Repository navigation
docs(specs): add functional specs for ten roadmap initiatives - #1293
Merged
Merged
Conversation
The roadmap says what each initiative must achieve. Nothing said what it does in detail, so an agent picking up an initiative had to invent the product decisions. These ten specs close that gap, each in the house format already used by commander-power-meter.md. Every spec carries a dated market sweep of the whole category, not only the named incumbents, and a verified account of what the code does today. Three findings from the sweep shaped the specs: - Generation is commoditised, verification is not. Six products sell conversational Commander building; one documents that its own results vary between runs. None publishes which claims were checked against card data and which are model reasoning. - Collection-to-deck is crowded. SuperBrew, DeckFinder, ManaBuilder, MTG Agents and ManaForge all rank by coverage. None excludes copies already committed to another of the player's decks, and none separates Oracle identity for legality from printing identity for price. - Reproducibility is absent category-wide. Five hand simulators report numbers without a seed, an interval or a sample size. Two corrections to earlier assumptions, both from reading the sources: - I-14's "show the query before it runs" is already shipped by OffMeta. What is unoccupied is deck context surviving an edited query, and French. The spec scopes to those. - I-20's reference, Archidekt collaborators, grants direct real-time write access, not review. Propose-and-accept exists nowhere in the category, so the risk is overbuilding, not falling behind. I-13's spec records the UI audit: no card art anywhere, a design system document describing tokens the code does not have, a dead 3D stack with zero imports, and borders at 1.51:1 and 1.33:1 against a 3:1 requirement. Each spec ends with delivery slices and their dependencies, so the three scheduled agents can take work without colliding. Verified: tsc clean, lint clean, 3509 tests passing, every relative link in the new directory resolves. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Codex Review Summary: ❌ FailedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the
 |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
KaelSensei
added a commit
that referenced
this pull request
Nov 8, 2026
…ives (#1294) Follows #1293, which landed the ten specs for the committed initiatives. That PR was merged before this work was pushed, so it ships separately. The roadmap ended at I-21. Three agents take work from it every day, so the pipeline needed depth, and depth is only useful if it is grounded. Six initiatives, each from a verified gap in a second market sweep covering angles the first pass missed - Discord, accessibility, cEDH meta sources, offline apps, collection apps and localisation: - I-22 Verification as a public surface. Six products generate decks and none publishes a verification contract. The validator is already deterministic and /api/v1 already exists. - I-23 The pod's own surface. Otterly MTG, Mana Grove and an open-source Commander bot already answer bracket questions in Discord. None carries a pod's declared Rule 0 agreement, which is I-15's artifact. - I-24 Decks a screen reader can play from. Blind players adapt cards with NFC tags read by a phone; braille fails because marked cards risk disqualification. No deckbuilder emits output shaped to be heard. - I-25 A third locale, chosen on evidence. Every AI deckbuilder found is English-only. Nine catalogues sit dormant here, and the spec refuses to pick from catalogue convenience. - I-26 Collections arriving from elsewhere. Scanning is already a non-goal, so import is the only honest way a collection is populated. - I-27 Learning while building. Tools assume the vocabulary and teaching lives in video. Nobody explains a rule at the moment it fires, and the repo already carries mtg-rules and mtg-learn skills. Four things the sweep ruled out are recorded so the next pass does not re-derive them. The most useful: tournament win rates stay blocked. EDHTop16, the best candidate, publishes commander meta share and entry counts but no win rates, no match-level results and no public API; MagicMeta and LucidPuzzle derive their tiers from it. The 2027 specs are deliberately less detailed than the Next-horizon ones. Committing to interface detail eighteen months early produces precision that ages rather than guides. Six and not twenty, stated in the roadmap: beyond these the sweep stopped producing gaps this product is positioned to fill, and inventing more would fill a roadmap without filling a pipeline. Verified on current dev: tsc clean, lint clean, 3510 tests passing, prettier clean, every relative link resolves. Co-authored-by: Codex <agent-local@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
The roadmap states what each initiative must achieve. Nothing stated what it does, so an agent picking one up had to invent the product decisions — and three scheduled agents now pick work from it every day.
These ten specs close that gap. Each follows the house format already used by
commander-power-meter.md, and each carries a dated market sweep of the whole category plus a verified account of what the code does today.What the sweep found
Broad, not limited to the named incumbents. Three findings shaped every spec:
Generation is commoditised; verification is not. Six products sell conversational Commander building (ManaTap, Farseek, DeckCheck, MTG Master, ManaForge, CommanderSalt), several free. DeckCheck's own support page documents that its AI results vary between runs. None publishes which claims were checked against card data and which are model reasoning. That is I-02's opening, and it is a presentation change over a validator that already passes its tests.
Collection-to-deck is crowded; the ownership detail is not. SuperBrew ($6/mo), DeckFinder, ManaBuilder, MTG Agents and ManaForge all rank decks by collection coverage. None states that a copy already committed to another of your decks is excluded, and none separates Oracle identity (legality) from printing identity (price).
Reproducibility is absent category-wide. ManaTap, ScrollVault (50 000 hands in WASM), MyDeckStats, KrakenTheMeta and Archidekt all simulate thousands of hands. Not one publishes a seed, an interval or a sample size. For a product positioning on checkable output, that is the widest opening found.
Two corrections to earlier assumptions
Both came from reading the sources rather than trusting the roadmap's framing:
The specs
I-13 records the UI audit
Measured on a local dev server at 1440×900 and 390×844, contrast computed from
src/styles/globals.css:design-system.mddescribes tokens the code does not have —grep --gold src/returns 0 results.three,gsap,@react-three/*,@types/three, all with 0 imports; the README still advertises a 3D landing.--accentas text is 3.09:1 against 4.5:1, used atMoxfieldProfileResults.tsx:102ontext-xs.Built for the automations
Every spec ends with a delivery-slices table carrying dependencies. That is the part the three daily agents need: each slice is independently reviewable, and the dependency column says which cannot start yet. Without it, two agents can pick overlapping slices on the same day — which already happened once this week.
Scope discipline
Documentation only. Per
CLAUDE.mdthis is achore/docsPR becausedocs/product/roadmap.mdis touched (ten one-line spec links plus a pointer to the new directory). No initiative status changed and no scope box was ticked — a spec is a plan, not a record.docs/project/changelog.mdanddocs/project/progress.mdare untouched.Verification
npx tsc --noEmit— cleanpnpm lint— zero warningsTZ=UTC pnpm test— 344 files, 3509 tests passingprettier --writeon all eleven new filesdocs/product/specs/resolves to an existing file (checked programmatically)SKIP_E2E=1on the push: Docker is unavailable in this environment and the diff is Markdown only.For review
Competitor rows are claims read from each product's public pages on 2026-11-08, not evidence. Each spec says so, and each should be revalidated before a row justifies implementation.
🤖 Generated with Claude Code