utbyggingsplan_trinnvis follows chapter 17.1: the grid balance of the
first stage, the payment condition for a module's power plants, S49,
and in T-K the two stocks of chips and the queue for replacements.
tidslinje gives MW IT in operation per type of power (S50). S48-S54
are pure functions in ligninger.py, and Inndata carries the new
quantities, the scenarios for the loss of demand and the paths of
u_alt. lag_plan returns a Trinnvisplan for T-F and T-K; regn_program
refuses T until it is taken through the cash flow.
Tested on made-up numbers only (59 new tests, hand-computed build-outs).
Nothing is computed for the hypothesis.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Clarifications 70 to 75 in chapter 16: nuclear power before new grid is
refused, the grid balance in years with an export stop, the payment
condition for a module's power plants, two sensitivities of 17.8, the
paths for the loss of demand and for u_alt, and the timeline (S50).
None of them changes a number for values within the ranges of the
parameter file. The parameter file and the pre-registration are unchanged.
Part 10 of the review log: the project's code for the build-out of T-F
and T-K and for S48-S54 against the independent reader's code, on
made-up sets only: 42 000 build-outs and 20 000 sets for S51-S54
without a difference. Nothing is computed for the hypothesis.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The loader knows the new checksum and the units of the new quantities
(g CO2 per kWh, cubic metres per MWh). 10 000 draws of 91 quantities with the
same seed; the four new ones are the last columns. All columns have new
numbers, because the random numbers are laid out row by row. The stream for
yearly events is byte for byte the same; only its description records the new
checksum. Tests for the counts, the new quantities and the checksum in the
pre-registration. README, CHANGELOG and CLAUDE.md updated.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
One deviation, before anything is computed. modell/parametere.json version 1.1,
SHA-256 4b8809bc56abdd8b2872b89eecffd8453199eef627110276dd10aecea8307f16.
- Thirteen new quantities for chapter 17 (module size, lead time and cap of
the first stage, the response to lower demand, the two demand scenarios,
the emissions of the power the compute would otherwise use, cooling water).
- Seven new sensitivities, and the three lines clarifications 53, 64 and 65
announced (the median in 2025 kroner, the rate for the state's cash flow,
the settlement priced with the cost of equity).
- None of the 124 frozen parameters has a new value, range or distribution.
Four things changed from the candidates after the source check (chapter 17.8
and part 9 of the review log): the seawater range is 119-168, zero freshwater
use is an open assumption, the largest data centre case in the register is
180 MW, and 2.4 years is one dated case.
BREAKING CHANGE: the loader refuses the file until its checksum constant and
the stored draws are updated; that is the next commit.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The 30 entries chapter 17 of the specification cites that the project had not
looked up itself are now checked against the primary source: 23 in the files
the researchers and controllers downloaded, 2 in pages fetched now, 1 through
a search tool, and 4 in Statnett's register of grid connection cases, read in
the browser filter by filter. Each entry says what was looked up and what was
not. K26-006 also records which scenario its rows belong to.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Approved 2026-11-12, after the choices were put forward with the independent
reader's findings and before anything is computed. The parameter file is
unchanged.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Chapter 17 of the specification (S48-S54): solution T, "stepwise", in two
versions (T-F and T-K), the timeline, a threshold and two labelled scenarios
for compute on the users' own machines, emissions with a deduction, and
cooling water. The pre-registration gets T-F and T-K in the table of
alternatives, in M7 and in clarifications 1, 3, 10 and 11, new clarifications
13-16 and a dated entry. An independent reader coded the chapter from the text
alone in three rounds; the findings and what was done are in part 8 of the
review log.
Nothing is computed. The parameter file is unchanged (checksum 4582835...f6cc);
the new quantities are candidates until the deviation from the freeze. The
eight choices in 17.7 are put to the project owner and not yet approved.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The operator approved the four recommendations after the source work in
areas 22-26: a new stepwise solution T in two variants measured on the same
terms as A, B and C (A must also beat T in part II), no gas power, on-device
AI tested as a threshold and two marked scenarios without changing the
verdict that counts, and the basis for avoided emissions and cooling water.
Decided before anything is computed. The alternatives table, the wording of
M7 and how T is computed follow with the specification. The frozen
parameters are unchanged.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Third round of source work, before anything is computed: what can be in
operation within a few years on the existing grid (22), power sources with
short lead time (23), stepwise build-up and switching power solution over
time (24), AI compute on users' own machines (25), and the two quantities
the model lacked parameters for: avoided emissions and cooling water (26).
188 entries, one researcher and one independent verifier per area:
114 confirmed, 52 corrected, 22 added by the verifiers, none unconfirmed.
22 load-bearing entries were checked a third time against the primary
source and carry an "Egenkontroll" line.
The model, the frozen parameters and the pre-registration are unchanged.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The operator approved the corrections to the comparison of ownership forms
on 2026-11-12, before anything is computed, and decided that the addition to
the direction is worked in before the run over the draws is coded.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
eierskap.py: the residual value and the sale at the final year, the levies,
the private owners' cash flow and the settlement over the draws, the present
value with rho_D without the debt flows, the required return before tax, and
equal values in the ranking. Also the median of the petroleum cash flow in
2025 kroner, the state's cash flow next to M6, the shares shown next to each
verdict, and the sources' figures for M5 (i).
Made-up numbers only: 424 tests pass, 7 documented misses. The project's
code and the independent reader's code agree to within one krone on 1 500
made-up paths.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The operator's four answers of 2026-11-12: clarifications 41-60 approved,
clarifications 55-59 entered in the pre-registration as clarifications 9-12,
the addition to the direction entered, and the announced deviation from the
parameter freeze recorded. The result of the large comparison of part 5c is
in part 6 of the review log.
Clarifications 62-67 correct the weaknesses of clarification 61 in the
comparison of ownership forms: residual value at the final year, the levies,
the private owners, the rate the state's cash flow is valued with, and the
cost of capital before tax. An independent reader coded them from the text in
three rounds (20, 5 and 2 findings); part 7 of the review log. Clarification
68 holds the sources' figures for M5 (i), checked against the downloaded
primary sources, and 69 the figures shown next to each verdict.
The parameter file is unchanged. Nothing is computed for the hypothesis.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The accident cost stays in S11 and S31 when the additionality rule binds;
the peak year of M6 is taken among operating years; tax depreciation and
carried losses lose real value with price growth; the special tax deducts
the corporate tax, has a return allowance and carries a negative basis with
interest, so it only meets return above r_e; ownership forms are compared
only with forms that pass the EEA assessment; the overall verdict needs a
negative median to fell a solution and takes whether conditions can be
named; M5 does not hold when both (i) and (ii) are broken; a non-positive
price without the programme is an error in M3. Made-up numbers only.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Clarifications 41-61 in chapter 16 and part 6 of the review log. The
independent readers found 60 + 17 + 5 + 1 + 0 points. Weaknesses in the
comparison of ownership forms that are not corrected are listed in
clarification 61. The parameter file and the pre-registration are unchanged;
no result for the hypothesis has been computed.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Chapters 7 to 11 of the specification, on the made-up set only:
- nivaaisert: the power cost per facility, year and vintage on levelised
form (S9, S11 with clarifications 2, 11, 13 and 15), S15, S19, S27, S42
- eierskap: the business cash flow with every asset booked, tax, S28, S29
and S30 for the four ownership forms, the criteria over the draws
- skala: S31 year by year, the first year with stable operation, M6
- miljo: S32 year by year, the rules of M5 (i) and (ii)
- dommer: S37, the verdict per target, the step of the ladder, the overall
verdicts; tested on invented outcomes
- kontantstrom: each asset is recorded as it is paid, and the price of
compute exposes the competitor's cost and the markup
This is an intermediate step. The readings are not settled: an independent
reader is coding the same from the specification alone, a first comparison
agrees on the levelised cost, S19, S42, the cash flows, M3 and M4 and
differs on depreciation at the final year, tax, debt and two definitions.
Chapter 16 gets the clarifications before any result is computed.
Nothing is computed for the hypothesis; the parameter file is unchanged.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The project owner answered the three questions from part 5b on 2026-11-12
("Ja til dine anbefalinger"), before any result is computed:
- chips are a stock from the solution's first possible operating year, not
from the decision year (clarification 25 rewritten; the old reading stays
as a switch that is not part of the verdicts)
- renewable power in C enters operation the same year as the facility
(clarification 24 stands)
- the other choices in clarifications 21-40 are approved
The default in the code is flipped, seven tests that assumed a stock from
2027 are redone, and the comparison with the independent reader's code is
rerun with its switch set the same way: same numbers in all 9 471 paths
that can be built. The parameter file is unchanged.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
- utbygging.py: when each facility starts and which vintages of IT hardware it
gets (S22, S33), with chips as a stock, replacement before new facilities and
the export rule.
- kraftsystem.py: the fitted price lines and the power and capacity balances
year by year (S23, S24, S25, S38).
- kontantstrom.py: the yearly cash flows per facility, the additionality rule
by iteration, the residual value and the net present value for Norway (S26,
S34, S35, S40, S45), for the solutions A, B, B', C and R.
- hendelser.py and data/aarshendelser.npy: the stream of random numbers for the
export rule, one uniform number per draw and year 2028-2100, from its own seed.
- inndata.py: one set of values in the units of the equations.
- ligninger.py: one function per equation for chapters 4 to 6.
Tested on a made-up set of values only (tests/conftest.py); no result for the
hypothesis has been computed. 271 tests pass. Seven are documented misses
against a source: the five from before, and the price lines for NO3 and NO4,
which miss one of NVE's twelve scenarios by 6.0 and 7.7 per cent.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
A second independent reader coded the build-out, the power system and the cash
flow (S12, S22-S26, S33-S35, S38-S41, S45 and the sharing of S47) for all five
power solutions from the specification alone. It found 28 ambiguities, and ten
more when it coded again from the clarifications. Clarifications 21-40 in
chapter 16 settle them; clarification 29 (the nuclear units in B) was rewritten
twice after the reader showed that the first two rules were wrong. With the
final text the two implementations agree on 10 000 made-up build-out paths.
The clarifications that are choices are marked and put to the project owner.
Whether chips are counted from the decision year or from the first possible
operating year is left open until the owner has answered. The parameter file
and the pre-registration are unchanged, and no result has been computed.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Clarifications 5, 10 and 11 in chapter 16 are choices, not only readings.
The project owner approved them on 2026-11-12, before any result, and decided
that the yearly production loss NVE assumes for wind is not added: the
deviation from NVE is to be stated in the report and the freeze stands.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
One pure function per equation for chapters 1 to 3 of the specification
(S1-S21, S36, S43-S47), a strict reader for the frozen parameter file with
checksum, validation and unit conversion, and the 10 000 Monte Carlo draws of
the 87 drawn quantities, stored in data/ and reproducible bit for bit.
Tests: unit tests and edge cases per equation, the reader, the draws, and
known answers from Argonne, NOU 2026: 4, Epoch and NVE with the sources' own
assumptions. Five comparisons miss the 5 % tolerance and are kept as strict
expected failures with the measured deviation. No result for the hypothesis
is computed. ruff ALL and mypy strict pass.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
An independent reader coded the key equations from the specification alone
and listed 19 ambiguities (4 blocking). Chapter 16 of the specification now
says how the text is to be read, dated 2026-11-12, after the parameter freeze
and before any result. Equations, equation numbers and the parameter file are
unchanged. The review log records each finding, what was done, and the
outcome of the known-answer tests.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
K12-001: the method article behind the Argonne study (Hanna et al. 2024) read
in text; what the test needs is quoted, and what could not be read is stated.
K16-011: the committee's spreadsheet reproduces 132.2 øre/kWh but gives 222.1,
not 219, for the high scenario. K11-004: NVE's investment figure for hydro
already includes interest during construction.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The project owner approved the eight proposed clarifications of the
pre-registered criteria on 2026-11-12, as recommended, after the
parameter freeze and before any code or result exists. They are entered
under «Presiseringer» with a dated entry and the reason for each under
«Endringer». Three wordings in the criteria text are aligned with them.
The specification now refers to them as clarifications instead of
proposals, and the rule for conditions uses the 10th to 90th percentile.
Equations and parameters are unchanged; modell/parametere.json keeps its
checksum.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Commit 01b27fa, SHA-256 4582835…f6cc. The criteria are unchanged.
Eight proposed clarifications are with the project owner and are not
entered.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
124 parameters for the model in modell/spesifikasjon.md (draft 3), all
amounts in real 2025 kroner, one distribution per parameter, each with
the source bank entries it rests on. 80 follow from confirmed entries,
1 is derived from them, 35 are open assumptions with a stated range and
8 are set-up choices. The file also fixes the conversion rule, the
Monte Carlo set-up and the list of sensitivities.
Frozen before anything is computed and before any code exists. The file
is not edited after this commit; a change is a deviation with a new
checksum and a dated entry in the pre-registration.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Draft 3 has 47 equations. Changes follow the pre-read of draft 2
(44 findings), two independent reviews of draft 3 (55 and 22 findings)
and an independent control of the corrections (39 findings). Each
finding was checked by the project before it was corrected; the log
says what was done with each.
Main changes: all parameters in 2025 kroner; explicit cash flows per
solution; the price of compute set per MW sold, so that unsold capacity
is not charged to Norway alone; power plants on the grid are never
credited a surplus (S45); the reading of the Norwegian nuclear cost
range is drawn (S46); dimensioning, build-out and residual value are
specified. Eight proposed clarifications of the pre-registration are
listed in chapter 15 and are not entered there.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Five entries get the figures the project itself read in the primary
source and that parameters now rest on: the remaining rows of table
5.17 in the consultant report (K16-002), the assumptions sheet of the
commission's spreadsheet (K16-012), all twelve power balances in NVE's
2035 scenarios (K21-087), the intermediate years of the carbon price
sensitivity paths (K09-014) and table 6-1 of the Argonne study (K12-001).
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Second round of the source bank, before the parameter freeze: eight
tracks for the open points 10-17 in the specification (areas 13-20),
the three unconfirmed entries in area 11 and the outdated entries
(area 21). One researcher and one independent verifier per track.
286 new entries (157 confirmed, 93 corrected, 36 added by verifiers,
1 not confirmed). 57 verified corrections applied to existing entries,
each with the previous wording kept in the control line. 53 entries
checked by the project itself against the primary source (Egenkontroll).
README and the cross-area review updated.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Approved by the project owner after two independent reviews of the
model specification, before anything is computed and before any code
exists: overall verdict, M2 against the price of compute, verdict per
price scenario with fixed build-out paths, the M4 requirement, M3 in
normal and dry years, the M6 numerator and year, a symmetric M7
threshold, a grid-purchase reference run, and the limit for state risk.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Draft 1 was read by two independent reviewers (a subject reviewer and a
devil's advocate). Draft 2 corrects the equations they found wrong
(levelised price, token cost unit, welfare sign, double counting) and
models off-grid operation with unit size limits and reserve units.
Findings that touch the pre-registered criteria are listed as open
decisions, not applied.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Pre-registered criteria (M1-M7), approved before anything is computed,
and a source bank of 662 entries in twelve areas, each checked against
its primary source by an independent verifier.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>