The recommendation engineMy Comics Collection (MCC)2026 combines four families of signals to suggest comics suitable for each collector:inventory collection(series owned, publishers, eras, creators),reading history(issues read, completion of arcs, series skips),the wishlist(purchasing intentions, budget priorities), andexplicit ratings(notes 1 to 5 on the issues read). The algorithm calculates a vector similarity between your profile and around 220,000 referenced comics (Marvel, DC, Image, Dark Horse, IDW, French indie Panini, Urban, Glénat, Bliss), then weights by four use cases:next run to read,sleeper to speculate,completing a run in progress,indie discovery. Cold start: 5 ratings are enough to calibrate the taste. VF budget and availability filters applied on output.
The MCC comics recommendation engine version 2026 answers a simple but structurally difficult question:“What should I read or buy next?”". This question seems trivial for a beginner who follows two or three series, but becomes critical for an advanced collector who follows 40 current series, has 1,000 to 5,000 issues, and must decide between around twenty editorial options each month. The MCC engine tackles this complexity by crossing four families of signals (inventory, reading, wishlist, ratings) with a similarity graph between comics which contains approximately 220,000 nodes linked by 3.4 million edges (similar series, shared creators, crossed eras, narrative tags). The result is a personalized list, classified by relevance score, filterable by monthly budget and by French availability (Panini Comics, Urban Comics, Glénat, Bliss Editions).
This guide dismantles the internal workings of the MCC 2026 engine, from thecold start(first user rating) to stable convergence (after 50 ratings), including the weighting of the four use cases and the two essential operational filters. Each section illustrates the mechanism with a concrete practical case: beginner without a collection starting from scratch, and experienced collector with 1,000 comics. The objective is twofold: to allow the collector to understandWhysuch recommendation appears in his feed, and give him the levers to manually adjust the engine (voluntary rating, wishlist update, serial blocking) in order to direct the suggestions towards his real priorities.
The MCC recommendation engine in 2026: architecture and signals
The MCC 2026 engine is structured around avector user profileupdated with each interaction. This vector of dimension 384 (compressed from 1,024 by supervised autoencoder) summarizes the taste of each collector on the editorial, narrative, aesthetic and temporal axes. Four signal families feed this vector continuously.
Signal collection: physical and digital inventory.Each issue added to your MCC collection (manual entry, CSV import, barcode scan) feeds the profile vector by weighting the corresponding axes: publisher (Marvel, DC, Image, Dark Horse, IDW, Boom, Vault), era (Golden, Silver, Bronze, Copper, Modern, Contemporary), creator (separate screenwriter and artist), genre (superhero, horror, sci-fi, crime, slice of life), format (single issue, TPB, hardcover, omnibus). A collection of 100 Spider-Man Marvel Bronze Age produces a very different vector than a collection of 100 Image Comics Indie 2020-2026. The collection signal accounts for 35% of the final score in steady state. SeeMarvel comics for startersto understand Marvel input logic.
Reading signal: what have you actually read?MCC separates inventory (issues owned) from reading (issues actually read). A collector can own an Amazing Spider-Man Bronze Age long-box without having read all 200 issues. The engine therefore tracks the status separatelyluvia explicit marking (MCC mobile and desktop check box) or via digital reading session (Marvel Unlimited, DC Universe Infinite, ComiXology integration if the user connects their accounts). The reading signal weighs 25% of the final score and makes it possible in particular to distinguish the profiledrive(who reads everything he buys) from the profilepure collector(who owns without reading). This distinction radically changes the recommendations: a reader primarily sees narrative suggestions (sequences of arcs, next runs to read), a collector primarily sees suggestions for completion or speculation.
Wishlist signal: declared purchasing intentions.The MCC wishlist contains the issues that the user plans to acquire in the short or medium term. It is segmented by priority (high, medium, low) and by target budget (maximum acceptable price). The wishlist signal weighs 20% of the final score and particularly influences use casesnext run to readetcompletion run. If you have 12 issues from Uncannymajor 80s narrative epic. SeeDC comics for startersto understand the wishlist logic by publisher.
Signal rating: explicit notes on the issues read.Each issue read can be rated from 1 to 5 stars in MCC. This signal is the most powerful of the engine (40% of the final score during the cold start phase, falling to 20% in stabilized mode), because it expresses an explicit qualitative judgment, where collection, reading and wishlist express ambiguous behaviors (a comic possessed is not necessarily loved, a comic on a wishlist is not necessarily purchased). User ratings also feed the MCC community base, which aggregates around 12 million individual ratings at the end of May 2026, and makes it possible to calculate a weighted average by series and by outcome. The engine crosses individual rating and community rating to estimate the probability that you will like a given comic.
These four signals converge into a unique profile vector, recalculated at each interaction (collection addition, reading marking, wishlist addition, submission rating). The recalculation takes place in less than 200 ms on the server side, which allows an almost instantaneous update of the displayed recommendations. The profile is stored encrypted (AES-256) on the MCC backend side and is never shared with third parties, in accordance with European GDPR policy. For an exploration of key series to integrate into a collection or wishlist, seecomics Image to start.
Similarity algorithm: series vectors, creators, eras
The algorithmic heart of the MCC 2026 engine is based on acomics similarity graphwhich contains approximately 220,000 nodes (single issues, series, creators, editorial events) connected by 3.4 million weighted edges. Each comic is represented by a vector of embeddings of dimension 384, calculated by a multi-task model trained on four crossed signals.
Embedding series.Two series are considered close if they share creators (writer, designer, inker, colorist, letterer), eras (temporal overlap), editors, and narrative universes (Marvel 616, DC New Earth, Image independent universes). Amazing Spider-Man Bronze Age 1970-1980 and Daredevil Bronze Age 1979-1986 have a similarity of 0.84 on the series vector, because they share the publisher Marvel, the Bronze era, the partial artist Frank Miller (Daredevil) and Gil Kane (Spider-Man), and the universe 616. Conversely, Amazing Spider-Man Bronze and Sandman Vertigo 1989-1996 have a similarity of 0.17: different publishers (Marvel vs DC/Vertigo), disjointed eras (Bronze vs Copper), foreign narrative universes (superheroes vs dark fantasy).
Embedding creators.The engine tracks 14,600 referenced comic creators (writers, artists, inkers, colorists, letterers, editors). Each creator is himself a vector of dimension 256, aggregated from his published works. This structure allows a recommendationby author: if you rate three issues by Jonathan Hickman (Secret Wars, House of X, Ultimate Spider-Man) 5/5, the engine pushes Avengers (Hickman 2012), New Avengers (Hickman 2013), and Powers of Creator granularity is essential for profilesauthorwho follow individual screenwriters across publishers.
Embedding eras.The engine divides comics history into seven formal eras: Platinum (1897-1937), Golden (1938-1955), Atom (1956-1969 transitional), Silver (1956-1970), Bronze (1970-1985), Copper (1984-1991), Modern (1992-2011), Contemporary (2012-2026). Each comic is anchored to a primary era by publication date, allowing recommendations to be filtered by time preference. A pure Silver Age collector will not see Contemporary suggestions unless their profile contains at least 8% post-2012 comics. The threshold is manually adjustable in the MCC settings.
Embedding narrative tags.An additional layer traces 84 narrative tags (street level, cosmic, time travel, multiverse, body horror, courtroom drama, heist, post-apocalyptic, coming of age, etc.) automatically extracted by an NLP model trained on official Marvel, DC, Image synopses, and on community reviews (Comic Book Roundtable, ICv2, MCC community). These tags allow cross-referencing of recommendations: a fan of Daredevil Born Again (street level + courtroom drama) sees Punisher MAX (Garth Ennis, street level), Gotham Central (Brubaker, courtroom drama), and Powers (Bendis, street level + courtroom drama).
The final similarity between two comics is a weighted linear combination of the four embeddings: 35% series, 30% creators, 15% eras, 20% narrative tags. This combination is recalibrated every six weeks based on aggregated user ratings. Engine performance is measured by arecall@10(probability that a comic rated 5/5 by the user appears in the top 10 recommendations): this score reaches 71% in May 2026, compared to 58% in January 2025, thanks to the enlargement of the graph and the refinement of creative embeddings. SeeMarvel vs DC vs Collectible Imagefor cross-editor logic.
Taste calibration: cold start in 5 ratings
Lecold startis the most delicate phase of any recommendation engine: how to offer relevant suggestions to a user about whom the system knows nothing? The MCC 2026 engine solves this problem with a three-step hybrid method that converges into just 5 user ratings.
Step 1: initialization by editorial selection.At first launch, MCC offers the user a grid of 16 emblematic covers covering a wide spectrum: Amazing Fantasy #15 (Silver Spider-Man), Action Comics #1 (Golden Superman), Watchmen #1 (Copper Moore), Sandman #1 (Modern Gaiman), Saga #1 (Contemporary Vaughan), Ultimate Spider-Man #1 (Contemporary Hickman), Walking Dead #1 (Modern Kirkman), Bone #1 (Modern Smith), East of West #1 (Contemporary Hickman), Sex Criminals #1 (Modern Fraction), Paper Girls #1 (Contemporary Vaughan), Daredevil #181 (Bronze Miller), Crisis on Infinite Earths #7 (Copper Wolfman), Maus (Spiegelman), V for Vendetta #1 (Copper Moore), Y: The Last Man #1 (Modern Vaughan). The user selects those that attract him visually. This first selection gives a raw initial vector with a precision of 31% on recall@10.
Step 2: first targeted rating.The engine immediately prompts the user to note5 specific outcomeson a grid extended to the 80 flagship series of the base. The 5 proposed outcomes are chosen to maximize thegain information: they cover the editorial axes (Marvel, DC, Image, indie), the eras (Bronze, Modern, Contemporary), the tones (superheroes, serious indie, horror, slice of life). Typically: Amazing Spider-Man #129 (1st Punisher), Watchmen #1, Saga #1, Walking Dead #1, Ultimate Spider-Man #1. Each rating out of 1-5 reduces the uncertainty of the profile vector. After 5 ratings, the recall@10 rises to 52%.
Step 3: iterative refinement.The engine then produces 10 initial recommendations, 3 of which are explicitly markedexploration(average similarity, chosen to test uncertain areas of the profile). As the user adds comics to the collection, marks readings and submits ratings, the profile is refined. The recall@10 reaches 65% after 25 ratings, 71% after 50 ratings, and caps at 73-74% beyond that. The MCC 2026 engine is therefore operational from the 5th rating, with maximum quality reached around the 50th.
For the beginner who starts with 0 collections, the recommended method is: initial visual selection of 16 covers, then rating of the 5 proposed outcomes, then acceptance of the 3 recommendationsexplorationto discover unexpected series. This routine takes around 8 minutes and provides access to a useful stream of recommendations as soon as the registration session ends. For an in-depth test, seebeginner collector: pillar guideand use the toolfind your comic.
Reco by use case: next run, sleeper, completion, indie discovery
The MCC 2026 engine does not produce a single generic list: it segments its recommendations intofour use casesdistinct, presented as separate tabs in the interface. Each use case applies a different score calculation, with specific weightings on the four families of signals.
Use case 1: next run to read.For a user in reading mode, MCC offers the nextruncoherent narrative to begin after recent readings. The score favors the reading signal (40%) and the rating (30%): if you rated Daredevil Born Again by Frank Miller 5/5, the engine pushes Sin City (Miller) if you also rate DKR, or Punisher MAX (Garth Ennis) if you also rate Preacher. The proposed runs typically have 12 to 60 outcomes, forming a complete narrative arc. The filter automatically excludes current series (open-ended) unless the user has explicitly activated the modecurrent series accepted.
Use case 2: sleeper to speculate.For a user in speculation mode, MCC offers underpriced comics with high potential for added value. The score combines collection signal (40%, to adapt to the actual budget), wishlist (20%, to align with intentions) and an external non-user signal: themarket trendcalculated over the last 90 days by crossing eBay, MyComicShop, Heritage Auctions, GoCollect Comics. A typical sleeper has a stable rating of €15-40 over 12 months, a sales volume of at least 8 transactions/month, and a potential catalyst (film/series announcement, renowned screenwriter's run, release of 1st appearance of a secondary character). Seecomics spec 2026: key issues to be assembledfor the complete sleeper grid.
Use case 3: run completion.For a user who started a run without finishing it, MCC offers the missing issues to complete the collection. The score favors the collection signal (60%) and the wishlist (25%). The motor detectsholesin partially collected runs: if you own Amazing Spider-Man #121-126 and #129-135, the engine pushes issues #127, #128, #136 which complete the arc to the top of the list. The functionality extends to variants: if you collect Ultimate Spider-Man 2024 Cover B variants, MCC pushes the missing Cover Bs first. This mechanic is the most popular usage among advanced collectors (1000+ issues), with 64% of total time spent in MCC according to May 2026 session statistics.
Use case 4: indie discovery.For a user whose profile is dominated by the majors (Marvel + DC > 80% of the collection), MCC offers a suitable indie selection. The score favors themaximum deviationon tone embeddings and narrative tags, without straying too far from creator and era preferences. A Marvel Bronze Age fan will see Daredevil by Frank Miller (marvel mainstream) if unread, then Sin City (indie Miller) which exports the tone to indie. A Marvel Modern fan will see Saga (Vaughan) then Paper Girls (Vaughan) then East of West (Hickman) then Lazarus (Rucka). Indie discovery causes 18% of pure Marvel/DC users to converge towards a mixed collection over 12 months (MCC internal study May 2026). SeeCalendar Image Independent Comicsfor the detailed list.
Filter by budget: monthly and maximum per issue
The MCC 2026 engine applies two budget filters at the pipeline output to adapt the recommendations to the collector's financial reality. Without these filters, a beginner at €50/month would receive recommendations from Amazing Fantasy #15 (€1,800 minimum) that are completely unsuitable for their budget. The budget filter is essential for the operational relevance of the engine.
Monthly budget filter.The user declares his monthly comics purchase envelope in the MCC parameters (by default 50 €, 100 €, 200 €, 500 €, unlimited). The engine calculates the aggregate cost of the 10 proposed recommendations and ensures that it does not exceed 1.5 times the monthly budget (to leave room for choice). A budget of €100 generates a list of 10 items whose total cost is between €80 and €150. The filter also applies to sleeper and completion use cases: no proposal at €300 for a user €50/month. The filter also adapts to the timing: if the user has already purchased for €80 this month, the new recommendations will be adjusted to the remaining €20.
Maximum filter per issue.In addition, the user declares the maximum acceptable priceby issue, which may be less than his monthly budget (a collector 200 €/month can refuse any unit purchase greater than 60 €). This filter eliminates expensive one-off pieces that would saturate the monthly budget and prevent diversification. For speculation collectors, the maximum filter per issue is typically aligned with the targeted sleeper range (for example €15-40 for a modest sleeper profile).
The two budget filters work in cascade: first the maximum per issue (eliminates overly expensive parts), then the aggregated monthly budget (balances the final list). The recommendations passed at the filter output are then reordered bybudget-adjusted score, which weights items slightly positively below 60% of the maximum per issue (to encourage efficient purchasing). This logic is compatible with the toolsfree estimateand the catalogcomicswhich display price ranges updated weekly.
Filter by availability VF: Panini, Urban, Glénat, Bliss
For the French collector who prefers reading in French, MCC 2026 includes a filter by VF availability based on four major publishers: Panini Comics (Marvel France, some DC), Urban Comics (DC France), Glénat Comics (partial Image, partial IDW), Bliss Editions (indie Image, Boom, Vault). This integration is based on the official catalogs updated monthly and on the MCC community database which tracks upcoming VF releases.
Strict VF mode.The user activates the modeStrict Frenchdans les paramètres MCC. The recommendations are then filtered to only offer comics available in an official French edition (single issue VF rare, newsstand format, bookstore format, complete, omnibus, deluxe). Strict VF mode reduces the base of eligible comics from 220,000 to around 38,000 (17%), but ensures that each suggestion is immediately purchasable in French. This mode is used by 43% of MCC France users as of June 8, 2026.
VF mode preferred, VO accepted.The user activates the modeFavorite French version. Recommendations remain drawn from the entire database, but comics available in VF have moved up in the rankings (bonus of +15% on the final score), while comics in VO only receive a penalty of -8%. This mode allows mixed VO/VF reading while favoring French access. It is used by 38% of MCC France users.
VO mode indifferent.No VF filter is applied, the engine works on the complete database of 220,000 comics. This is the default mode for advanced users who can read fluently in English. Used by 19% of MCC France users.
The engine also integrates the monitoring ofupcoming VF releasesannounced by Panini, Urban, Glénat, Bliss in their quarterly catalog. A series announced for December 2026 at Urban Comics is moved up in the recommendations from September 2026 if it matches the profile, with a badge“VF to be published”and the forecast date. This anticipation allows the collector to pre-order directly on publisher sites (Panini, Urban, Glénat, Bliss), specialized booksellers like Pulp's Comics, Album, Comics Zone, or platforms like BDFugue and Pulpitis. The guidecomics France collector guide pillardetails the VF pre-order strategy.
Beginner practical case: 0 collection, MCC start
Camille, 28 years old, has just discovered the comic book universe through Marvel and DC films. She doesn't have any comics in her collection. She opens an MCC account on June 8, 2026 and wants to start a collection without wasting her budget of €50 per month. This is how the engine supports it.
Registration and initial calibration.Camille downloads MCC mobile, creates her account, and arrives on the 16 covers calibration screen. She is visually drawn to Saga #1, Watchmen #1, Ultimate Spider-Man #1, Paper Girls #1, Sandman #1. The engine extracts an initial vector leaning towardscontemporary + serious narrative + indie/Vertigo + Vaughan/Hickman. She moves on to the rating stage: 5 issues are offered to her: Amazing Spider-Man #129 (Marvel Bronze), Watchmen #1 (Vertigo Copper), Saga #1 (Image Contemporary), Walking Dead #1 (Image Modern), Ultimate Spider-Man #1 (Marvel Contemporary). She rates respectively 3/5, 5/5, 5/5, 4/5, 4/5. The engine refines: profilecontemporary narrative + indie + Vaughan, low affinity Marvel Classic Bronze.
First recommendations.MCC produces a first list of 10 items weighted by use case next run to read (use case 1 by default for an account without collection). Top 3: Y: The Last Man #1 (Vaughan, Vertigo, €18 complete Urban), Paper Girls TPB Vol.1 (Vaughan, Image, €14 VF Urban), East of West Vol.1 (Hickman, Image, €16 VF Urban). Aggregated budget for 10 items: €95 before filter. Camille declared a budget of €50/month, so the filter demotes the more expensive items. Top 3 after filter: Y: The Last Man Vol.1 Urban €18, Paper Girls Vol.1 Urban €14, Saga Vol.1 Urban €16. Total €48. Camille buys Saga Vol.1 (which she has already rated 5/5 on the single issue) and Paper Girls Vol.1.
Convergence at 24 months.Over 24 months at €50/month, Camille builds a collection of 86 TPBs and 12 single issues (104 items). The engine gradually refines: going from recall@10 = 52% in month 1 to 71% in month 12. In month 24, Camille's profile shows 78% Image/indie, 12% DC/Vertigo, 6% Marvel Contemporary, 4% other. She discovered Stray Bullets (Lapham), Stray Dogs (Fleecs), Local Man (Fleecs/Seeley), Department of Truth (Tynion), Something is Killing the Children (Tynion). The engine converted a mainstream MCU beginner into an enlightened indie collector, in less than 24 months, with a total budget of €1,200. Seebeginner collector pillar guidefor the complete method.
Collector practical case: 1,000 comics, MCC optimization
Olivier, 41, has a collection of 1,040 comics as of June 8, 2026, accumulated over 18 years: 60% Marvel Bronze + Copper, 25% DC Copper + Modern, 10% Image Modern, 5% indie miscellaneous. He imported his CSV inventory into MCC in March 2025 and has been using the recommendations for 15 months. This is how the engine optimizes its collection in advanced mode.
Dominant use case: run completion.The engine identifies 47 partially collected runs in Olivier's base: Amazing Spider-Man #129-180 misses issues #144, #157, #169; Daredevil #168-191 misses issues #173, #185; New X-Men Morrison #114-154 misses issues #131-133. The engine produces a priority list of 28 missing issues ordered bynarrative tension(issues that complete a major narrative arc as a priority). It pushes to the top of the list: Amazing Spider-Man #144 (1st appearance Gwen Stacy clone, €42 raw NM), Daredevil #173 (Elektra arc, €28 raw NM), New X-Men #131 (Imperial arc Morrison, €18 raw NM). Over 12 months at €100/month completion budget, Olivier fills 24 holes out of 28.
Secondary use case: sleeper speculation.In parallel, the engine traces 14 sleepers compatible with Olivier's profile (Bronze + Copper). Top 3: Marvel Premiere #15 (1st appearance Iron Fist, €110 raw NM, sleeper since announcement of the Disney+ series remake January 2026), Doctor Strange Vol.2 #1 (Bronze, €28 raw NM, sleeper since the announcement of the film Doctor Strange 3 MCU phase 6), Werewolf by Night Vol.1 #32 (1st appearance Moon Knight, €850 raw NM, sleeper on the next Disney+ wave). Olivier adds the 3 sleepers to his high priority wishlist.
Tertiary use case: independent discovery.The engine detects that Olivier is underexposed to contemporary indie (only 5% of his collection). It pushes a selection of 5 indie series compatible with its Bronze/Copper profile (serious tone, epic narrative): Black Hammer (Lemire, Dark Horse, parallel to Bronze Marvel), Powers (Bendis/Oeming, Icon, street level Bronze-like), Stray Bullets (Lapham, El Capitan, crime), East of West (Hickman, Image, epic sci-fi), Lazarus (Rucka, Image, dystopia). Olivier tests Black Hammer Vol.1 (€18 VF Urban) which he rates 5/5, which opens up the entire Black Hammer sub-universe (Sherlock Frankenstein, Doctor Andromeda, Skulldigger) for priority recommendation.
15 month MCC assessment for Olivier.1,040 → 1,218 issues in collection (+178 net), including 64 completion issues filling 11 partial runs, 8 speculation sleepers acquired (average rating +18% over 15 months according to GoCollect), 24 contemporary indie issues (Lemire, Hickman, Rucka, Vaughan) with average rating 4.3/5. Total budget spent: €4,800 over 15 months. Current estimated value of the collection: €28,400 (vs. €22,600 in month 0). The MCC engine optimized budget allocation and oriented purchases towards narrative and financial value at 15 months.
FAQ: MCC recommendations in 8 questions
How many ratings are needed for the MCC engine to work well?
The MCC cold start converges into 5 minimum ratings. At this stage, the recall@10 (probability that a comic rated 5/5 appears in the first 10 recommendations) reaches 52%. Accuracy continues to increase until the 50th rating, where it peaks at 73-74%. For daily operational use, 25 ratings are sufficient (recall@10 = 65%). Beyond 50 ratings, the improvement is marginal.
Are MCC recommendations biased towards Marvel and DC?
No, the MCC 2026 engine is volume-neutral on publishers. The database contains 220,000 comics distributed approximately 38% Marvel, 26% DC, 19% Image, 8% Dark Horse, 4% IDW, 5% other (Boom, Vault, AfterShock, indie). The user profile dictates the final editorial weighting: a 100% indie collector receives 100% indie recommendations, a 100% Marvel collector receives 100% Marvel. The use caseindie discoveryalso allows you to exit a profile dominated by the majors if the user wishes.
Can I block a series that I no longer want to see recommendations for?
Yes, MCC allows you to block a series, creator, publisher or era from the advanced settings. The block is applied immediately and persists. For example, you can block all Spawn series (Image McFarlane universe) if you no longer wish to see any Spawn recommendations, or block screenwriter Brian Michael Bendis if his style does not suit you. The blockage is reversible at any time.
Does the MCC engine also offer VF Panini/Urban recommendations?
Yes, MCC integrates the official catalogs Panini Comics (Marvel France, some DC), Urban Comics (main DC France), Glénat Comics (Partial Image, IDW), Bliss Editions (indie Image, Boom, Vault). The modeStrict Frenchonly offers comics available in French edition. The modeFavorite French versionmoves the VFs up in the ranking without blocking the VOs. The modeVO indifferentworks on the entire database of 220,000 comics.
How does MCC distinguish a reader collector from a pure collector?
MCC separates inventory (issues owned) from reading status (issues marked read). If you have 500 issues and have only marked 80 as read, the engine identifies a profilepure collectorand directs the recommendations towards the completion run and sleeper speculation axes, reducing the weight of the next run use case to read. Conversely, if you read 95% of what you own, the profiledrivewill dominate and narrative recommendations will take precedence. The transition between the two profiles is continuous, calibrated by the reading/possession ratio.