BYTEPATH++ realignment

Web mention sweep (past 2 months, 3 parallel agents):

• Quiet window overall: ~33 verified third-party mentions across all three terms; zero press, zero HN, no confirmed Reddit; X/Twitter unswept (auth-gated) and Reddit crawler-blocked — flagged as real coverage gaps.
• SNKRX (25 mentions): the Serpent Squad situation is the biggest live thread — a licensed MIT re-release on Steam (app 4281080, UncleChen Games) kept a "Your game is stolen @DEV" thread active through June–July, surfacing explicit sequel demand; GLIMPO flagged as a second SNKRX-alike. SNKRX hit an all-time-low $0.83 (72% off) during the Summer Sale (Jun 25–Jul 9), driving a review cluster (14 positive / 1 negative in-window). Small-channel YouTube burst the week of Jul 15–18 (Ashley Eck 3-parter, Backlog Bureau commentary video).
• BYTEPATH (6 mentions): two fresh positive Steam reviews (one: "pre-empted the Vampire Survivors craze"); Automattic's kandelo project adopted BYTEPATH as a Chromium rendering-regression test (PR #843 fixes an RGBA4 canvas regression that broke the 127-gray displacement shaders and adds a CI test that boots the game); project-based-learning PR #893 fixed all 16 stale SSYGEN→a327ex tutorial links.
• a327ex (2 mentions): the tutorial link-maintenance PR + a Chinese auto-battler design doc citing a327ex as a mechanics authority.
• Incident: the first SNKRX agent opened a YouTube deals video in the browser pane that autoplayed audio on the machine; playback was torn down (pause/mute, then removeAttribute('src')+load()) and the sweep relaunched with a hard no-video-playback rule. The video ("STEAM UNDER $1 OVERWHELMINGLY POSITIVE GAME DEAL", CultOfMush) was later verified from metadata: April upload, zero SNKRX content — not a qualifying mention.

Full Steam review-corpus analysis (all 4,336 SNKRX + 316 BYTEPATH reviews):

• Method: full dump via the appreviews API (cursor-paginated, all languages) with fetch_reviews.py; stats + deep-read chunking with prep_analysis.py (every negative + every substantive positive ≥120 chars = 1,566 SNKRX + 185 BYTEPATH selected); 8 parallel chunk-analysis agents; synthesis delivered in chat and saved as steam-market/reports/24-own-games-review-corpus.md with raw dumps in steam-market/data/ and scripts in steam-market/scripts/.
• SNKRX tells a five-act story: launch/streamer surge (Northernlion ~30 mentions; weekly update cadence WAS the marketing) → abandonment transition (hostile reviews are retroactive edits citing the announced-but-unshipped rewrite) → controversy year (~80% of negative surface reputational, decoupled from product — "maybe literally the best $3 game to ever exist" inside a thumbs-down) → Brotato era ("better along basically every dimension") → 2025-26 equilibrium ("charming complete classic"; by the final month nobody mentions the dev or any comparison game — the slate is clean, so an update lands as resurrection, not obligation).
• Cross-era product signal, ranked: shop RNG/build-assembly (the immortal complaint — players spec the fix themselves: banish/pity/pool-thinning/class-weighted odds); the NG+ structural flaw (+1 segment enlarges the hitbox exactly as one-shots begin — "harder difficulty by making your avatar fatter and slower"); unavoidable damage; the agency paradox ("the game would not suffer at all if the player had no control" at launch, corner-camping endgame in 2026; the reviews' own answer: segment adjacency synergies); opacity (players call "imbalance" partly because numbers are hidden; global achievement stats audit balance publicly); performance as a fun tax (swarm builds lag exactly at the power fantasy); friend/foe color collisions; the pause-doesn't-pause bug trio.
• Price evidence: ~55 explicit would-pay-more statements ($5 to $15-20, three literally requesting a price raise) vs ~2 "overpriced" ever — but "for the price" is the hedge in a third of tail-era positives (sentiment is price-subsidized), so a reprice must ride the quality pass. BYTEPATH's darkest datapoint: a JP review quotes the dev stopping updates "because it didn't sell" at 205 yen — reviewers themselves concluded it was priced too low to sustain.
• BYTEPATH findings: the tree IS the product (~100/160 positives; "The magic is in the menus"; nobody ever asked for a smaller tree, only tools — search, PoB-style stat totals, build slots); the modern negative case is three-headed (trust violations — escape-wipe 24 helpful; the performance ceiling ~24 mentions, builds crash at their peak; post-VS standards — "upgrades should change your playstyle, not pump stats") plus a door problem (9/14 modern negatives are sub-1.5h ejections); the tutorial halo is real ("Hello Arnold.") and weaponized exactly once ("he teaches game dev and shipped this").
• Key synthesis (became the remake's organizing principle): players accept chosen stakes and revolt against un-consented loss; SNKRX's sanctioned ALT-F4 scumming and BYTEPATH's escape-wipe rage are the same lesson from both sides. "The original punished players by accident; the remake should demand commitment on purpose."

BYTEPATH++ pivot (owner decision):

• Owner: the tower design "is trying to do too many things at once… I've not settled on a design that seems clear or good yet and its been like a week. So what I'd like to do instead is do the safest thing possible, just make a game that is exactly like BYTEPATH, but improve it where it can be genuinely improved."
• Assessment agreed: it's a return to the settled steam-market plan, not a retreat; the tower design stacked three unproven load-bearing bets; the corpus is unusually kind to a faithful remake ("$10 game with a permanent 80% off discount" is the business case verbatim).
• Decisions: (1) combat = ORIGINAL tank-ship controls (cursor-is-ship dropped); (2) FULL content redesign — passives/actives/abilities/ships and therefore the tree; (3) progression = PoE-style characters — level to endgame, build goes crazy, play endgame, then new character or push to max; max level ≈ whole tree explored; a character normally walks only part of the tree; NO forced deletes; explicitly not an incremental; (4) tower design archived via folder rename.
• Scope: NOW = content (inventory → owner keep/cut pass → redesign → implement); LATER = high-level loop design. Actual coding starts next session.

Reorganization and content extraction:

• E:/a327ex/BYTEPATH++ renamed to E:/a327ex/BYTEPATH++-old (git history intact); fresh BYTEPATH++/ created with design.md (project document: decisions, organizing principle, 8 review-corpus mandates, commitment notes, references, open questions) and reference/ docs.
• Source-path trap discovered mid-extraction: the old design.md pointed at D:/2017-2018 code/code/bytepath, which is a PRE-RELEASE snapshot (no audio, no class system, 6 of 14 enemies; a proc calls an undefined addBuff — would crash). The real shipped open-source release is D:/2020-2021 code/old/BYTEPATH (Player.lua 2,396 lines, rooms/Classes.lua). All four inventory agents were resumed and redid their work against the shipped source; snapshot-only facts preserved as appendices (incl. the dev's balance stat-budget worksheet and todo).
• Inventories written (724 checkbox items for the keep/cut pass): tree.md (833 nodes — 24 big/79 medium/730 smalls, 139 stat keys, flat 1/3/6 SP costs, max_tree_nodes=80 +5/loop, no-refund respec, two 100× value bugs + one 10× flagged); player-kit.md (17 attacks, 9 ships, 40 in-run classes, the in-run Item system, ~45 procs, six genuine shipped bugs incl. Bomber's 0.25-vs-25 double-attack typo); enemies-arena.md (14 enemies incl. Glitcher miniboss + disabled Rotator; adaptive Director max(3*(kills/created)^4,1), HP scaling past d40, Item/Key clocks; walls bounce for 10 not insta-death); meta-structure.md (console shell, 40 Steam achievements, 69-file AIRGLOW "Memory Bank" audio inventory, effects intensity settings 0-10 — the feature reviews credited with saving recommendations, loop NG+ +10 SP/+5 node cap, 8-key secret "escape" ending, two-file bitser save split).
• reference/visual-language.md (618 lines): the look extracted from BYTEPATH++-old — 5-stage post stack (displacement → bloom → 3x upscale → additive combine → playscii CRT → flash), born-white-colored-later idiom, terrain visuals, owner-tuned baked values flagged do-not-revert (u_disp 0.0326389, ghost 0.102431, crush 0.215278, bloom 0.35/0.3/3/1.33), 15 porting gotchas, and the .rogue not-yet-ported list. Hard constraint recorded in design.md: the 1px-outline language depends on Anchor 3 engine fixes (inset-ring stroke, sized layers, raw-GL aux samplers) — stock Anchor 2 renders it invisible; decide engine basis at implementation start.
• reference/inventory/rogue-kit.md: BYTEPATH.rogue gameplay inventory — 9 attacks, 61 passives (20 projectile mods/21 stat smalls + remap curves/5 constructs/5 on-kill/5 cycle procs/5 specials), 12 enemies, 56 spawn formations (19 reachable), devices/DATA economy. Structural finding: the prototype's meta was façade (devices never changed the attack, DATA unearnable, constructs unacquirable, the run a ~31s scripted chain); mechanics real, game never wired. Catalog corrections: finale never fires (counter resets at 4, trigger ==5), supershot every 8th not 6th, amplify's DoT rider dead code.
• Memory updated: project_bytepath_plus_plus.md rewritten for the pivot (+ MEMORY.md index line) — records the source-path trap, the engine dependency, and that the next step is the owner's keep/cut pass.

Notable connective findings:

• Shipped BYTEPATH already contains proto-versions of the new direction: loop NG+ raises the tree-node cap +5/loop ("more tree as you deepen"), and respec was already no-refund — the review complaint was its illegibility, not its existence.
• The tree's real shape (88% stat smalls) is exactly what decision 2's redesign is up against, per the corpus's "upgrades should change your playstyle" bar.

Only the summary of this log is public. Private because this is BYTEPATH++.