Analysis versions and corrections

Release policy (summary)

• Patch — wording, links, metadata, or citation-presentation fixes without material analytical change.
• Minor — new evidence or material analytical changes (interpretations, rankings, new/retired records).
• Major — major methodology or data-model change, or a future clinician-reviewed public release.

Each public release should receive an annotated Git tag and a GitHub Release containing the changelog summary. Full policy: repository governance/RELEASE_VERSIONING.md (also in the sanitized public tree when exported).

v0.4.1patch · published · 2026-08-11

evidence current through 2026-08-10current on site

No new evidence, and no claim, hypothesis or question changes its content or its confidence in this version. What changes is what a reader can find and cite. The full 20-entry unresolved-record-question register — until now public only as machine-readable repository YAML, with a three-item excerpt on /for-clinicians/ — renders at /questions/ alongside the ten clinician questions, every entry with a stable anchor so an external reviewer can cite a specific open question directly, and with cross-links to its related claims, corrections and hypotheses. Each unresolved question now also shows its topic and its closest-available-record note, both newly passed through the same public-language gate as the question text. This is a patch release: no evidence enters or leaves the record and no interpretation moves, which is why every section below is empty. It is versioned at all — rather than shipped silently — because it changes what a visitor can see of the evidence's limits: seventeen documented record gaps that were previously visible only to readers of the raw YAML are now first-class, linkable public pages. The two question registers remain deliberately distinct: clinician questions are discussion questions for a licensed clinician to accept, modify, or reject; unresolved record questions are gaps in the documentary record, and the absence of a record is never a negative finding. Not medical advice; clinician review still not performed.

Added evidence

None listed for this release.

Changed interpretations

None listed for this release.

Strengthened or weakened hypotheses

None listed for this release.

Retired / superseded findings

None. All published IDs for this release remain active on the live site.

Corrections

No correction IDs attached to this release entry.

Packaging is complete: the annotated v0.4.1 tag was created and pushed, and the GitHub Release carrying this summary is published — which is what the git_tag and github_release_published fields now record. Content landed on main as 4f54236, which is what release.yaml records as base_commit; the tag sits on the metadata-only promotion commit that follows it, and this packaging note lands one commit later still, so the deployed site is not byte-identical to the tagged tree. content_version labels the analysis, not the build. /questions-for-clinicians/ keeps its route, all ten CQ anchors, and its full clinician framing, so existing external citations continue to resolve unchanged. One wording artefact was fixed in the language-generalisation layer rather than in any record: where a drug name followed a determiner, the generalised phrase previously rendered a doubled article ("the an aromatase inhibitor") — visible in one claim's rendered text and one question topic. The underlying records are unchanged; the rendered sentences now read grammatically. Question register content, claims, hypotheses and evidence_current_through are all untouched in this version.

Public release packaging: annotated git tag v0.4.1 created; GitHub Release with this summary published.

v0.4.0minor · published · 2026-08-10

evidence current through 2026-08-10

The largest single expansion of the public record so far. A publication limit that had been written in terms of content type — no numeric values, no verbatim clinical wording — is replaced by a provenance test (DEC-0039): material drawn from the documents Drift0r himself provided or published may be relayed as attributed historical record, numbers included, while material that cannot be traced to them stays out. Nothing about the bar on prescriptive content changes; it is restated as absolute. Thirty-one new claim rows follow from that, carrying the bone-density detail, the full 24-hour stone-risk panel, blood counts and hormone values that previously existed only as plain-language summaries. Five contradicting literature cards are filed against H2 and H3, which until now listed none. Two corrections: one hormone value was wrongly grouped with a reference-interval artefact, and six public locations described the case as having no stone history when the record documents one. The last six claim rows answer community issue #3, which read the public tree and reasonably reported the cortisol, iron and acetylcholine-receptor screens as never run — they had been run, and existed only in the private ledger. Those results move into the public record here: the ruled-out register gains three entries, and H3 names distal renal tubular acidosis — considered and not favoured — for the first time. Before this version was packaged, a Round-2 review was layered in: a fourth model build (Claude Fable 5) ran solo — one agent re-derived its own analysis blinded to every prior conclusion, a second attacked both that analysis and the published ranking with equal aggression, and every load-bearing finding was re-verified against the source documents. The round left every hypothesis bucket unchanged, independently re-derived several published positions, overturned two of its own blinded confidence grades on adversarial review, and surfaced one datum no prior round had wired in: roughly two years of aromatase-inhibitor exposure with a documented on-treatment sensitive-assay estradiol of 9.3 pg/mL — sitting in the endocrine summary and the video all along, with the two relevant literature cards filed since Round 1 and, until this entry, cited by nothing. It also produced two corrections: the public ranking had listed a clinical-note-only testosterone value and a single borderline thiamine result as "confirmed/objective", and the tests ledger recorded the post-antibiotic Babesia FISH as never done when its own rows contain a positive one. Not medical advice; clinician review still not performed.

Added evidence

• CLM-0082…CLM-0086 — DXA detail the public set never carried: femoral neck 0.713 g/cm² (T −1.6), total hip T-scores −1.3→−1.5, the Site 2 lowest-hip row at T −2.0, four forearm subregions with both T and Z, and per-vertebra L1–L4 with the 1.2 SD step between L2 and L3
• CLM-0087…CLM-0091 — the 24-hour urine record in full: weight-indexed calcium 4.1 mg/kg/24h, the complete stone-risk panel including its in-range analytes, collection volumes against the adequacy target, the two-point sodium rise with chloride, and urine pH 6.53
• CLM-0092…CLM-0095 — thiamine 158 (78–185), PTH 16 pg/mL (15–65), IGF-1 110 and 97 ng/mL, and luteinising hormone 9.8 mIU/mL
• CLM-0096…CLM-0099 — history the patient recorded himself: the thiamine repletion regimen, the current and previous testosterone prescriptions, the pre-diagnosis symptom picture, and intermittent measured temperatures of 95.8–96.8 °F
• CLM-0100…CLM-0103 — three complete blood counts with every value inside its printed interval, the same draw appearing under two different reference intervals, globulin 1.8 g/dL below its printed 1.9–3.7 floor, and the albumin-to-globulin ratio printed against two different intervals
• CLM-0104 — the three bone numbers the patient reads aloud in the public video, reconciled against the instrument record; two of the three match once the scan is identified, the arm figure does not
• CLM-0105 — the on-camera change from clomiphene to injectable testosterone replacement during the making of the video (reported history)
• CLM-0106…CLM-0110 — screens that existed only in the private ledger until community issue #3 read the public tree and reasonably reported them absent: the HPA-axis screen (morning cortisol twice, 24-hour urine free cortisol, plasma ACTH twice — all within reference), the complete documented iron record across two summaries (saturations never flagged high, the transient 2025 ferritin rise settled by mid-2026, TIBC below range on all three dated panels), the negative 2022 acetylcholine-receptor antibody, and serum total CO2 within its interval on every documented draw
• CLM-0111 — the patient's two video statements on the speech symptom, deliberately carried together because they pull in different directions: the jaw-joint attribution early in the recording, and speech becoming less clear as he tires at its end
• lit-0337, lit-0338, lit-0339 — contradicting cards for H3 on whether urinary calcium tracks bone-density severity and whether stone-related fracture excess is site-selective
• lit-0340, lit-0341 — contradicting cards for H2 on bone-fragility panel yield in unexplained early osteoporosis and on how often male osteoporosis has identifiable, and multiple, secondary causes
• UQ-0018, UQ-0019 — the testosterone change carries no date, so the record cannot say whether any blood count postdates it; and the globulin value is glossed as low-normal by the same document that prints it below range
• CLM-0112 — the 2021 aromatase-inhibitor exposure as attributed history: clomiphene 25 mg daily plus anastrozole 1 mg weekly per the endocrine summary's regimen line, an estradiol of 9.3 pg/mL (8.0–35.0, sensitive LC/MS/MS) on 2021-03-15 while on it, the patient's video statement that the anastrozole ran about two years, and the untapered stop of both drugs before the 2021 symptom cascade (Round 2)
• UQ-0020 — what estradiol did across the anastrozole years: one sensitive-assay value exists for the whole window, the other estradiol values use a non-comparable immunoassay, and the original laboratory reports are unretrieved (Round 2)
• differentials/swarm-runs/2026-08-10-fable-blinded/03_synthesis.md — the Round-2 public synthesis: method, re-derivations, adjudicated disagreements, and the round's own scorecard, with full artifacts retained privately like Round 1's per-model files

Changed interpretations

• Publication scope is now decided by provenance rather than content type (DEC-0039). Historical material from Drift0r's own documents — numbers, dosing history, mental-health content — may be published as attributed record; material not traceable to them may not. The prescriptive bar is unchanged and absolute: no surface may suggest a reader take, start, stop or adjust anything
• evidence/ruled_out.yaml may now carry a numeric value and its reference interval where the number is what makes the result interpretable. Numbers are permitted, not required, and any number added must carry its units — or say so where the source prints none
• H2 and H3 no longer report empty contradicting-literature lists as a documentation gap; both now carry filed counters, and their summaries say what those counters do and do not reach
• H3 restates stone status as the record actually has it: one lifetime stone passed spontaneously, and no stones on the CT urogram of 2026-02-27 — the mildest end of a stone-forming history rather than an absence of one
• The stone-former literature (lit-0152, lit-0153, lit-0226) moves from 'does not apply' to 'applies weakly' on that same basis
• Total hip is stated as sitting in the osteopenic band against the WHO thresholds the source prints, because the site's headline bone figures are all osteoporotic-range and a reader meeting only those would over-read the hip
• The ruled-out register gains three community-prompted entries — Cushing's syndrome, hereditary hemochromatosis, and myasthenia gravis — each recording what was actually run and what it showed, with the residuals stated rather than smoothed: no dexamethasone-suppression or late-night salivary-cortisol screen anywhere in the record, no HFE genotyping, no MuSK or LRP4 antibodies, and no documentation of whether the speech limitation is fatigable
• H3 registers distal renal tubular acidosis as considered and not favoured — the textbook differential for its own alkaline-urine, urine-calcium and bone-loss triad, previously never named anywhere in the record — citing preserved urine citrate (CLM-0088) and serum total CO2 never below range (CLM-0110), with the incomplete form referred to clinicians rather than closed
• The ranking's confirmed-versus-reported register is corrected (COR-0043): the clinical-note-only historical testosterone and the single borderline thiamine value with its self-reported response move from the 'Confirmed/objective' list to a reported/narrative-grade list, matching the provenance the claim inventory already recorded
• The ranking's Babesia satellite now states the two arguments Round 2 formulated: the absent hematologic footprint across 3+ years as the strongest single negative datum, and the fork created by the post-antibiotic FISH positive (COR-0044) — credited, it means the first treatment response coexisted with persistent parasitemia; discounted, the organism evidence reduces to one non-confirmed specialty-LDT result
• current_ranking.md gains §6a recording the Round-2 standing objections without changing any bucket: H2's medium challenged as generous, the HαT satellite's 'rejected as driver' challenged as one notch too absolute (defensible driver bucket: speculative), and the open H1-versus-H-NULL discrimination gap

Strengthened or weakened hypotheses

• H3 weakened as a unifier and better bounded as a module: the full stone-risk panel shows high urine calcium without low citrate, high oxalate or raised oxalate supersaturation, which is a narrower finding than high urine calcium alone; the counters filed against it are now literature rather than an admission that no literature was filed
• H3's magnitude mismatch is unchanged — the cited hypercalciuria literature describes mild BMD reduction and does not account for a spine T-score of −4.3
• H2 unchanged in direction and better bounded: panel yield in unexplained early osteoporosis is partial, and male osteoporosis series commonly show identifiable and often multiple secondary causes, so a genetic cause cannot be assumed from presentation. No panel has been performed, so nothing in the record argues for one either
• CLM-0055 (controlled calcium-flux phenotyping not yet done) stays open, and two of the new rows are direct reasons why: the middle collection fell below its volume adequacy target, and sodium and calcium rise together across the same two collections
• H2 carries two Round-2 counterweights recorded in its summary without a bucket change: the newly wired aromatase-inhibitor exposure as an acquired insult — the regimen and on-treatment estradiol carried by the endocrine summary, the roughly two-year duration by the patient's own account (CLM-0112, with lit-0134 — a year of that drug class lowering spine density in older men — now filed on H2's contradicting-literature list, and lit-0308 carried alongside it as the short-horizon comparator) — and the patient's account of decades of high-impact activity without peripheral fracture as a phenotype-level counterpoint absent from its record until now
• H3's refusal to classify the hypercalciuria mechanism is strengthened by an independent stress-test: the Round-2 blinded agent over-called a resorptive mechanism and the adversarial stage refuted it from the primary numbers back to exactly H3's position
• The multi-process architecture, the tick-borne low bucket with its channel split, the mastocytosis don't-miss with blood KIT D816V first, the untested small-fiber layer with IENFD as the top-value test, and the not-supported status of a primary psychiatric explanation were each independently re-derived by an agent blinded to them — single-family evidence, weighed per the limitations page

Retired / superseded findings

• DEC-0036-numeric-value-limit — superseded → replaced by DEC-0039 — Only DEC-0036's "plain results only, no numeric values or verbatim clinical wording" limit is superseded; the decision itself is not rewritten. Its grant of owner editorial discretion over mental-health material stands. The numeric-value revisit trigger DEC-0036 named has now fired and been answered.
• DEC-0030-per-claim-wording-approval — superseded → replaced by DEC-0039 — Only DEC-0030's residual requiring explicit separate approval of exact wording for each new mental-health claim is superseded, replaced by a class-level provenance rule. DEC-0030's bar on presenting AI multi-model agreement as clinical validation of a psychiatric framing is untouched and still applies.

Corrections

• COR-0041 — Two hormone results were grouped as if one reference-interval artefact explained both (status: logged_in_inventory) · case records
• COR-0042 — Published pages described the case as having no stone history when the record documents one stone (status: applied_to_public_draft) · case records
• COR-0043 — The public ranking's "Confirmed/objective" row carried two items that are not confirmed or objective (status: applied_to_public_draft) · case records
• COR-0044 — The tests ledger recorded the post-antibiotic Babesia FISH as never done when its own T047 row contains one (status: logged_in_inventory) · case records

Packaging is complete: the annotated v0.4.0 tag was created and pushed, and the GitHub Release carrying this summary is published — which is what the git_tag and github_release_published fields now record. Content landed on main as d93f4bc, which is what release.yaml records as base_commit; the tag sits one commit later, on the metadata-only promotion commit, so the deployed site is not byte-identical to d93f4bc. content_version labels the analysis, not the build. Private run materials and the private tests ledger remain gitignored. The new numeric rows are transcribed from the compiled summaries Drift0r provided, and the original documents are not redistributed. This entry was amended once before packaging, to fold in the same-day Round-2 review and its paired-review fixes.

Public release packaging: annotated git tag v0.4.0 created; GitHub Release with this summary published.

v0.3.0minor · published · 2026-08-07

evidence current through 2026-08-07

Plain-language landing rewrite plus a new public register of commonly suggested explanations and what was actually tested. A paired Codex/Grok pre-publication review then corrected several places where public copy read more strongly than its claim rows: the mast-cell entry shipped as a clean negative despite elevated tryptase and an incomplete mastocytosis exclusion, the autoimmune entry dropped its documented denominator, and reported history was rendered as documented fact on the landing page. The register now carries visible claim ids and passes the same publication gate as every other medical surface. Search indexing is recorded as enabled by owner decision. Not medical advice; clinician review still not performed.

Added evidence

• CLM-0078 — celiac panel (tTG/DGP/EMA) negative 2025-10-01 with total IgA sufficient
• CLM-0079 — thyroid panel with antibodies euthyroid across draws 2021–2025
• CLM-0080 — heavy metals panel below reporting cutoffs 2026-06-19
• CLM-0081 — Ehlers-Danlos reported by the patient as previously cleared (reported history)
• evidence/ruled_out.yaml — public register of twelve commonly suggested explanations, test name / date / plain result only

Changed interpretations

• Mast-cell entry no longer reads as a clean negative: tryptase was above range on both documented draws (CLM-0030), the record attributes that to patient-reported hereditary alpha-tryptasemia (CLM-0031), and the KIT assay's method and limit of detection are undocumented, so systemic mastocytosis is not fully excluded (CLM-0032, CLM-0065)
• Autoimmune entry restated to three documented CRP and three documented ESR draws with no value after 2025-08-05, and 'unrevealing on the panels run' rather than 'completely negative' (CLM-0044)
• Mold/CIRS marked as a patient-reported negative rather than an objective one (CLM-0052)
• Psychiatric entry no longer asserts the question closed: the evaluation history other than the MMPI-2-RF is patient-reported, the record supports reversal of the SSD label specifically, and treating the medical findings as primary is a research position not a clinical adjudication (CLM-0046, CLM-0047, CLM-0048)
• Thyroid entry bounded to 2021–2025; COVID entry restated from 'documented onset' to reported onset from a patient-compiled timeline (CLM-0051)
• Landing case-in-brief attributes the joint-pain narrative (CLM-0045) and the 2017 onset (CLM-0051) to the patient rather than presenting them as documented
• AI methods disclosure no longer claims exact-build completeness — model families are recorded, exact builds are not recoverable for the earliest rounds

Strengthened or weakened hypotheses

• H3 plain title/summary: the gap is missing bone-formation markers in the documented record, not 'bone tests never run' — DXA, CTX and serial 24h urine calcium are all documented, and CLM-0011 notes absence of documentation is not proof never drawn
• H2 plain title made interrogative: CLM-0050 is not_verified and calls a monogenic/constitutional cause only the best single-primary research alternative
• H4 plain title made interrogative: CLM-0043 is contested and puts Bartonella-spectrum disease at a low–medium residual contributor without commercial confirmation

Retired / superseded findings

• DEC-0023 — superseded → replaced by DEC-0037 — DEC-0023 deferred removal of noindex to a separate patient decision and DEC-0027 rejected enabling indexing at launch. The site was already shipping noindex: false with no decision recording it. DEC-0037 records the owner's indexing decision, and the open item that DEC-0023 wanted Drift0r's separate say-so on indexing specifically.

Corrections

• COR-0033 — Mast-cell ruled-out entry published as a clean negative (status: applied_to_public_draft) · case records
• COR-0034 — Autoimmune ruled-out entry dropped its documented denominator and as-of boundary (status: applied_to_public_draft) · case records
• COR-0035 — Mold/CIRS patient-reported result presented as objective (status: applied_to_public_draft) · case records
• COR-0036 — Psychiatric ruled-out entry asserted the question closed (status: applied_to_public_draft) · case records
• COR-0037 — Reported history rendered as documented fact on the landing page (status: applied_to_public_draft) · case records
• COR-0038 — Hypothesis plain titles overstated their claim rows (status: applied_to_public_draft) · case records
• COR-0039 — Ruled-out register bypassed the publication gate and the site's source-trail promise (status: applied_to_public_draft) · case records
• COR-0040 — Public documents asserted indexing was disabled while the site shipped indexable (status: applied_to_public_draft) · case records

Content landed on main as e278d54, which is what release.yaml records as base_commit. The v0.3.0 tag was created on db3b105, one commit later: db3b105 restores the full contribution link labels on the compact CTA variants, a presentation-only fix that carries no version bump. The deployed site is therefore not byte-identical to e278d54; content_version labels the analysis, not the build. Private run materials and the private tests ledger remain gitignored; the four new claim rows are sourced from it but publish test name, date, and plain result only.

Public release packaging: annotated git tag v0.3.0 created; GitHub Release with this summary published.

v0.2.0minor · published · 2026-08-06

evidence current through 2026-08-06

Material corrections from the 2026-08-06 multi-model adversarial audit (Claude/Grok/Codex). Fixes patient-agency reversal on nodule biopsy, bounds inflammatory-marker claims with denominators, restores specialty-LDT and negative-test hedges, expands treatment-response domains, retypes HαT as reported history without genotype, adds same-scanner L3–L4 DXA within-LSC claim, tightens public “is rejected” language, removes unsourced high-VUS-rate wording, renders literature titles on the clinician packet, and reclassifies lit-0206 as caution/counterweight on H5. Not medical advice; clinician review still not performed.

Added evidence

• CLM-0077 — same-scanner L3–L4 BMD series with LSC (bone density summary)

Changed interpretations

• CLM-0041 agency: patient was refused exploratory biopsy (not patient refusal)
• CLM-0044 CRP/ESR bounded to 3+3 documented draws as-of 2025-08-05
• CLM-0037/0038 LDT contested label + negative ≠ exclusion hedges
• CLM-0039 treatment response named domains (joints, night sweats, nightmares, REM); durability restated to the summaries' own wording (sustained ≥6 months; second course larger/more durable) — the prior blanket 'non-durable' was unsupported
• CLM-0043 same durability descriptor corrected to documented joint-pain relapse after the first course
• CLM-0031 kind reported_history; ranking no longer lists HαT genotype as confirmed
• Clinician packet literature lines show title/year (+ applicability notes when approved)
• Public language: bare 'is rejected' → not supported by the presently available record

Strengthened or weakened hypotheses

• H2/CQ-004: remove unsourced high VUS rate; keep modest yield + VUS interpretation caution
• H5: lit-0206 moved from supporting to contradicting (diagnostic-pitfalls review)

Retired / superseded findings

• COR-0009-vus-frequency-clause — superseded → replaced by COR-0030 — Only COR-0009’s VUS-frequency clause is superseded; yield language retained. COR-0009 register entry itself is not rewritten (append-only).

Corrections

• COR-0022 — Nodule biopsy agency reversed (patient was refused, did not refuse) (status: logged_in_inventory) · case records
• COR-0023 — CRP/ESR “repeatedly normal” lacked denominator and as-of date (status: logged_in_inventory) · case records
• COR-0024 — CLM-0037 missing specialty LDT contested / not-FDA-cleared label (status: logged_in_inventory) · case records
• COR-0025 — CLM-0037 / CLM-0038 missing negative ≠ impossible hedge (status: logged_in_inventory) · case records
• COR-0026 — CLM-0039 collapsed treatment-response domains (status: logged_in_inventory) · case records
• COR-0027 — CLM-0031 typed as observed_fact without genotype instrument (status: logged_in_inventory) · case records
• COR-0028 — Same-scanner L3–L4 DXA series missing from claim inventory (status: logged_in_inventory) · case records
• COR-0029 — Bare “is rejected” passed public-language gate (status: applied_to_public_draft) · case records
• COR-0030 — Unsourced “high VUS rate” / “VUS common” frequency claims removed (status: applied_to_public_draft) · case records
• COR-0031 — Printed clinician packet rendered bare literature IDs (status: applied_to_public_draft) · case records
• COR-0032 — lit-0206 miscategorized as support on H5 (status: applied_to_public_draft) · case records

Remediation landed on main and was deployed to the live site on 2026-08-06; the annotated tag and GitHub Release carry this summary. evidence_current_through advanced to 2026-08-06 for the adversarial audit and primary-source rechecks. Private run materials remain gitignored.

Public release packaging: annotated git tag v0.2.0 created; GitHub Release with this summary published.

v0.1.3patch · published · 2026-08-06

evidence current through 2026-08-05

Initial stable public release of the research evidence portfolio. This promotes the reviewed v0.1.3-rc.9 candidate without changing medical evidence, claims, literature polarity, working hypotheses, or confidence. Search indexing is enabled. The portfolio remains incomplete, AI-assisted, not medical advice, and not clinician-reviewed; licensed clinicians must verify all underlying records and findings.

Added evidence

None listed for this release.

Changed interpretations

• No analytical change from v0.1.3-rc.9; stable-release and indexing metadata only

Strengthened or weakened hypotheses

• No hypothesis strength change

Retired / superseded findings

None. All published IDs for this release remain active on the live site.

Corrections

No correction IDs attached to this release entry.

Promoted after the quiet custom-domain deployment and live HTTPS verification. Clinician review remains not performed, and publication permission does not imply endorsement.

Public release packaging: annotated git tag v0.1.3 created; GitHub Release with this summary published.

v0.1.3-rc.9patch · release candidate · 2026-08-05

evidence current through 2026-08-05

Checkpoint I.1.1 / J.1 credibility and launch-safety pass. Citation audit v1.2.1 source-grounds 32 of 34 public-used works against Europe PMC abstracts or open publication locators; 2 works (lit-0058, lit-0089) are bibliographic-only (no abstract or local full text — not counted as source-grounded; catalog-only comparisons removed; human review still pending; semantic_verified_count unchanged). Indexing polarity test uses isolated DRIFT0R_RELEASE_YAML fixture (never mutates live release.yaml). Contribution CTA links are ≥44px mobile tap targets. Print report validation rejects missing or empty content_version. Artifact-integrity tests strip ambient DRIFT0R_* so Pages CI env cannot poison negative gates. noindex remains true; clinician review not performed.

Added evidence

• LAUNCH_CITATION_FIDELITY_AUDIT v1.2.1 — 32 source-grounded + 2 bibliographic-only (lit-0058, lit-0089) of 34 public-used works
• DRIFT0R_RELEASE_YAML isolated fixture for noindex:false polarity build scan

Changed interpretations

• Contribution CTA primary links styled as full-width ≥44px rows on mobile
• validate_all fails closed on missing/empty print QA content_version
• lit-0058 and lit-0089 reclassified bibliographic_only (NO SOURCE TEXT) — not source_grounded_ai

Strengthened or weakened hypotheses

• No hypothesis strength change

Retired / superseded findings

None. All published IDs for this release remain active on the live site.

Corrections

No correction IDs attached to this release entry.

Checkpoint J.1 amends this entry for honest source-grounded counts and CI env isolation. Checkpoint J.2 / J.2.1 are presentation-only UX (record previews, homepage composition, mobile atlas carousel, compact provenance) — no medical content changes. Post-J.2.1 follow-up adds homepage Current working synthesis (H1–H3 from approved data) and decluttered desktop Evidence Atlas edges (legend + a11y + table carry verbs). Nothing pushed, deployed, tagged, or indexing-enabled.

Public release packaging: annotated git tag planned (v0.1.3-rc.9); GitHub Release with this summary planned. Not created — this version was not packaged as a public release.

v0.1.3-rc.8patch · release candidate · 2026-08-05

evidence current through 2026-08-05

Checkpoint I.1 focused remediation. Visitor-facing indexing copy derives from release.noindex (false-polarity publication build scan); source-grounded citation audit for public-used works with exact statements and source passages (human review still pending; semantic_verified_count unchanged); mobile landing chrome compacted so the project headline enters the first viewport; mobile QA extended with route markers, scroll samples, interactions, uncapped touch counts; print QA report version must match release.yaml. noindex remains true; clinician review not performed; permission is not endorsement.

Added evidence

• Source-grounded LAUNCH_CITATION_FIDELITY_AUDIT v1.1 (exact public statements + source passages)
• Indexing false-polarity publication build scan (noindex:false → no stuck disabled copy)
• Mobile QA route markers, scroll screenshots, home h1 first-viewport gate

Changed interpretations

• All visitor-facing indexing strings derive from release.noindex helpers (homepage, print meta, legal)
• Homepage hero reordered: headline before long caveat; layout disclaimer omitted on home for mobile fold
• Citation fidelity not claimed fully passed — AI source-grounded only; independent human review pending
• Print pipeline records content_version_matches_release; validate_all rejects stale print report version

Strengthened or weakened hypotheses

• No hypothesis strength change; citation audit honesty tightened (many prior templated supports → partial)

Retired / superseded findings

None. All published IDs for this release remain active on the live site.

Corrections

No correction IDs attached to this release entry.

Annotated Git tag and GitHub Release deferred until public launch. Nothing pushed, deployed, tagged, or indexing-enabled in this remediation pass. Physical device smoke remains owner-only after quiet deploy.

Public release packaging: annotated git tag planned (v0.1.3-rc.8); GitHub Release with this summary planned. Not created — this version was not packaged as a public release.

v0.1.3-rc.7patch · release candidate · 2026-08-05

evidence current through 2026-08-05

Checkpoint I pre-launch remediation. Custom-domain URL single-source (base /), indexing single-source from release.noindex with robots.txt allowing crawl, production OG image and summary_large_image social meta, creator/independence attribution, launch-cited literature fidelity audit (AI semantic only; semantic_verified_count unchanged), automated mobile QA, browser-print safety smokes, public concise roadmap export, community ops docs, private unpublish runbook and source-snapshot manifest, coverage no-write verification. noindex remains true; clinician review not performed; permission is not endorsement.

Added evidence

• LAUNCH_CITATION_FIDELITY_AUDIT for 44 launch-cited works (identity + AI semantic statuses)
• 1200×630 Contour OG/X image (generic framing only)
• Automated mobile viewport QA report and owner real-device checklist
• Public concise ROADMAP via export rewrite; CoC, SECURITY, moderation/FAQ/72h ops

Changed interpretations

• Indexing: release.yaml:noindex drives HTML robots meta and provenance; robots.txt permits crawl
• Canonical/OG/issue URLs derive from build-time DRIFT0R_SITE_URL / PUBLIC_REPO_URL (no hard-coded origin split)
• Print disposition renamed to PASS_WITH_TEXT_AND_AUTOMATED_LAYOUT; browser-block smokes for /case/ and prediction-matrix
• lit-0057 presentation: guideline silent on FISH LDTs; smear/PCR is published standard (project inference)
• Standard verification no longer rewrites coverage.yaml (UTC as_of); public export ships concise ROADMAP only

Strengthened or weakened hypotheses

• No hypothesis strength change from citation AI pass; lit-0001 retained as historical/retired framing only

Retired / superseded findings

None. All published IDs for this release remain active on the live site.

Corrections

No correction IDs attached to this release entry.

Annotated Git tag and GitHub Release deferred until public launch. Nothing pushed, deployed, tagged, or indexing-enabled in this remediation pass. Non-COR presentation fixes recorded under changed_interpretations: lit-0057 FISH language (guideline silent on FISH LDTs; smear/PCR is the published standard — project inference, not endorsement); coverage verification no longer rewrites coverage.yaml as_of on UTC day boundary; public export no longer ships the private operational ROADMAP body.

Public release packaging: annotated git tag planned (v0.1.3-rc.7); GitHub Release with this summary planned. Not created — this version was not packaged as a public release.

v0.1.3-rc.6minor · release candidate · 2026-08-05

evidence current through 2026-08-05

Public research-preview candidate (Checkpoint H.1 series). Contour hybrid site with audited claims, working hypotheses, clinician questions, literature cards, and print routes. Patient publication permission obtained; clinician review not performed. noindex remains enabled. Analysis versioning and structured changelog introduced so future material changes are traceable without a historical-site archive.

Added evidence

• Public claim inventory ship-set (CLM-*) with source trails and evidence-type registers
• Working hypotheses H1–H5 and H-NULL with support/contradiction structure
• Clinician questions CQ-001–CQ-010 and unresolved record questions UQ-0001–UQ-0017
• Literature cards and identity attestation coverage for launch-cited works

Changed interpretations

• Public language gate and specialty LDT dual-channel framing on medical surfaces
• Primary-record honesty: no original lab/DXA/clinic notes reviewed by this project
• Browser-print policy: only designated print routes emit multi-page medical content

Strengthened or weakened hypotheses

• Round-1 multi-model synthesis recorded as research ranking only (not clinical diagnosis)
• H-NULL retained as qualitative base-rate honesty baseline, not a disease competitor

Retired / superseded findings

None. All published IDs for this release remain active on the live site.

Corrections

• COR-0001 — Hip BMD change vs LSC: source-attributed, not free-floating “significant” (status: logged_in_inventory) · case records
• COR-0002 — “Osteoporosis” is WHO-by-T-score language in compiled summary, not a verified clinician diagnosis string (status: logged_in_inventory) · case records
• COR-0003 — “Catastrophic osteoporosis” internal research language (status: applied_to_public_draft) · case records
• COR-0004 — Historical T ~34 ng/dL provenance is incomplete (status: logged_in_inventory) · case records
• COR-0005 — KIT “negative” lacks method and LOD (status: logged_in_inventory) · case records
• COR-0006 — CTX 616 lacks fasting / time-of-day documentation (status: logged_in_inventory) · case records
• COR-0007 — Specialty Babesia/Bartonella must not read as confirmed infection (status: logged_in_inventory) · case records
• COR-0008 — SSD reversal scope is documented in mental-health summary (status: logged_in_inventory) · case records
• COR-0009 — Gene-panel yield honesty (~9%) must not become “high yield” (status: logged_in_inventory) · case records
• COR-0010 — Catalog duplicate works inflate public counts (status: identified) · case records
• COR-0011 — Cross-scanner DXA absolute BMD not comparable (status: logged_in_inventory) · case records
• COR-0012 — Dry beriberi label is interpretive around documented deficiency + phenotype (status: logged_in_inventory) · case records
• COR-0013 — Machine-readable compound probability vocabulary (status: deferred) · case records
• COR-0014 — Public medical page generation must not ship raw overstatement surfaces (status: logged_in_inventory) · case records
• COR-0015 — ISCD Official Positions near-duplicate cards (status: logged_in_inventory) · case records
• COR-0016 — Structured hypothesis and clinician-question instances required (status: applied_to_public_draft) · case records
• COR-0017 — Nine literature cards had wrong or dead DOI/PMID identifiers (status: applied_to_public_draft) · case records
• COR-0018 — lit-0294 inclusion/exclusion and gene-panel yield figures (status: applied_to_public_draft) · case records
• COR-0019 — Source class is independent of verification_status; no primary instrument records in repo (status: applied_to_public_draft) · case records
• COR-0020 — Defect-card # Links sections still carried old identifiers after frontmatter fix (status: applied_to_public_draft) · case records
• COR-0021 — lit-0104 applicability: uremia review must not support non-CKD patient cytokine inference (status: applied_to_public_draft) · case records

Annotated Git tag and GitHub Release with this summary are required at public launch; not created in private monorepo remediation passes.

Public release packaging: annotated git tag planned (v0.1.3-rc.6); GitHub Release with this summary planned. Not created — this version was not packaged as a public release.

v0.0.0-draftmajor · draft · 2026-08-05

evidence current through 2026-08-05

M0–M3 audits, schemas, validators, and private monorepo scaffolding. No public medical site pages yet.

Added evidence

• Claim inventory, literature catalog, and research schemas established

Changed interpretations

None listed for this release.

Strengthened or weakened hypotheses

None listed for this release.

Retired / superseded findings

None. All published IDs for this release remain active on the live site.

Corrections

No correction IDs attached to this release entry.

Pre-site draft baseline.