systematic review in the Cochrane tradition, as a public repo: standing queries instead of
stale searches, plots as YAML instead of PDF images, judgments in git history instead of editorial
memory. this surface is exhibit 001 — the cannabis & cancer corpus as an
evidence map: mechanistic and early-phase, honestly not poolable.
not medical advice. preclinical strength is not clinical proof in either direction.
This is an evidence library you can re-run — not a brochure you have to trust.
Empty cells are findings — not failures.
exhibit 001 is early-phase and honestly not poolable — cannabis & cancer, named once.
Claims arrive graded or not at all — not medical advice.
What if the empty cells were the most honest answer?
A pooled number on an evidence map is a category error — this page refuses it.
Empty cells are findings — not failures.
exhibit 001 is early-phase and honestly not poolable — cannabis & cancer, named once.
Empty cells are findings — not failures.
This collection contains early research. It does not show that a treatment works.
Plant combination
Synthetic isolate
CBD isolate
Supportive care
Evidence present
Evidence present
No evidence in this collection
Antitumour research
Early research
No evidence in this collection
No evidence in this collection
Empty cells are the finding — this corpus is early-phase and honestly not poolable.
honest seams, visible on contact: not medical advice · not poolable · preclinical strength is not clinical proof in either direction.
WHY THIS EXISTS — the four things Cochrane cannot retrofit
Cochrane
BiGen
searches
run by hand; reviews go stale for years
standing queries on a schedule — living is mechanical
access
paywalled across most of the world
a public git repo — clone is the whole access story
data
forest plots are images inside PDFs
plots are YAML — anyone recomputes the number
dissent
write a letter to the editor
fork, change the rating, show your work
a Cochrane forest plot is an assertion you trust; a BiGen plot is a computation
you re-run. architecture: bigen/ARCHITECTURE.md ·
integrity schema: bigen/integrity/SCHEMA.md ·
design law: DESIGN-BRIEF-02.
every record below was verified against PubMed via E-utilities on 2026-08-19 — not against recollection.
STATE OF THE EVIDENCE · EXHIBIT 001 (cannabis & cancer) — one summary, three reading levels (identical numbers, only prose changes)
reading level is a preference, not a tier — nobody proves anything to get the
plain version, and technical is not a reward (DESIGN-CONSTRAINTS §5). all three render from the same constants.
EXHIBIT 001 · EVIDENCE MAP — cannabis & cancer
this corpus is mechanistic + early-phase: nothing in it is poolable, so no pooled estimate
will ever render on this page. a pooled number on an evidence map is a category error — the exact error this
object type exists to prevent (BRIEF-02 §2 negative control). where a published pooled analysis exists
(Zhang 2015, lung cancer), it is cited as their computation, never rendered as ours.
claim
verdict
sources
verdict set: ✓ SUPPORTED · ≈ MIXED · ? UNTESTED · ✗ REFUTED · ⊘ REFUTED-AS-UNIVERSAL.
UNTESTED means no study has asked the question — a finding, not a gap in our work and not a negative result.
claims arrive graded or they do not arrive (DESIGN-CONSTRAINTS §6).
THE GAP MAP — oncology trials by intent × formulation: the empty cells are the finding
three of six cells occupied. empty is information: no synthetic-isolate antitumour
trial exists, and no CBD-isolate oncology trial of any intent exists. every occupied cell cites its trials below.
EXHIBIT 002 · EVIDENCE MAP — the COA panel undercount: a total wearing a panel's name
the founder's catch, documented from primary regulation: regulators require 4–16 analytes
against ~160 cannabinoids identified in the plant — a COA's "Total Cannabinoids" line is a panel floor
wearing a total's name. this corpus is regulatory tables + same-sample assays: not poolable — no pooled
estimate will ever render here. source of record:
RECEIPT_CANNABINOID_PANEL_UNDERCOUNT (Seat 3, 37a432f; Canada IHR figure corrected same-day at 72661ef in the sibling receipt).
claim
verdict
sources
the display law this exhibit exists to carry: the surfaces never render
"total cannabinoids" without naming the panel — total (n of ~160 known) —
three laws sharing one grammar: unmeasured ≠ zero · saturated fat ≠ one number · total ≠ panel sum.
the two honest boundaries — ship with the finding, always
the gap compound COUNT, not MASS — the "10–30% of mass unidentified" figure traces to 403'd trade press; never cited
the matrix NO figure describes seed — every source is inflorescence, plant, resin, or extract; the seed ceiling is the contamination mechanism, not panel width
and gate CB-4: does the project commission its own broad-spectrum
assay on its own COA — the one measurement nobody else publishes, with the funding flag firing on ourselves
(integrity Law A's next generation — the specimen ships first or the detector is decorative).
THE REGISTER — the search strategy is config that runs, not prose somebody retypes
loading…
sweep receipt: all 13 PubMed-indexed records in this
corpus checked against "Retracted Publication"[pt] on 2026-08-19 — 0 flagged.
Hussein 2014 is not PubMed-indexed: manual watch, no structured flag exists to fetch. this page is static and keyless;
the register re-verifies on schedule in the repo.
STUDY CARDS — the integrity layer, per study, symmetric by law
defaults are unclear / not_assessed, never favourable — an unassessed study displays as
unassessed. integrity renders beside the finding and never auto-downgrades it: a fork may weight it, the library
does not weight it for you (SCHEMA Law B).
THE INSTRUMENT — why the integrity layer exists (Lundh 2017, Cochrane methods review)
manufacturer-sponsored → …
RR
95% CI
favourable efficacy results
1.27
1.17–1.37
favourable conclusions
1.34
1.19–1.51
low risk of bias from blinding
1.25
1.05–1.50
results–conclusions agreement
0.83
0.70–0.98
read rows 3 and 4 together: sponsored studies score better on standard
methodology — the bias is not in the method section, it is in the sentence after the table. that sentence is what
results_conclusions_gap measures. Lundh et al. 2017, MR000033.
the instrument pointed at ourselves first:BiGen's own ECS map
attributed CBD's stem-cell effect to an "AMPK→SIRT1/autophagy pathway." AMPK appears nowhere in the source's
abstract, keywords, or MeSH terms (In Vivo 2026;40(1):222–234, PMID 41482390).
logged as CORRECTIONS-01 EDIT 1 — a library that only runs its instruments on other people's work has not implemented them.
THE MAP AS DATA — clone it, fork the verdicts, show your work
exhibit 001 as YAML (what this page renders; lives in the repo as bigen/evidence-maps/)
loading…
SOURCES — cite or the cell stays silent
External citations — each link opens in a new tab so this page and session stay put.