Add fantasy football auction draft tracker Firefox add-on

[RaphaelLikesCoding]

A read-only Firefox sidebar add-on that watches a live fantasy football auction draft and turns it into an analytical layer: who was taken, for how much, what every team has left, what they still need, and what the player on the block is actually worth in this market.

Lands in a new top-level fantasy-auction-tracker/ directory. Nothing in the existing BLIS-Net codebase is touched.

Design decisions worth reviewing

It reads the DOM and the WebSocket, not the screen. The original ask was framed as "look at a browser screen," but a content script already lives inside the page, so pixel-scraping is a last resort rather than the design. Three layers, best first:

Layer	Mechanism	Confidence
ws	page-world shim wrapping window.WebSocket, mirroring frames	1.00
dom	MutationObserver over the draft board	0.85
ocr	canvas fallback (designed for, not implemented)	0.55

The WS tap is strictly passive — it never sends, blocks, or alters a frame. Both implemented layers run simultaneously; the store deduplicates, so the DOM layer silently covers whatever the WS mapping misses instead of leaving a hole you notice only after the draft.

State is an append-only event log. Budgets, rosters and analytics are a pure function of that log (src/core/reducer.js — pure, total, deterministic). That is what makes replay, undo, and mid-draft page-refresh recovery work at all. Low-confidence observations are quarantined in a review queue rather than applied, and unresolved problems surface as alerts — silence never means "fine."

The platform seam is data, not code. Site knowledge lives in selector profiles, so repairing a redesign is a config change rather than a rewrite. That matters when a platform can reskin the week before your draft.

Analytics

• Inflation — remaining money ÷ remaining par value, with the $1-per-open-slot floor removed from both sides (the naive form overstates it, since pinned dollars can never chase value)
• Max bid per team — money left − $1 × other open slots; the real ceiling a rival can reach, usually well below their raw budget
• Positional scarcity — startable players left vs. starting slots still to fill, with replacement level derived from roster settings and flex demand split across eligible positions
• Tier cliffs — the drop to the next player at a position, which is what justifies paying over par
• Budget pressure — which teams are nearly locked into $1 bids
• Bid advice — walk-away (par × inflation) and ceiling (+ a configurable slice of the tier cliff), both hard-capped by your own max bid so it never advises a bid you cannot legally make

Valuations are imported from CSV (alias-matched headers; FantasyPros exports work as-is) and rescaled to the league's total money, so a sheet built for a $200/12-team league does not silently bias every suggestion in a $300/10-team one.

Testing

src/core/ has no browser dependency, so the whole analytical layer runs under node --test with no DOM shim — 57 tests, all passing.

tools/simulate.js generates a deterministic 12-team $200 auction (192 sales, every team landing on exactly the cap) plus a matching valuation CSV. tools/replay.js rehearses the full pipeline against it. The replay tests assert end-to-end that no player is drafted twice, no team overspends, every dollar is accounted for, and reducing a prefix of the log equals replaying to that point — the property page-refresh recovery depends on.

Indexing the valuation lookups (they were linear finds inside a reduce, i.e. O(sales × players) on every event) took the suite from 14.1s to 1.8s.

What is not finished

• The NFL.com and CBS selector profiles are provisional. They were written without access to a live auction draft room and are marked as such in profiles.js. tools/calibrate.js derives the real ones from a mock draft.
• The WS field mappings need a real capture, so the WebSocket layer ships in record-only mode: it observes and captures frames, and the DOM layer does the work until a mapping is confirmed.
• OCR is designed for but not implemented — only needed if a platform renders its draft board to <canvas>.

The platform was given as "NFL.com / CBS / other" without a specific pick, so the core and adapter layer were built platform-agnostic; narrowing to one platform is a profile + mapping change, not a code change.

Scope

Read-only and local. It observes your own draft and advises; it does not place bids, automate any interaction, or send anything anywhere — all state stays in browser.storage.local. Auto-bidding would put the user crosswise with platform terms of service, so it is deliberately out of scope.

---

Generated by Claude Code

[RaphaelLikesCoding]

Pushed 188c87b — adds the tiered target board with adjustable price targets, plus the CBS testing work.

Target board

Tiers group players who are interchangeable to you, so losing one only matters when the tier runs dry. Each tier is flagged healthy, critical (≤2 names left) or exhausted.

Price targets are per-player and override the market-derived ceiling entirely — set $42 and the Live tab says PASS at $43 regardless of what the CSV thinks. Every price and tier is an editable cell that saves on change, not on keystroke, so a half-typed "4" of "45" never briefly becomes the live ceiling mid-bidding.

Inflation adjustment (optional, off by default) scales targets with the market, since a $40 target in a 1.2× market is really a $48 target. When on, the adjusted figure renders beside the typed one so it is never a silent override.

Plan feasibility is the number that earns the tab — it fits open targets into remaining roster slots and compares the cost against money in hand:

target board
  tier 1:  0/6 open  won 0 ($0)   lost 6   EXHAUSTED
  tier 2:  1/6 open  won 2 ($41)  lost 3   CRITICAL
  tier 3:  5/6 open  won 0 ($0)   lost 1

plan: $127 of targets for 6 of 13 open slots, $99 available -- SHORT BY $35

Walking into "$12 left, plan needed $60" is the classic auction death; this makes it impossible. Rehearsable via npm run replay -- ... --targets fixtures/sample-targets.json --me team-4.

On testing against CBS

I could not reach cbssports.com from this environment — the network policy returns 403 at the proxy on CONNECT — so live calibration did not happen and the CBS selectors remain unverified. Two things were done instead:

1. The DOM adapter is now genuinely tested against a real DOM (linkedom, dev-dependency only — the add-on still ships dependency-free). fixtures/cbs-draft-room.html is a CBS-shaped room built deliberately messy: a header row that parses to nothing, $58 / 58 / $7 price formats, a suffixed name, a D/ST alias, and a row with no position. The full chain from DOM mutation through to league state is asserted end to end. This proves the adapter handles a room of that shape — it does not prove CBS uses those class names.

2. tools/selftest.js closes that gap in the room, in one paste, and is itself tested. It separates FAIL (container present, selector genuinely wrong) from EMPTY (nothing has happened yet) from NO-SCOPE (parent absent) — conflating those is what sends you chasing selectors that are already correct. __auctionSelfTestLive() additionally confirms the room is observable at all, rather than canvas/shadow-DOM rendered.

Two bugs found while building this

Both were caught by the demo/tests rather than by inspection, and both were real:

• LEAGUE_CONFIGURED erased your own team identity. It merged its entire payload, so a config event carrying myTeamId: null silently wiped the value you set — disabling your max-bid cap, the no-slot verdict, and the whole target plan. Config merges now ignore null/undefined. Regression tests added.
• The self-test scoped row selectors to the first matching row, which is usually the table header. That reported correct selectors as broken and let a genuinely wrong one pass as "waiting for a sale" — precisely the false confidence the tool exists to prevent. It now scopes across every matching row.

124 tests pass (was 57).

---

Generated by Claude Code