Prompt and caching are the two weakest levers available: output tokens are ~96% of a Quiver call, Arrow's API has no cache parameter, and the gap already closed itself through local asset reuse
The question
Verbatim: "What prompt or caching changes to the Quiver art pipeline would bring Scribble Works' AI-art spend back under the $20/mo autonomous-spend ceiling?"
Context: the 2026-10-09 unit-economics review projected studio art at about $71/month against the $20/month autonomous line in [[project_rdco_spending_authority]], and asked whether batching, caching or prompt-shortening could close the gap. That projection has since been superseded by measurement, which changes the answer.
What we already know (from the vault)
- The $71 was a projection, not a measurement, and it did not happen. The 2026-10-09 review (F3) took $2.37/round measured over 10-01 to 10-08 and multiplied by a 30-round month to get a $36-$118 range, "~$71 at mean" ([[2026-10-09-unit-economics-cost-basis]]). The 2026-10-16 review measured the next eight rounds at $7.14 total, about $0.89/round, and about $0.34 per new game across 21 game pull requests ([[2026-10-16-unit-economics-cost-basis]], F1). Per-round cost fell about 62% in one week, driven by art reuse and code geometry.
- Measured October art is about $29.51 at 10-16, not $71 ([[2026-10-16-unit-economics-cost-basis]] F4, from
~/.claude/state/studio/rounds.md:846-1006 plus the 10-09 figure): $18.47 for 10-01 to 10-08, $7.14 for 10-09 to 10-16, and $3.90 for two refreshes. The same review's forward pace was ~$43 by 10-31, ~$53-57 with the six refreshesqueue.md:796 requires. - Spend is already at $0/round and games still pass the gate. Ray imposed a $0 Quiver hard cap on the 10-17 round because October was past the $20 line; that round closed 4/4 gate PASS with Quiver $0, and the 10-18 round did the same (
rounds.md:1016-1022, :1031-1041). #452 shipped on 7 library reuses and 0 Quiver calls. This is the strongest evidence in the record for the reuse lever, and it is measured, not modelled. - Ruling 20 already makes code the default for anything derivable. The founder's correction of 2026-09-02 is explicitly about cost basis: "If we write code it can be run as many times without incremental cost... there is still a difference between costing nothing and costing just a little" ([[2026-08-31-product-model-rulings]]:80-81). Geometry in code is on-contract; illustration in code is the 2026-10-01 regression ([[feedback_use_real_art_tools]]) and is off the table.
- Reuse today is manual path-copying, and the catalog that would automate it is unmerged. On
origin/main(46ed046a) there are 0 files undercontent/art/, noart:findscript, and 0 standalone.svgfiles: art lives inlined in 136game.htmlfiles. PR #428 ("Reuse drawn art: art library Phase S (113-drawing catalog, duplicate gate, art:find)", #396) is OPEN DRAFT with 100 files. So the $0 rounds were achieved by grepping existing game HTML and byte-copying path markup from open pull-request heads, which is why merge-order coupling and the R53 rework list exist (rounds.md:1005, :1019, :1022).
What the web says
- Arrow 2 is already the cheaper of exactly two models.
mcp__quiverai__list_models(called 2026-10-19) returns onlyarrow-2(isDefault: true) andarrow-2-telos("Excels at complex SVG creations and drains usage much faster"), bothpricing.kind: token_usage. Published rates are $4/M input and $20/M output forarrow-2, $6/M and $30/M forarrow-2-telos([[2026-10-14-quiverai-tos-commercial-art-license]], sourced from QuiverAI'sllms-full.txtandapi.quiver.ai/v1/openapi.json). Telos appears nowhere in studio state, Scribble Works docs, or the printables project. There is no cheaper tier to move to; the only available direction is 1.5x more expensive. - Output tokens are the entire cost story, and the measurement is ours. The 2026-09-18 bake-off recorded
usageper call: input was a flat ~722 tokens while output ran 6,800-21,100 tokens per SVG, and "reasoning is billed as output" (content/engine-spike/bakeoff-arrow2/scorecard.md,IMPLEMENTATION-NOTES.md:38-57). At $4/M input, 722 tokens is $0.0029. Against the measured production mean of about $0.080/call, input is roughly 3.6% of the bill. max_output_tokensdoes not bound spend. Two bake-off calls returned 20,013 and 21,147 output tokens against a 16,384 cap, because reasoning tokens are billed as output and appear not to be capped by that knob (IMPLEMENTATION-NOTES.md:56-57).reasoning_effortis an accepted request field and was left at Arrow's default in every run on record.- There is no cache parameter in the Arrow API. The documented request body is
model,prompt,n,instructions,reasoning_effort,max_output_tokens,attributes.viewBox, and unknown fields are rejected with a 400 (IMPLEMENTATION-NOTES.md:42). Nocache_controlanalogue exists, and no QuiverAI page reached in this research mentions prompt caching. Generic Anthropic-style prompt-caching semantics do not transfer to this surface. - Subscription credits and API spend are separate pools.
docs.quiver.ai/app/pricing(retrieved 2026-10-19) states: "API calls use the API Platform prepaid balance, not weekly or prepaid App credits," and "QuiverAI offers monthly plans with included weekly App credits."quiver.ai/pricing(same date) lists Go $8/mo, Plus $20/mo, Pro $40/mo, Enterprise custom, with included weekly app usage of $5 / $15 / $40, "Resets every monday UTC (no rollover)," overage "via prepaid balance + optional auto-top-up," and 20% off annually. - Batching is real but sized for a problem Scribble Works does not have. The one-call icon sheet plus bounding-box split measured about 3x cheaper per icon, but only above 12 subjects, and the output has no groups (243 flat elements on 09-19) so splitting needs bbox clustering ([[2026-10-18-quiver-grok-art-tool-decision-framework]]). Measured drawings per Scribble Works game run 0-11, so the sheet threshold is almost never reached. It is a coloring-book lever, not a studio one.
- Quiver's own library reuse is unavailable on our path.
mcp__quiverai__search_librarywith a broad Scribble Works query returned{"items": []}on 2026-10-19. The REST generations endpoint returns SVG inline and persists no Artifact, so there is nothing for the app library to index. Reuse viasearch_libraryis not a lever for this pipeline; reuse via the local catalog is.
Convergences and contradictions
- Convergence: the vault and the vendor agree that the drawing, not the prompt, is the bill. Published rates put output at 5x input, and our own
usagelogs put input at a flat ~722 tokens against 6.8k-21.1k output. Both roads lead to the same conclusion, which is that the question's two named levers are bounded at a few percent before quality is even considered. - Contradiction, and it is the premise: the $71 figure is stale by two revisions. The backlog note and the 10-09 escalation both carry ~$71; the 10-16 measurement says the per-round basis that produced it fell 62% within a week, and the 10-17 and 10-18 rounds ran at exactly $0 with full gate passes. Anything written against $71 today is arguing with a number the studio has already beaten.
- Contradiction worth flagging: every dollar in this brief is imputed, not billed. The coloring-book implementation notes record that the "Quiver balance endpoint 403 (key lacks
billing_read) -> Quiver cost = provider-reported tokens x list price ($4/M in, $20/M out)" ([[2026-10-06-composition-test-implementation-notes]]). So $71, $29.51, $0.34 and $0.080 are all token counts multiplied by published list price. None has been reconciled against an invoice or a prepaid-balance drawdown, and the ledger cannot help:cost_micros_artsums to $15.33 across 63 rows with newestcreated_aton 10-07, andinput_tokens/output_tokensare 0 or NULL on every row.
Synthesis for RDCO
The question asks about the two worst levers on the board, and the arithmetic says so before taste does. Prompt shortening touches input tokens only. Input is a flat ~722 tokens per call, $0.0029 at the published $4/M, against a measured production mean of about $0.080 per call: 3.6% of the bill. Deleting the entire prompt would save about $1.06 on a $29.51 month. Even allowing that production prompts run three times longer than the bake-off's, input tops out near 11%, so halving prompt text buys 2-5%. Caching is worse than weak, it is absent: the Arrow generations endpoint accepts seven body fields and rejects unknown ones with a 400, so there is no cache parameter to set, and if there were it would act on the same 3.6%. The levers that matter all act on output tokens, which are the SVG's own path data plus reasoning billed as output, and none of them is a prompt or a cache.
What actually closed the gap was local asset reuse, and it closed it harder than any prompt change could. Per-round art went $2.37 (10-01 to 10-08) to $0.89 (10-09 to 10-16) to $0.00 on both 10-17 and 10-18, with 4/4 gate PASS on each of those two nights. That is a 100% reduction on new-game art, achieved by the round brief's standing order to search content/art and existing game SVGs before generating, plus ruling 20's code-geometry default for grids, mazes and tables. Two structural notes keep this honest. First, the mechanism is not Quiver's search_library, which returns empty for this account because the REST path persists no Artifacts; it is grepping 136 inlined game.html files. Second, the catalog that would make it systematic, PR #428's 113-drawing catalog with a duplicate gate and art:find, is still an OPEN DRAFT, so today's reuse depends on byte-copying from unmerged pull-request heads, which is what generates the merge-order coupling and the R53 rework list. The lever is proven; its plumbing is not merged.
The residual driver is refresh rework, which is a merge-throughput problem wearing a cost problem's clothes. Refreshes measured $2.28 (#368) and about $1.62 (#383), five to seven times a new game, and queue.md:796 attaches one to every game pull request older than #425. The 10-16 review's own pace arithmetic is the tell: about $43 by 10-31 on new games alone, about $53-57 once six refreshes land. Those refreshes exist because 0 of 21 game pull requests merged and main last moved on 10-11, blocked on the production D1 catalog token open since 10-11. So the cheapest dollar in this whole analysis is not a prompt edit; it is the founder's decision on the merge-as-hidden-pending-review proposal and the D1 token, which retires the refresh rule's trigger entirely.
One structural finding deserves separate weight, because it could make the metering question moot rather than optimised. QuiverAI runs two disjoint pools: subscription plans carry included weekly app credits (Go $5, Plus $15, Pro $40, resetting Mondays UTC with no rollover), while "API calls use the API Platform prepaid balance, not weekly or prepaid App credits." Studio builders fetch QUIVERAI_API_KEY through op-service-account.sh and hit the REST generations endpoint, so Scribble Works art is drawing the API prepaid balance, and the subscription's weekly app allowance is being paid for and left unused. Go at $8/month includes roughly $21.70/month of app usage and is comfortably autonomous; Plus at $20/month includes roughly $65/month but sits exactly at the founder-gated threshold. If the MCP send_prompt surface debits app credits rather than the API balance, then a flat $8/month plan could absorb the studio's entire measured art volume and the variable-cost problem becomes a fixed-cost line. The supporting evidence is suggestive rather than conclusive: MCP returned insufficient_credits on 09-19 while REST worked, which is what an exhausted weekly no-rollover pool looks like, and today's list_models shows generate with allowOnDemandCredits: false and state: "ok". I am labelling this an inference, not a finding, because no page reached documents which pool MCP debits and the billing endpoint 403s.
Recommended configuration, with the arithmetic
Baseline, imputed: about $29.51 October-to-10-16; measured $0.34 per new game since 10-09; measured $0.080 per call over the 54 calls logged across #434 and #437-#446; about 2.6 new games per round, roughly 79 per month.
| # | Change | Measured effect | Confidence |
|---|---|---|---|
| 1 | Keep arrow-2; never arrow-2-telos |
$0 saved, prevents a 1.5x regression | High (vendor rates; Telos unused today) |
| 2 | Merge #428; make art:find a mandatory pre-step before any generate call |
$0.89/round to $0.00 on 10-17 and 10-18, gate 4/4 both nights | High (measured, two rounds) |
| 3 | Cap new drawings at 3 per game | Heavy games were #434 (11 calls, $0.74) and #438 (9, $0.81), both pre-pass driven; 3 calls = ~$0.24 | Medium |
| 4 | In the blind legibility pre-pass, swap from catalog before regenerating | #439 did exactly this (sun swapped #400 to #398): 1 call, $0.07 for the whole game | Medium-high |
| 5 | Extend ruling 20 code geometry to every derivable element | #448, #451, #452 shipped at $0 | High (founder ruling + measured) |
| 6 | One-call icon sheet only above 12 subjects | ~3x cheaper per icon, threshold almost never reached in studio games | Medium |
| 7 | Cap refreshes at 3/month pending the merge decision | $1.62-$2.28 each measured | High on unit cost, medium on count |
| 8 | Lower reasoning_effort one step |
Unknown; untested in every run on record | Low on magnitude |
Path to under $20, stated as three different numbers because they are three different things:
- The $71 was a projection and is superseded. Do not carry it forward.
- 79 games x $0.34 (the 10-09 to 10-16 average) = $26.86/month, still over.
- 79 games x $0.163 (the 10-16 round's actual average, 3 games for $0.49, which is the round that already followed items 2-5) = $12.88/month, under with about $7 of headroom.
- Add refreshes: six at the measured $1.62-$2.28 = $9.72-$13.68, which breaks the ceiling on its own. Three refreshes = $4.86-$6.84, giving $17.74-$19.72. Under $20, with almost no margin.
The binding constraint is therefore refresh count, not prompt or cache design. Items 2 through 5 are already sufficient on new-game art; items 7 and the merge decision are what decide whether the month lands under the line.
Unverified assumptions in that arithmetic, named explicitly:
- Every dollar is imputed (provider-reported tokens x list price) because the balance endpoint 403s for want of
billing_read. Nothing is reconciled to an invoice. - 79 new games/month is extrapolated from eight days at 2.6/round; the slot count while merges stall is an open founder decision.
- $0.163/game assumes the catalog hit rate holds. New subject classes force generate calls, and bands 2_3, 7_8 and 9_10 are untouched, so the rate cannot be assumed stable.
- $0.080/call is the production mean for isolated icons. Scene-class or high-detail art measured $0.219-$0.426 in the bake-off, three to five times higher, so any style shift invalidates the line.
- Three refreshes/month is a proposed cap, not a measured rate. The live
queue.md:796 rule implies more. - Whether the subscription price itself counts against the $20 ceiling is undecided. If Go at $8/month is already billed, the API budget is $12/month and even the $12.88 line is over.
Quality risk, ranked by cost saved per unit of risk
- Catalog reuse (best ratio). Saves ~$0.89/round at near-zero risk, because the drawing already passed critic and founder review. The real risk is sameness and blast radius: the turtle shell that "reads as a soccer ball" now appears in #444, #445 and #447 (
rounds.md:990, :1003), so one weak asset propagates. Cheap fix, redraw the one asset once. - Code geometry for derivable elements. $0 marginal, deterministic, founder-ruled on-contract. Hard boundary: geometry only. Pushing it into illustration is the 2026-10-01 regression.
- Capping calls per game at 3. No measured link between drawing count and gate outcome; #443 used 8 drawings and still drew a "page a bit sparse" note. Count is not the quality driver.
- Swap before regenerate in the legibility pre-pass. Low-moderate risk: a swapped drawing fits the slot's composition less exactly than one drawn for it.
- Lowering
reasoning_effort. Unknown saving, and the highest chance the founder notices. Output tokens are path count, and the bake-off's structural audit ties path count to style: 40 paths for the flat corgi against 246 for the doodle probe and 300 for the engraving probe. Cutting reasoning plausibly cuts exactly the density that reads as hand-drawn, on an axis where Arrow already under-delivers (the engraving probe scored 3/5 charm and "only gestures at engraving"). That is the founder's single strongest stated taste ([[feedback_design_taste_high_personality]]). Test blind before adopting. - Simplifying art style to cut output tokens. Biggest per-call saving, three to five times, and the worst idea here: it is the flat/minimal direction the founder has explicitly rejected. Do not offer it as a cost measure.
- Prompt shortening (worst ratio). Bounded at 2-5%, and it carries real quality risk, because the 10-18 framework's law is that Arrow obeys semantic, content-level instructions and ignores global ones. The prompt is the only working control surface, so shortening it trades the one thing that steers the drawing for a rounding error on the bill.
Why this is in the vault
It answers decision 1 carried in escalations.md since 2026-10-09 and re-raised 10-16 ("Monthly Quiver ceiling?") with measured per-call and per-game numbers instead of the stale ~$71 projection, and it tells the founder that the ceiling decision he is being asked for is really two different decisions: a per-game art cap that items 2-5 already satisfy, and a refresh-count cap that depends on the unanswered D1 token and merge-as-hidden-pending-review proposals. It also redirects queue item #396/#428 from "art reuse would be nice" to "art reuse is the only lever that measurably worked," and it kills prompt-shortening and prompt-caching as queue candidates with arithmetic rather than opinion.
Open follow-ups
- Reconcile imputed cost against actual billing. The API key lacks
billing_read, so the balance endpoint 403s and every figure here, including $71 and $29.51, is tokens times list price. Needs a founder-side key-scope change before any of this is auditable. - Which plan is the account on, and does the subscription price count against the $20/mo ceiling? Open since #396 Q5 ("which Quiver plan, for the license field") and mitigation 1 of [[2026-10-14-quiverai-tos-commercial-art-license]]. Founder decision plus account access.
- Does MCP
send_promptdebit weekly app credits rather than the API prepaid balance? If yes, a flat $8/mo Go plan may absorb the whole studio volume and the metering problem dissolves. Evidence is suggestive (09-19insufficient_creditson MCP while REST worked;allowOnDemandCredits: falseongeneratetoday) but undocumented. Testable only after the billing-scope item above. - What does one step down on
reasoning_effortdo to output tokens, and to blind-scored charm and print-readiness? Untested in every run on record, and it is the largest unmeasured output-side lever. Needs a fixed prompt set and blind scoring, not a judgment call. - How many refreshes does
queue.md:796 actually require, and is the rule still right? It is downstream of the merge-as-hidden-pending-review proposal (escalations.md:589, :623) and the D1 token blocker. Founder decision. - How does catalog hit rate decay as games move into new subject classes? At 113 drawings the 10-16 to 10-18 rounds hit near-100%, but bands 2_3, 7_8 and 9_10 are untouched and there is no model of the decay. Measurable from the reuse-versus-generate split once #428 merges.
- What is the blast radius per weak reused drawing? The turtle shell reached three games before anyone flagged it. Measurable from the catalog plus the pull-request reuse graph.
- Is there a vector-generation vendor priced per image rather than per output token? A flat-per-image vendor would make cost insensitive to drawing complexity, which is the direction the founder's taste pushes. Desk-researchable from vendor pricing pages.
Related
- [[2026-10-16-unit-economics-cost-basis]]
- [[2026-10-09-unit-economics-cost-basis]]
- [[2026-10-18-quiver-grok-art-tool-decision-framework]]
- [[2026-10-14-quiverai-tos-commercial-art-license]]
- [[2026-08-31-product-model-rulings]]
- [[2026-09-16-scribble-works-model-cost-ceiling-refresh]]
- [[2026-10-06-shopify-launch-plan]]
- [[2026-10-06-composition-test-implementation-notes]]
- [[2026-10-01-accepted-spec-v0]]
- [[2026-08-31-studio-charter]]
- [[project_rdco_spending_authority]]
- [[feedback_use_real_art_tools]]
- [[feedback_design_taste_high_personality]]
Sources
Vault
01-projects/printables-product/reviews/2026-10-16-unit-economics-cost-basis.md(F1-F6: per-round $7.14/8 rounds, ~$0.34/game, October ~$29.51, ledger stale)01-projects/printables-product/reviews/2026-10-09-unit-economics-cost-basis.md(F3 ~$71 projection, F5 ~$0.97/critic-accepted draft)01-projects/printables-product/2026-08-31-product-model-rulings.md:76-82 (ruling 20 and the 2026-09-02 cost-basis correction)06-reference/research/2026-10-18-quiver-grok-art-tool-decision-framework.md(job-class table, icon-sheet ~3x, REST-over-MCP note, failure modes)06-reference/research/2026-10-14-quiverai-tos-commercial-art-license.md(plan list;arrow-2$4/$20 andarrow-2-telos$6/$30 per 1M)01-projects/coloring-books/2026-10-06-composition-test/tools/implementation-notes.md(REST viaop-service-account.sh; cost = tokens x list price)01-projects/coloring-books/2026-10-06-bakeoff/tools/implementation-notes.md("balance endpoint 403 (key lacks billing_read)")
Primary operational state (read directly 2026-10-19)
~/.claude/state/studio/rounds.md:934-1041 (per-round and per-game Quiver call counts and dollars; $0 rounds on 10-17 and 10-18)~/.claude/state/studio/escalations.md:565-640 (carried ceiling decision; D1 token blocker; merge-as-hidden proposal)~/.claude/state/studio/dispatches/2026-10-1*.md(round briefs: "Quiver art <=$3/game, search content/art + existing game SVGs BEFORE generating"; secrets viaop-service-account.sh)~/.claude/state/studio/eval-ledger/ledger.sqlite(schema pluscost_micros_art= $15.33 over 63 rows, token columns 0/NULL, newest 10-07)RayDataCo/scribble-worksorigin/main46ed046a: 0 files incontent/art/, noart:find, 0 standalone SVGs, art inlined across 136game.html; PR #428 OPEN DRAFT, 100 filescontent/engine-spike/bakeoff-arrow2/scorecard.mdandIMPLEMENTATION-NOTES.md:38-70 (per-call usage, 722 input tokens, 6.8k-21.1k output, reasoning billed as output,max_output_tokensnon-binding, seven accepted body fields)
Vendor / MCP (retrieved 2026-10-19)
mcp__quiverai__list_models— onlyarrow-2(default) andarrow-2-telos;token_usagepricing;generatewithallowOnDemandCredits: falsemcp__quiverai__search_library— returned{"items": []}for a broad Scribble Works query- https://quiver.ai/pricing — Go $8, Plus $20, Pro $40, Enterprise custom; included weekly app usage $5/$15/$40; "Resets every monday UTC (no rollover)"; "Overage via prepaid balance + optional auto-top-up"
- https://docs.quiver.ai/app/pricing — "API calls use the API Platform prepaid balance, not weekly or prepaid App credits"; "QuiverAI offers monthly plans with included weekly App credits"