Touchstone Assay
Dated SIU — SIU-2026a
$0.0089
13 Sep 2026 provisional
Correction notice — published after this print, disclosing a fact that does not change dated_siu:

3 corrections: 2 precision, 1 pricing

  • 13 Sep 2026 — This print was published (2026-09-13, automated daily run) before two real bugs in this project's own code were fixed (commit 48ba69c), both affecting its own cost_of_production_usd. (1) RunRecord.usage.cached_input was captured by every provider adapter but never priced by this project's cost formula -- a real, unconditional gap, not specific to any one provider. For this print specifically: gemini-3.1-pro-preview: 84506 cached tokens, $0.016901; grok-4.6: 7296 cached tokens, $0.000004 -- together understating the published cost_of_production_usd ($1.4109) by $0.016905. This is not a hypothetical cache-adoption policy (see docs/methodology.md's sensitivity block, cachePolicyVariant, which is unrelated) -- this cache use actually happened and was actually billed. (2) Separately, and larger for gemini-3.1-pro-preview specifically: the Google adapter discarded a real, separately-billed API call's usage whenever Gemini's mandatory reasoning consumed the whole completion budget and triggered an internal retry with a bigger one -- only the retry's usage was recorded, not the first, truncated call's. This fires on nearly every gemini-3.1-pro-preview call. Its real input-token count and full raw response were never saved for this print's own run records (captured only after the fix), so this piece cannot be reconstructed precisely, unlike (1) above. Neither bug is corrected on this print's own signed cost_of_production_usd or dated_siu -- a published print's signed body is never edited -- so this print remains status: "provisional" permanently. See docs/methodology.md's Index governance for the full explanation.
  • 25 Sep 2026 — dated_siu previously rounded to a fixed 4 decimal places rather than 4 significant figures — found live and fixed the same day, 2026-09-25 (docs/methodology.md's Rounding section). At this print's price level that was too coarse to show real day-to-day movement. This print's signed body is never edited — the published figure (0.0089) stands. Recomputed from this print's own published basket_costs and weights (Σ weight × basket_cost, rounded under the corrected 4-significant-figure rule, reproducible by anyone from this print's own published figures alone), the real value is 0.008856.
  • 25 Sep 2026 — Correction to this print's own rounding correction note above: that note's phrase "the real value is X" overstated what the recomputation shows. It restores rounding precision only — the precision a fixed 4-decimal-place rounding rule removed, recovered by recomputing Σ weight × basket_cost from this print's own published figures. It is not a claim that the recomputed figure is this print's true economic value. This print's own basket_costs are also affected by the reasoning-token and/or cached-input pricing correction disclosed elsewhere in these notes — usage that was under-priced before that fix landed — so the recomputed rounding figure is the unrounded version of a value already known to need that separate correction too, which this recomputation does not make. See docs/methodology.md's Rounding section for the general statement of this limit.
— 0.0% since 12 Sep 2026 — Unchanged from the previous print.
Equal weighting, disclosed: basket costs span 35.9× across the 10 qualifying constituents. 4 of 10 constituents (40%) sit at or above the mean basket cost and together account for 79% of the summed basket cost Dated SIU is computed from — equal weighting still gives each of them the same 1÷10 nominal share. Dated SIU therefore tracks the more expensive end of the basket more closely than a demand-weighted index would. See the methodology's Weighting section, and Frontier SIU / Commodity SIU, the tier prints published alongside this one for exactly this reason.

Exchange-rate table

ModelUSD / SIUSpread to indexSIU per $1
deepseek-v3.2 $0.0018 -79% 550.2
llama-3.3-70b-deepinfra $0.0007 -92% 1382.3
mistral-small-3.2-24b-instruct $0.0007 -93% 1530.9
qwen-2.5-72b-instruct $0.0025 -72% 398.5
claude-sonnet-5 $0.0234 +164% 42.7
gpt-5.1 $0.0091 +3% 110.0
gemini-3.1-pro-preview $0.0204 +130% 49.0
grok-4.6 $0.0170 +92% 58.7
claude-haiku-4-5 $0.0077 -13% 130.3
gpt-5.4-mini $0.0053 -40% 189.2

Floor & market spread

Not yet published — the floor needs a measured hardware-rental rate for the reference configuration, which this print does not yet include.

Sensitivity

The print recomputed under alternative cache and batch-discount policies, so the effect of those policy choices is stated rather than hidden inside one number.

Policy variantDated SIUDeltaApplies to
cache-enabled: 40% of T2 input at 10% of input price $0.0063 -28.9% deepseek-v3.2, llama-3.3-70b-deepinfra, mistral-small-3.2-24b-instruct, qwen-2.5-72b-instruct, claude-sonnet-5, gpt-5.1, gemini-3.1-pro-preview, grok-4.6, claude-haiku-4-5, gpt-5.4-mini
batch-discount: 50% off input and output $0.0044 -50.0% deepseek-v3.2, llama-3.3-70b-deepinfra, mistral-small-3.2-24b-instruct, qwen-2.5-72b-instruct, claude-sonnet-5, gpt-5.1, gemini-3.1-pro-preview, grok-4.6, claude-haiku-4-5, gpt-5.4-mini

Methodology, signature & anchor

Methodology
v0-draft
Cost of production
$1.4109
Signature
0xdc7cdad8…cfd0bb24
Public key
0x030c598e…e25cb299
Anchor tx
0xdfdee369…5212153f (base-sepolia)
Raw runs
https://github.com/petrkrulis2022/dated-siu-inference/tree/main/data/runs/2026-09-13

Dated SIU is an assessment of qualifying inference work under this basket version, model/provider configuration and quality gate — not a prediction of production agent costs, which vary with workflow design, tool reliability, retry policy and context handling.

Verify this yourself

Don't take this page's word for it — every value below is checkable independently.

Body hash
0xff385e4d…7fae8aa5
Recovered signer
0x284ff2f8…d48f845a — matches the TouchstoneAttestation publisher below
TouchstoneAttestation
0xF6070179…93297190 (base-sepolia)
  1. Recompute this print's body hash from the published JSON (JCS-canonicalised, signature fields excluded) and confirm it matches the body hash shown above.
  2. Open the contract link above on Blockscout, go to "Read Contract", call publisher(), and confirm it matches the recovered signer shown above.
  3. On the same "Read Contract" tab, call postedAt(bytes32) with the body hash shown above and confirm it returns a non-zero timestamp.

All prints