integrity alerts held out of findings (routed for manual re-read), never treated as a safety finding or an accusation.
The boundary-clustering review adds round-number, digit-preference and nearest-neighbor screens, and every statistical callout now carries an explicit significance tier (interesting · unlikely · statistically significant · highly significant) with the threshold stated. Producer flag rates now show a 95% confidence interval and mark low-sample groups, so a small sample is never read as a firm conclusion.
Every product listed anywhere in the report now shows its test date and a clickable link to its COA — every section, every time.
COA links no longer wrap or bleed between columns, tables now use the full width of the page (no more cramped columns with wasted space), and the lab-result-change comparisons read as big, obvious before → after values.
A fully-verified recent-year report was previously being branded "DIAGNOSTIC — not ready" because heavy metals are only required on a fraction of products. That readiness logic is fixed: it now judges the panels that are actually tested everywhere, so a clean, fully-verified report reads as a clean pass with warnings.
Every flag remains a lead to verify against the official COA — never a conclusion, and CannaScope is not affiliated with the State of Connecticut, any lab, or any producer.
- 🛰️ Honest live-verification gate — the cover states how many products were re-verified against their live source this run; a 0-live run is unmistakably stamped
CACHE-REPLAY, and--live-verify/--validateforce genuine live re-fetching. - 📊 "Convenient Lab Result Groupings" — the statistical screen for results clustering just below a pass/fail limit (binomial / z-score / chi-square / Fisher / cliff-effect / 0–100 Convenience Score) — a review signal only, never a fraud claim.
- 🧮 Laboratory Data Consistency Flags & Biologically Implausible High THC Flower Review (accurate renames), Coverage Integrity Summary, decluttered/centered cover blocks, tier-banded producer table, Unicode-clean
µg/kgrendering.
- 🔎 Look up a product by name / brand / batch / COA number → a plain-English PDF.
- 🏛️ Statewide report over any date window → one downloadable PDF.
- 🧪 Reads each COA for 🦠 microbials (yeast & mold, aerobic bacteria, pathogens) · ⚗️ heavy metals · 🌾 pesticides · 🧴 residual solvents · 🍞 mycotoxins · 🌿 cannabinoids/potency · ✅/❌ pass-fail.
- 🛡️ Live-first (the cache is only a speed hint; live wins), multi-product COA isolation (never cross-attributes), date-window integrity, below-detection aware, and fail-loud (an empty or non-reconciling report stops; it never ships a silent blank).
- 📦 Downloadable PDF + CSV exports for every section + a technical appendix.
# Statewide (live-first). Add --validate for a full-window, 100% live-verified forensic run.
python CannaScope_CT_V17.py statewide --since 2024-01-01 --until 2026-06-10
python CannaScope_CT_V17.py statewide --since 2024-01-01 --until 2026-06-10 --validate
# Single-product consumer report
python CannaScope_CT_V17.py concern --exampleEach OS zip bundles the self-contained CannaScope_CT_V17.py (embeds the triple-verified COA dataset) + README + requirements + LICENSE + install/run scripts. Python 3.9+; pip install -r requirements.txt.
CannaScope CT is an independent, informational transparency tool — not medical, legal, or professional advice, and not affiliated with or endorsed by the State of Connecticut, any lab, or any producer. Every flag is a lead to verify against the official COA, not a conclusion. Provided as-is, no warranty. All prior releases are preserved; this is an additive release.
🔧 Technical changes (V17.2) — for developers / auditors
Detection / classification (ANALYSIS_VERSION → 17.2.0):
- Over-limit-under-PASS:
assess_extractionno longer routes a top-PASS + genuine body over-limit to UNCERTAIN/held; it marksp._coa_pass_overlimitand publishes the finding (ambiguous FAIL-status without a verified numeric exceedance still routes to review). Newcoa_pass_overlimit_lines(p)(robust on cache rows),COA_PASS_OVERLIMIT_FLAG, debugcoa_pass_with_overlimit_lines, run-log count, and a "COA Marked Pass — Over-Limit Line Items" section. Surfaces in analyte/contaminant tables + producer/lab aggregation automatically. - Cannabinoid math:
thc_conflictnow also flags Total Cannabinoids < sum(THCA + Δ9 + total CBD). - Confidence:
assess_extractioncredits extraction breadth via_extraction_richness(promotes, never demotes). - Year-readiness:
COAFormatLearner._core_coverage/_verdictjudge metals coverage relative to where reported (not ÷ all products), so sparsely-tested metals no longer force NOT-READY; arsenic presence regex no longer matches the English word "as".
Accounting / validation backbone (single source of truth):
VerificationAccounting+build_verification_accounting(fail-loud reconciliation),record_accounting_buckets(record partition, fail-loud Unknown),consistency_audit(build-time,sys.exit(5)on contradiction),write_remediation_report(Final Remediation Report file). All wired beforebuild_pdf. TiersFINDING_VALIDATION_FAIL/DIAGNOSTIC_COVERAGE_INCOMPLETE; dual Report/Findings status on the cover.
New analyses / helpers: heavy_metal_coverage_rows, top_cannabinoid_flower_rows, statistical_screens + significance_tier, _wilson_ci + CG_MIN_GROUP_SAMPLE.
Presentation: date + clickable COA on every product table; coacell fixed (8pt, splitLongWords=0, wordWrap=None) so COA links never wrap/bleed; _fit_widths scales all tables to ~96% of usable width (stretch-capped, no overflow); big centered before→after in lab-change cases.
Tests: _test_verification_accounting.py, _test_presence.py, _test_consistency_math.py, _test_overlimit.py, plus S8/S10 assertions in _test_convenience.py. Build: python3 _make_v17.py. Forensic: --validate (needs the certifi CA bundle in restricted shells).