ECO-05: the scenario register frozen before any result (R = ETC's trailing-year miner revenue USD 48.7 M from the public page's own figures, 8 October 2026, IGN 0.063 implied; the P12 worlds verbatim; the pass envelope verbatim; the competitive core declared; the sustainability rule declared; BLOCKED items named); the 5090 mapping control into the floor file (the multiply-shift term zero at the knee)

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

docs/analysis/class-v6/eco-05-scenarios.md

@@ -0,0 +1,116 @@
+# ECO-05 scenario register: stress success, contraction and cheap electricity (frozen before any result)
+
+Case ECO-05, registry row `docs/plans/igneum-2.0-test-registry.json` suite ECO, gate G4, profile P12 ("Five-year
+coexistence envelope"), priority GATE, owner the economics lead and an independent reviewer, status NOT RUN at the
+time of this register. Written 8 October 2026, 18:4x UK, branch `class-v6-floor-sram`, floor lane 3, on the research
+lane's word of 18:3x UK under the approved Test and Acceptance Standard. **This file is frozen before any ECO-05
+result is computed; the result cube (`eco-05-results.md`) cites it and changes nothing in it. Unknown inputs are
+BLOCKED, never guessed; nothing is weakened to pass.** The model is the coexistence model
+(`docs/analysis/class-v6/coexistence-model.md`), its reference population (`reference-population.md`) and the
+profitability surface (`floor/sram-and-floor.md` section 4.4); every row of those is modelled unless labelled measured.
+The founder is not named.
+
+## 1. The sourced revenue reference R
+
+| Item | Value | Source and date | Label |
+|---|---|---|---|
+| The reference chain | Ethereum Classic (ETC): a public GPU-mined chain (Etchash, GPU and ASIC mined, the GPU share not published; the standard's default when no better source exists) | | the standard's default |
+| ETC block reward | 2.5600 ETC per block | miningboard.com, https://miningboard.com/mining/ethereum-classic/halving, read 8 October 2026 (the page's data "15 minutes ago" at 18:3x UK); the page states a constant reward with no halving schedule, which conflicts with ETC's published 5M20 reduction schedule (a 20 percent cut every 5,000,000 blocks); the page's figure is carried as read and the conflict is a limitation | claimed (the page) |
+| Block time | 13.0 s | the same page | claimed |
+| ETC price | USD 7.83 (down 6.45 percent in 24 hours) | the same page, the same reading | claimed, point in time |
+| Network hashrate | 143.42 TH/s | the same page | claimed |
+| Blocks a year | 31,557,600 / 13.0 = 2,427,508 | arithmetic | modelled |
+| ETC minted a year | 2,427,508 x 2.56 = 6,214,420 ETC | arithmetic | modelled |
+| **R, the trailing-year miner revenue** | **USD 48.7 M a year** (6,214,420 x 7.83); USD 133,000 a day | arithmetic on the page's figures | modelled |
+| R's range | USD 40 M to 55 M a year on the year's price range (USD 6.51 in August 2026 to 8.84 in May 2026, the same site's monthly pages via the search summary of 8 October 2026) | the search summary; a point-in-time price times a year's emission is not a trailing sum, which is a limitation | approximate |
+| The IGN price that the spec's year-1 emission implies for R | 0.77 B IGN to miners in year 1 (3,168,808,781 base units per DAA second, 10^8 per IGN, 80 percent to miners, the 30-day ramp): **USD 0.063 per IGN** at R; 0.016 at 0.25R, 0.25 at 4R, 0.63 at 10R | the spec's constant; arithmetic | modelled; never an assumed appreciation |
+| The rejected alternative | Kaspa: ASIC-mined since 2023 (not a GPU reference); its public page read the same way (https://miningboard.com/mining/kaspa/halving, 8 October 2026: 93 KAS per 1.0 s block, USD 0.04, 340 PH/s) gives USD 117 M a year, carried here only as a cross-check of R's order of magnitude | claimed; rejected as a GPU reference |
+
+## 2. The mandatory worlds (the standard's P12 list, verbatim as axes)
+
+| Axis | Values | Note |
+|---|---|---|
+| Revenue | 0.25R, R, 4R, 10R (USD 12.2 M, 48.7 M, 195 M, 487 M a year of miner revenue; IGN 0.016, 0.063, 0.25, 0.63) | the year-1 emission; the halvings in later years are the five-year run's, not this grid's |
+| Electricity | USD 0.03, 0.10, 0.25, 0.40 per kWh | the GPU side's tariff; the matched-tariff test puts the specialist at the same tariff; the heterogeneous test (section 4) puts the specialist at 0.03 against each GPU tariff |
+| Productive life | 1, 3, 5 years | the specialist's; the GPU entrant's horizon is 2 years with resale as the reference population's method |
+| Development cost | zero (sunk, the mandatory case), USD 20 M, USD 75 M | amortised over the specialist's fleet and life; the fleet is a third of the world's GPU-equilibrium hash (declared here; the whole-chain case is reported beside it as an adversarial combination) |
+| Business model | private mining (the operator self-mining at the unit ticket), hardware sales (the buyer pays twice the hardware term; the manufacturer's half) | |
+| External proving demand | none, normal (USD 2,000 a day at the base fee, spec 05's launch grid, 90 percent to provers), spiking (x10) | a second income for the GPU cohort's proving-capable classes (16 GB and up; the 12 GB tier a measured zero), none for any hash engine |
+| Grid size | 4 x 4 x 3 x 3 x 2 x 3 = 864 worlds per specialist | |
+
+## 3. The specialists and the honest cohort
+
+| Side | Rows | Energy and ticket | Label |
+|---|---|---|---|
+| Specialist 1 | the N2 SRAM die with the shadow core | 2.3x per joule vs the 5090 at its lock node-for-node (3.1x a node ahead), the reconciled unit ticket USD 1.0 per MH/s with the band 0.5 to 1.6 as adversarial ends | modelled (the adversary lane's reconciled rows; the chip model's energy) |
+| Specialist 2 | the stored-half hybrid board | 1.93x, USD 2.80 per MH/s | modelled |
+| Specialist 3 | the complete GDDR7 machine | 1.5x, USD 4.84 per MH/s | modelled |
+| The honest cohort | the reference population's 19 classes (block 1 the plan's eight, block 2 the eleven others), each as the existing owner and the new entrant at MSRP (the matched-tariff new-entry test) and at street | measured where the population says so, modelled knees elsewhere | per cell |
+| The competitive core | RTX 5090, RTX 5080, RTX 4090, RTX 3090, RX 9070 XT, RX 7600 8 GB, Intel Arc B580: seven configurations across three advertised vendors and five discrete generations (Blackwell, Ada, Ampere, RDNA 4, RDNA 3, Xe2) | declared before any result | the standard's P02 asks at least six across every advertised vendor and two generations |
+| The entry cohort | the 19 classes weighted by the reference population's installed-base counts (approximate) | | |
+
+## 4. The measure and the pass envelope (verbatim)
+
+- Cost per accepted MH/s-hour as the reference population defines it (annualised hardware plus power plus hosting,
+  failures and fees, over 97 percent accepted work; the specialist with farm hosting USD 0.02 per kWh on top and its
+  development cost amortised as section 2 says).
+- The cell ratio: the GPU new entrant's total cost per accepted work at MSRP over the specialist's total cost per
+  accepted work, at the SAME tariff (matched tariff). The heterogeneous-tariff cells (the specialist at 0.03 against
+  each GPU tariff) are published beside them and are not the envelope's test.
+- **The pass envelope, in every mandatory sustainable world: the matched-tariff median over the cohort at most 1.5;
+  the 90th percentile at most 1.75; no competitive-core cell above 2.0.** Published for every cell.
+- Sustainable worlds (declared before running, section 5): a world has enough funded demand for rational ongoing
+  service when, at the GPU side's own equilibrium (revenue per MH/s-hour set by the per-class supply curve with the
+  installed base as the cap), at least three purchasable GPU configurations across at least two advertised vendors
+  have positive modelled new-entry economics at MSRP, and at least 75 percent of the entry cohort by installed count
+  has positive marginal operating economics (owner cost below revenue per accepted unit). Worlds that fail that test
+  are collapse worlds and are retained as explicit safety and exit tests, never as profitable successes.
+- Not imposed anywhere: GPU market share, specialist production limits, token appreciation, full research-cost
+  recovery, automatic chip death. The 180-day rotation enters only as the 1-year life's reason, not as a death.
+
+## 5. The declaration of sustainable worlds, before running
+
+By the rule in section 4, read off the reference population's owner and entrant rows at each tariff and the
+per-class equilibrium (the coexistence model's section 9 rule), without any specialist present:
+
+| Revenue | 0.03 | 0.10 | 0.25 | 0.40 |
+|---|---|---|---|---|
+| 0.25R (USD 12.2 M a year) | to be read by the driver and published; expected collapse (the equilibrium hash under 2 TH/s, fewer than three classes with positive entry) | the same | collapse expected | collapse expected |
+| R (48.7 M) | sustainable expected | sustainable expected | to be read | collapse expected |
+| 4R (195 M) | sustainable expected | sustainable expected | sustainable expected | to be read |
+| 10R (487 M) | sustainable expected | sustainable expected | sustainable expected | sustainable expected |
+
+The driver reads every cell and publishes the declaration as found; an expectation above that the driver contradicts
+is reported as such, not revised. The sustainability test is applied with no specialist present (the GPU side's own
+economics), so that a world is not declared unsustainable because a specialist exists.
+
+## 6. The adversarial combinations (the independent reviewer may add; none removed)
+
+1. The die at the band's cheap end (USD 0.5 per MH/s) and a node ahead (3.1x), with development zero and private
+   mining, at the GPU's dearest tariff (0.40) and the specialist's cheapest (0.03): the heterogeneous worst case.
+2. The die taking the whole chain (development amortised over 100 percent of the equilibrium hash) rather than a third.
+3. Hardware sales at zero development: the manufacturer's half with no project to recover.
+4. The GPU entrant at street rather than MSRP (the market's price, not the list).
+5. A GPU generation at 1.5x per joule at the same price (the coexistence model's year-3 step) against the specialist a
+   node ahead.
+6. The spiking proving demand credited to the GPU side at the measured 5.5 percent proving efficiency rather than the
+   assumed 0.5.
+
+## 7. Outputs the results file must carry
+
+The full cube (every cell: world, specialist, median, p90, max core cell, the three envelope verdicts, the
+sustainability verdict and its two counts), boundary tables (the tariff and revenue at which each specialist crosses
+1.5, 1.75 and 2.0), a failed-world explanation per failing cell (which cohort classes carry the median, which core
+cell breaks 2.0 and by what term, hardware or power), the adversarial combinations beside the grid, and the model's
+uncertainty (the die's ticket band, the modelled knees, the approximate prices) stated as a range on the median.
+
+## 8. What is BLOCKED
+
+- The ETC page's "no halving schedule" against ETC's published 5M20 schedule: R is carried at the page's reward and
+  flagged; a second explorer read would settle it and is owed before G4.
+- The GPU share of ETC's hashrate (Etchash is ASIC-mined in part): unknown; R is the chain's whole miner revenue.
+- The Arc B580's watts (estimated, about 110 W) and the RX 7600's knee (estimated by the 9070 XT's grid): those two
+  core cells carry an estimated energy and are published with that label; the Intel lane's meter and the 7600's own
+  grid replace them.
+- No chip has been measured; every specialist row is modelled on the chip model's method and the adversary lane's
+  placed rows.

docs/analysis/class-v6/floor/sram-and-floor.md

@@ -341,7 +341,12 @@ GiB dataset FITS on the card's 8,176 MiB (6,732 MiB max alloc, the ds55 pack bui
 15.46 / 15.34 / 15.35 MH/s at 1 / 2 / 4 / 5.5 GiB (-1.9 / -2.6 / -2.5 percent), fingerprints equal to the 5090's at
 every size. The 8 GB tier pays under 3 percent of rate for the 5.5 GiB step where the locked 5090 pays 15: a 16-CU
 card is latency-bound by its own lane count at every size and the footprint never reaches its limit; with the rate
-flat its energy per hash moves with its watts alone (113 W at stock, today's card-in reading, approximate).** So the
+flat its energy per hash moves with its watts alone (113 W at stock, today's card-in reading, approximate).** **The mapping control landed (PC 1, the same program byte for byte, 18:3x UK): at the 1,300 MHz lock the 4 GiB
+mask path reads 1.780 microjoules and the 4 GiB multiply-shift path (2^30 minus 2^16 words) 1.782, +0.1 percent, so
+the mapping term is zero at the knee and the dataset size is the whole cost; against this card's 1 GiB rows (lock
+1.593, stock 2.248) the 4 GiB step is +11.7 percent at the knee and +0.7 at stock, the 5.5 GiB step +13.4 and
++4.3. Under the standard's P03 (no more than 5 percent of joules per accepted work against the paired tuned
+baseline) no step past 1 GiB passes at the knee on a 5090; the decision is the research lane's.** So the
 v6 epoch's 5.5 GiB floor holds the 8 GB tier in fact (section 3.2's worst-case "OUT" rested on the 75 percent rule
 with the cache resident; the measured allocation is 82 percent of the card with the cache freed, and the card mines
 at 97 percent of its 1 GiB rate), against the chip's ticket rising