Clean-room reimplementation track for DEFINE-UK 1.1 - #2
Merged
Conversation
…olver, sector skeletons Implements from the published Model Manual v1.1 only; the pinned upstream run is a numerical oracle (tests/test_oracle.py), never a source. Every equation must carry a manual_ref — enforced by the Equation dataclass. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…n; no upstream licence needed Drops the license request as a roadmap prerequisite — the clean-room build from the manual is the hosting path; upstream code remains an output oracle only. Results surfaces credit DEFINE-UK's authors and state the implementation is independent, matching the suite's adapted-model pattern. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Manual Table 4 vs ONS outturns (macro repo vintages), OBR EFO, DESNZ emissions, NDC path, and IMF/OBR multiplier benchmarks; flags the scenario-invariant multiplier columns in upstream Multiplier_Summary.csv as a probable computation artifact. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…gence recorded and gated Calendar anchored via the population path (run = 1987Q1-2040Q4). tests/test_replication_baseline.py runs against the cached pinned upstream output and skips when absent. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… licence Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
vahid-ahmadi
added a commit
that referenced
this pull request
Aug 12, 2026
An audit of the site flagged the claim that the emissions divergence had been "raised with the authors" as unsupported, because nothing in this repository recorded it. The claim was true — DEFINE_UK_1.1#1 exists and is open — but the auditor was right that it was untraceable from here, which is the same standard applied to every number in VALIDATION.md. Both queries are now recorded with their issue links: #1 on the Table 4 emissions vintage, and #2 on §3.3.2's six untabulated parameters and two undefined variables, which is why the section cannot be simulated forward as published, together with Eq. (84)'s factor-of-three disagreement and Eq. (61) contradicting the stated fuel-price normalisation. The findings were re-verified against the transcribed tables before being posted publicly: all eight symbols genuinely absent, Eq. (84) computing 0.9725 against a tabulated 0.3198, and the fuel-price normalisation corroborated independently by IC_FUELPS and IC_FUELPSR both tabulating 15.75, which is only possible at P_FUEL = 1. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
vahid-ahmadi
added a commit
that referenced
this pull request
Aug 12, 2026
…ical-mean calibration (#12) An independent audit re-derived every pinned discrepancy straight from the manual's printed equations, without importing this repo's sector modules, and re-checked all 661 calibration entries against the manual page each cites. **661/661 transcribed correctly** — the only misses are the two documented `SUBS_GVT`/`SUBS_PS` placeholders the tables genuinely do not contain — and every piece of arithmetic here reproduced. What did not survive was interpretation. **One claim was wrong and is withdrawn.** On Eq. (51) this repo said "the equation is the corroborated one, since Eqs. (49)+(50) imply 0.1598". That evaluates a *lagged* equation — Eq. (49) reads UC_{t-1} — at contemporaneous unit costs. Honouring the lag, the tabulated mu needs +0.91%/quarter unit-cost growth while Eq. (51)'s mu needs **-0.12%, i.e. deflation**; with the manual's initial period at 2022Q4 (Table 6 indexes the deflator at Q4 2022, INF_A = 0.107, quarterly nominal GDP growth 1.019) the evidence points the other way. Table 5 also says alpha1MU is calibrated to a *historical mean*, never to t=L. A first-period jump, not a manual contradiction. **Three more findings are the same class.** alpha_NELEC (Eq. 59) is "the mean over past data" and 2022Q4 was a gas spike — the right direction and roughly the right size for the 60% gap; alpha_DIVPPS (Eq. 94) and eta_PSB (Eq. 111) are likewise past-data means. The gaps are real and matter operationally, but none is the manual disagreeing with itself, and they are no longer presented as if they were. **Two mislocations.** Eq. (107)'s sign contradiction is not one of "four in section 3.3.2": the identical alpha*GO rule is printed at Eqs. (177), (231) and (284), and Table 6 tabulates all four transfers negative, out by 1.7-4.6x. And Eqs. (27), (104), (105), (132), (133) are one finding, not five — Table 6 deflates its entire capital block at 1.031 where the equations prescribe P_P = 1.035, which looks like a national-accounts investment deflator standing in for the production deflator. **Eq. (84) gains a likelier explanation.** (1+1.228)x0.4365 = 0.9725 against a tabulated 0.3198 implies a *negative* mark-up, which no historical mean can produce. Section 5 calibrates an electricity price rule section 3 never prints: Table 5's `t_ELECswitch` is the quarter of "the switch in the electricity price long-run formation rule" and Table 6's `P_ELECLR` is a "long run electricity price"; `P_ELECLR` occurs once in the whole PDF and neither appears in any printed equation. The 3.04x is most likely a missing equation, recorded as such. The findings that carry weight survive intact and better supported: Eq. (52) (Table 5 says alpha0WS was calibrated so the equation equals the observed value at t=L; it does not, and 0.6233 is exactly W/GDP), Eq. (61), Eq. (70), Eq. (108), Eq. (131), Eq. (136), the three sign contradictions, the LENDM residual (the seven model-determined entries sum to exactly 5.44 = -DIVN_ROW while national-accounts LEND sums to -0.001), and all of issue #2 — six parameters and two variables that occur in the body and in neither table. **Both filed upstream issues hold in full**, and the wrong claim never left this file. Also: the 1.78 green multiplier is recorded with the three properties that make it upper-leaning rather than comparable — a cumulative ratio to 2040 set against short-horizon IMF/OBR multipliers, an endogenous `SPEND_GVT` denominator of which investment is only 12.8% and social benefits fall when unemployment does, and a demand-led closure with Kaldor-Verdoorn productivity where >1 over fifteen years is what the model implies. `test_oracle.py` keyed its skip on `Rscript` being present, which is the wrong condition — the comparison needs upstream *outputs* — and it is rekeyed on the cached run, with a test that fails the moment section 3.3.3 lands so the outstanding gate stays visible in a green suite. 229 -> 236 passed; no test removed or weakened. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Starts roadmap track for hosting DEFINE-UK regardless of the upstream licence (parallels the license request, addresses #1's hosting path).
Equation/Registrywith requiredmanual_refand the manual's own equation taxonomy (identity/behavioural/calibrated); per-period Gauss–Seidel SFC solver that raises on non-convergence; ten sector modules mapped 1:1 to manual §3.1–§3.4.7 (no equations yet — they land per milestone).8 passed, 1 skipped locally (Python 3.13).
🤖 Generated with Claude Code