Changelog

Version 0.4.0 · Preview

1. 25 October 2026 · v0.4.0

   Exact-only confidence, range links, comparisons and constructs

   AlphaFold pLDDT is shown on this UniProt sequence only through a unique exact residue partner in the model; substitutions and ambiguous positions stay empty, and a range reads its own count of scored residues. Recorded predicted-model images use capture procedure 3.0.0. A link that carries a residue range selects it when the whole range lies in the entry, and says why when it does not; nothing is clipped. Compare with another entry aligns the sequence on screen and fetches the other entry afresh, and Try again always refetches. The construct cropper takes the current selection when asked, keeps typed start and end values until they are valid, and labels a C-terminal tag after the protein name. An uploaded PAE file is bounded before it is read and is cleared when the protein changes.

2. 18 October 2026 · v0.3.0

   Retained structure images and source receipts

   Record a chosen structure view against a saved task, with its PNG, the hash of the structure bytes actually loaded, selection and rendering details, and your explanation. Reopen the retained image and export its exact PNG and receipt without refreshing the source. Coordinates are not retained and the record does not provide interactive replay.

3. 7 October 2026 · v0.2.0

   Conservative cross-release residue placement

   AlphaFold residue tracks are withheld for missing sequence, ambiguous repeated fragments or a competing gapped alignment. Equal length alone no longer establishes residue correspondence. The per-residue AlphaMissense mean is explicitly an app-derived aggregation of individual-substitution predictions.

4. 6 October 2026

   Pins, studies and a page for each target

   A protein you look up is now remembered in this browser as a target, with UniProt's facts and its canonical sequence. Under the view header you can pin it, add it to a study, or open its own page, which shows the cached facts with their source and date, your notes, numbers, tags and files, the studies it is in, and this app's dossier for it. What you keep stays in this browser.

5. 5 October 2026

   Two views with a toggle, exports and faster variant loading

   The main screen is now a dossier (identity, headline facts and the structure first, a map of the whole sequence, evidence cards for structures, drugs and diseases, then the function text) or a workbench (the experimental entry and the AlphaFold model side by side over the sequence tracks), switched from the view header or with the v key; the choice is remembered on the device and carried in a copied link. The Export menu offers the PDB entries, drugs, and features and variants as CSV, the sequence as FASTA, the map and tracks as SVG or PNG and each 3D view as PNG, each with its provenance. The full variant list and the AlphaMissense scores are now read in a Web Worker, so a large download no longer freezes the page.

6. 1 October 2026

   Released on the mbio.tech Tools page

   The protein explorer went live: UniProt search, the default PDB entry beside the AlphaFold model with SIFTS residue mapping between them, sequence tracks with variants and AlphaMissense, and drugs from Open Targets, all fetched live in the browser.

No release summary yet: the app is in preview, and its first suite release will publish one here.

Version 0.2.0 · Preview

1. 24 October 2026 · v0.2.0

   Exports with their scope stated, cited FDA precedent, and dates labelled by source

   Trial exports say whether they hold the full loaded result or the current list across every page, and carry download limits, interrupted downloads and snapshot qualifications. The US products CSV keeps each date's source: original approval, NDC marketing start or unknown, and whether details failed to load. FDA precedent exports its filtered result as a cited CSV in a ZIP, with a citation on every row and the reasons evidence is missing. A joined product without a usable review interval now says so rather than reading as absent from FDA's compilation. On screen, an NDC marketing start date is labelled as such and is no longer shown as an approval on the Board, the timeline or FDA precedent. Saved research views keep the applied search, display settings and note, and reopen the search live.

2. 5 October 2026

   FDA precedent replaces historical odds

   The Historical odds view, the odds on the Board's connectors and the BIO, Informa and QLS Advisors figures behind them are removed. FDA precedent takes their place: pathway, designations, review time and published complete response letters for the condition's new molecular entities and new biologics, each cited to the FDA record. The connectors now count combined-phase trials.

3. 1 October 2026

   Released on the mbio.tech Tools page

   The drug pipeline explorer went live: live ClinicalTrials.gov searches with six charts, US approvals from openFDA, historical odds from the BIO, Informa and QLS Advisors report, and dated snapshots of seven example searches (ClinicalTrials.gov data of 25 September 2026).

No release summary yet: the app is in preview, and its first suite release will publish one here.

Version 0.2.0 · Alpha

1. 18 October 2026 · v0.2.0

   Captured FDA rows and real examples; review pending

   The pinned FDA capture replaces the invented store. Real product examples, source member checksums, explicit unknown dates and retained unresolved Paragraph IV rows accompany mechanically transcribed checks. Purple Book supplied text and reference-product expiry are preserved. The app remains alpha; independent regulatory review and the original incomplete fixture requirements remain open.

2. 13 October 2026 · v0.1.0

   First build on a synthetic example store

   The lookup, the rule-traced range of listed protection, the patents, exclusivities, generics and Paragraph IV views, the date window, the change log and cited exports, on an invented example store in the FDA file layouts while the daily capture starts; the first capture replaces it and the example label goes.

No release summary yet: the app is in preview, and its first suite release will publish one here.

Version 0.8.0 · Preview

1. 20 October 2026 · v0.8.0

   Published checks for pairwise DNA alignment, ahead of the aligner

   A published, hashed fixture set for the pairwise DNA aligner, written before the aligner ships, from an independent derivation on Biopython 1.88: 57 cases in three named modes (global, with every end gap penalised; semiglobal-read, with end gaps free on the read only; local), match +2, mismatch −3, a gap of k bases −(5 + 2(k − 1)) and ambiguity codes scored 0 and left out of identity. Every case records the score, the number of equally scoring alignments, which reference and read bases are paired in all of them, which differences are (and, for one that is not, the nearest bases fixed on either side), and the groups of equally scoring alignments with each group's extent; the alignment shown is checked against Biopython's equally scoring alignments sorted by the published tie order. Among the cases: tie-dense runs and repeats, ambiguity codes on both strands, reads across a circle's origin (one of 1,000 bases on a 6,646-base plasmid) and past one turn, reads running past a linear reference's ends, a read of N only, distant and banded placements, a circle where only some of 2,048 equally scoring alignments run past one turn, a read that scores the same on both strands over the same bases, chosen regions of a reference (one across a circle's origin), a read where no letter pairs, insertions before a reference's first base and at a circle's origin, and twelve refusals. The aligner itself is not in this version.

2. 17 October 2026 · v0.7.0

   Primers for one change in a template

   A new task designs site-directed mutagenesis primers in two schemes: overlapping complementary primers carrying the change, and back-to-back primers on a circular template whose product is linear and needs circularisation. Each primer carries its melting temperature against the mutant it matches exactly and its matched 3′ segment’s against the template; a melting temperature on a template the primer does not fully match is not given. The counts of candidates rejected under each rule are shown, and no rule is relaxed to produce a result. Published checks cover both schemes against an independent derivation.

3. 11 October 2026 · v0.6.0

   Primer receipts travel with their sequence

   Sending DNA to Plasmid carries retained primer sets and predicted PCR inputs in the browser record. Recorded temperatures retain their qualifications; inconsistent retained ionic conditions refuse. Published checks cover effective salt selection and historical records.

4. 11 October 2026 · v0.5.0

   Published checks for primer design

   The bounded primer design now has a published, hashed fixture set from an independent derivation: nine design cases (a hand-worked pair with tails, the lac operon's first 600 bases under the panel's constraints, with excluded regions, with 5′ tails, under Mg²⁺ with the Owczarzy 2008 correction, with a 200-base flank that caps the search, pUC19 with the target across the origin, and two searches that return no pair without relaxing a constraint) and four refusals, every melting temperature from Biopython, every window, count and ranking derived in the script, and exact binding by the exact-pair derivation. A separate, widely used primer-design program was run offline on the real templates as a recorded comparison, not as the oracle.

5. 10 October 2026 · v0.4.0

   Published checks for the design revision review

   Reviewing changes between two named revisions now has a methods entry and a published, hashed fixture set from an independent derivation: ten revision transforms, among them an origin rotation and a reverse complement that change coordinates without inventing mutations, a joined CDS whose edit is flagged and not transferred, a repeated sequence whose correspondence is qualified, a note-only change that makes no residue edit, a non-exact circular change withheld as data rather than resolved at an arbitrary origin, and a topology change recorded as data. The review records both sequence hashes and the method, and the edit list is a minimum-length script checked against an independent edit distance.

6. 10 October 2026 · v0.3.0

   Published checks for the PCR product and the fragment search

   The predicted PCR product and the exact fragment search over saved sequences now each have a methods entry and a published, hashed fixture set from an independent derivation: 30 PCR cases, among them origin-spanning, palindromic, tailed and overlapping primers and a refusal at every limit, and 26 fragment cases, among them palindromes reported once, overlapping hits, ambiguity codes as literal barriers and hits across the origin. Both agree exactly.

7. 9 October 2026 · v0.2.0

   Find and review a UniProt sequence

   The workspace can search UniProt by gene, protein name or accession. Choose a result, review the organism and residue count, then explicitly use the full canonical sequence. Existing input requires confirmation and saved-work safeguards. Accession, organism, release and retrieval date stay in the imported record and saved inputs. Pasted input stays local; only submitted search text and selected accessions reach UniProt.

8. 7 October 2026 · v0.2.0

   Explicit thermodynamic and genetic-code refusals

   Tm and secondary-structure calculations refuse nonphysical or unsupported ionic conditions. NCBI codes 27, 28 and 31 require biological context for their dual-use stop codons; stop-terminated translation and ORF detection are refused for these codes. Special hairpin-loop limitations remain explicit.

9. 6 October 2026 · v0.1.0

   Eleven operation pages, the workspace, hand-offs and the first verification sets

   The first release: reverse complement, translation in any frame with the 27 NCBI genetic codes, six-frame translation, open reading frames, GC percent, DNA and RNA molecular weight, motif search with IUPAC codes, protein molecular weight, isoelectric point and ε280 with tags and protease sites found and the numbers given before and after cleavage, oligo Tm by nearest neighbour with the Owczarzy 2004 and 2008 salt corrections under both concentration models, primer pair ΔTm and annealing temperature, hairpin and dimer ΔG at 37 °C from the SantaLucia and Hicks 2004 rules, and binding sites on a template. Raw, FASTA, multi-FASTA and GenBank pastes; inputs over 100 kb in a Web Worker to 1 Mb. Eleven answer-first operation pages and three versioned example pages (insulin, GFP, lacZα). Sequences and constructs kept as objects with Pin, Send to and Add to study; a protein's sequence goes to the calculators, a short peptide to the molecule app as SMILES, an accession to the protein app. Every number that depends on a model or a parameter table carries its method, and the translations, ORFs, protein numbers, melting temperatures, hairpin and dimer ΔG, DNA weights and peptide SMILES are checked against named external oracles on published fixtures.

No release summary yet: the app is in preview, and its first suite release will publish one here.

Version 0.15.0 · Preview

1. 17 October 2026 · v0.14.0

   A larger annotation library

   The annotation library grows from 50 to 90 parts, each taken from a feature its submitter annotated in an NCBI record (library 2). Where two parts reach the thresholds at one place, the map shows the best-scoring call and names the other beside it. Where a pair is marked as different elements, both are shown, the smaller drawn inside the larger. Maps that pin library 1 reopen with library 1. The limits now count the enzymes with no McClelland et al. row.

2. 17 October 2026 · v0.13.0

   Gel bands grouped by the rule the gel states

   A fragment now joins a band on the schematic gel only when it is within 5 percent of that band's largest fragment, so a band's fragments are within 5 percent of each other, as the gel says; before, a run of near sizes could be drawn as one band wider than that (1,000, 960 and 922 bp now draw as two bands). 5 percent is a judgement. The reference was changed first, with eleven grouping cases worked by hand; the correct, empty and reversed trio was judged again: 43 digests' bands moved and no picker verdict changed. The earlier trio file stays published beside the new one.

3. 11 October 2026 · v0.12.0

   Primer annotations from Sequence

   Sequence hand-offs admit predicted PCR products against their exact inputs and map full incorporated primers, including tails. Retained primer sets map only when their template hash, length and topology match; unmatched pairs remain in the source receipt with an omission warning. Supplied annotations, provenance and GenBank exports remain available.

4. 11 October 2026 · v0.11.0

   Assembly on your own sequences

   Restriction cloning, Golden Gate and Gibson design products from your inputs in the Assembly section. Circular strand cuts retain their signed separation across the origin; overlapping cleavage intervals refuse. Records retain original inputs and hashes, method versions, alternatives, junctions and qualifications. Accounting is checked against the recorded inputs and junctions. Product FASTA, GenBank and editable assembly inputs identify designed products, never evidence of a reaction.

5. 11 October 2026 · v0.9.0

   Circular assembly boundary references

   New independent references cover origin-spanning strand cuts, rotations, reverse complements and Golden Gate destinations. A second record reference preserves method-specific Gibson accounting and refuses the historical malformed accounting block. These fixtures precede the engine repair and own-sequence Assembly view.

6. 11 October 2026 · v0.8.0

   Designed-assembly record reference

   Eight independently written record cases cover retaining an assembly design, dropping stale product or topology claims, refusing malformed inputs, and reading existing constructs. The reference is published before the assembly workspace reader.

7. 10 October 2026 · v0.7.0

   Cloning corrections: a homology floor, both strands, either orientation

   Corrections to the cloning line before it merges. Gibson-style assembly now depends on a declared minimum terminal homology of 15 bp (Gibson et al. 2009; the commonly used working range is roughly 15 to 40 bases) and withholds the product below it, with that floor stated as a limit of the method and never as a recommended length; repeated homology is counted on both strands, so an overlap whose reverse complement sits elsewhere in the fragments withholds the product instead of showing an assembly that could invert. Golden Gate matches each part in either orientation, since a part and its reverse complement are one molecule, so an assembly no longer depends on which strand a part was supplied on; the double-stranded end primitive behind it keeps each end’s own overhang bases when a fragment is written out from the other strand. Restriction cloning states that the products shown are the one-insert circles whose ends anneal, vector self-ligation, multi-insert circles and concatemers not being enumerated, and that the worked example takes the longest enzyme-ended fragment on each side. The guide finder’s oracle is rewritten to enumerate independently, by one regular-expression search over the sequence and over the Biopython reverse complement with positions mapped back, with a case worked by hand base by base and a write-refusal guard; the guide table’s count is labelled “Other protospacer matches”, since it counts protospacer matches and requires no PAM. Fixtures republished with the new and corrected cases.

8. 10 October 2026 · v0.6.0

   SpCas9 guide sites, no scores

   Find every SpCas9 guide, a 20 base protospacer immediately 5′ of an NGG PAM, on both strands, with a PAM across a circular origin found once. Each candidate carries its coordinates and strand, the protospacer, the PAM, the GC fraction, the longest homopolymer run and the exact count of other occurrences of the protospacer within the sequence you pasted, on both strands. That count is of exact matches inside your sequence: it is not an off-target assessment and says nothing about the genome. No on-target or off-target score of any kind is given, no mismatch-tolerant search is run, and no genome-wide or specificity statement is made. SpCas9, NGG, 20 bases only. Checked against an independent enumeration published on the verification page.

9. 10 October 2026 · v0.5.0

   Gibson-style assembly, checked

   Join a user-ordered plan of 2 to 6 fragments on exact terminal overlaps into a circular or linear product, within a 50,000 bp ceiling. The overlap between two consecutive fragments is their exact shared terminal sequence, found base for base and collapsed once; this is sequence planning on double-stranded homology, so no overlap length is recommended or called optimal and no melting temperature is used, and fragments are taken as supplied. A missing or contradictory overlap, or an overlap whose sequence occurs more than at its two designed ends (repeated internal homology that would permit an alternative assembly), withholds the product rather than resolving it silently. Material is conserved, each overlap counted once. Checked against hand-designed assemblies, the product re-derived independently and reverse-complement and rotation giving the same product, published on the verification page.

10. 10 October 2026 · v0.4.0

    Golden Gate assembly, checked (BsaI, BsmBI)

    Assemble parts into a destination with one Type IIS enzyme, BsaI or BsmBI. The enzyme cuts outside its recognition sequence, so each part is released with defined overhangs and the recognition sites leave on the discarded flanks; the parts and the opened backbone join by their overhangs into one scarless circle, decided on the overhangs alone. A palindromic overhang, a duplicate or reverse-complement overhang conflict, an unresolved internal site or a cut that falls off a linear end is refused, not a warning to click past; no part is domesticated and no coding sequence changed, and no fidelity figure is given. Checked against hand-designed assemblies whose released overhangs and scarless products Biopython confirms, published on the verification page.

11. 10 October 2026 · v0.3.0

    Restriction cloning, checked

    Cut a vector and an insert with chosen restriction enzymes and ligate the recombinant circle. Two ends join only when their exposed overhangs and polarity anneal, decided on the bases alone and never on the enzyme name or a coordinate; a blunt or symmetric join that could go either way is shown as both products, not resolved silently; an end that depends on an unrecorded methylation state is qualified, never shown as a clean product; a cut that falls off a linear end refuses. Material is conserved, each shared overhang counted once. Nothing is repaired, filled in or mutated, and a designed product is not evidence that the reaction occurred. Checked against hand-worked clones round-tripped by Biopython, published on the verification page.

12. 10 October 2026 · v0.2.0

    A typed double-stranded end, checked

    The engine behind restriction cloning now represents a cut end by its exposed overhang, its polarity (5′, 3′ or blunt) and a methylation qualification, and decides whether two ends anneal on the exposed bases and polarity alone, never on the enzyme name or a coordinate. It is checked against hand-worked cases, each written out base by base, published on the verification page. No end is repaired, filled in or mutated, and a designed join is not evidence that a reaction occurred.

13. 6 October 2026 · v0.1.0

    Plasmid maps with library features, enzyme sites, methylation, primer sites and a diagnostic digest picker, in preview

    The first release, in preview: pasted sequences, FASTA and GenBank records, SnapGene .dna files read-only, the pUC19 and pBR322 examples and NCBI records fetched when you ask, up to 50,000 bp; circular and linear maps drawn as SVG and operable by keyboard; features from a 50-part annotation library built from submitter-annotated NCBI records, each with its identity, coverage, mismatches and gaps; the sites of 223 enzymes with dam and dcm methylation from McClelland et al. 1994, Table I, where a site of unknown sensitivity is never counted as a cut; six sequencing primers named by their makers' manuals; a digest picker comparing correct, empty and reversed constructs with a schematic gel; the plasmid's own weight with ng, pmol, copies and ligation amounts; explainers linked to the cell tour; exports as CSV, SVG, PNG, print, GenBank and FASTA. Checked against Biopython 1.88 on pUC19 and pBR322 and on 26 feature alignments, with the fixtures published on the verification page.

No release summary yet: the app is in preview, and its first suite release will publish one here.