junte/Model
1
0
Fork 0
mirror of https://github.com/Hestia-Homes/Model.git synced 2026-06-30 13:10:47 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

docs(audit): record stale-default-plan expectation + default-plan-only rule 🟪

Browse source 
Portfolio-796/scenario-1268 audit found the bulk of MEDIUM/HIGH anomalies
(already-meets-goal-with-works, zero-works-post-differs, plan-score-below-
baseline, and the non-fuel-switch slice of negative-bill-savings) trace to one
root cause: the persisted default plan is stale relative to the live model, so
it resolves on re-model rather than a code change. Confirmed via
run_modelling_e2e on three samples (stored vs live plans differ wholesale).

No new check added — the staleness is already surfaced by the existing checks
and addressed by the override-aware-rebaseline / persistence-fidelity work, so
a new check would only re-flag known divergence. Instead record the expectation
and the "audit the default plan only" rule in the skill Notes so the next
reviewer starts ahead. References kept to durable docs/adr (no PR numbers).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Khalim Conn-Kowlessar 2026-06-29 16:08:17 +00:00
parent 6c42065304
commit 54acc0388b
1 changed files with 21 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

24
.claude/skills/audit-ara-portfolio/SKILL.md
View file

@ -114,9 +114,27 @@ durable, compounding output of every audit.
## Notes
- Read-only on the DB. `run_modelling_e2e` is a dry run.
- **Audit the default plan only** — every check, and every characterisation
query, must filter `plan.is_default = TRUE` (the FE-shown plan). The runner's
`_QUERY`/`_ROLLUP_QUERY` already do; keep ad-hoc SQL consistent or the counts
mix superseded plans into the picture.
- **The stored default plan can be STALE vs the live model** — the persisted plan
is the output of an *earlier* modelling run; `run_modelling_e2e` re-models
against current logic + live EPC/solar. A stored-vs-live gap is the single
biggest driver of MEDIUM/HIGH anomalies on a not-yet-re-modelled portfolio, and
it spans more checks than the one below names. The fix is operational
(**re-model the portfolio, then re-audit**), not a code change — confirm a
sample with `run_modelling_e2e` before debugging the calculator.
- **Expected, not bugs** (until the override-aware-rebaseline + persistence-fidelity
PR deploys and the portfolio is re-modelled): much of `zero-works-post-differs`
and `plan-score-below-baseline` is the pre-fix baseline-vs-plan divergence and
should shrink after re-model — note it, don't re-debug it.
work — see `docs/adr/` — is re-modelled into the portfolio): much of
`zero-works-post-differs`, `plan-score-below-baseline`, `already-meets-goal-with-works`,
and the non-fuel-switch slice of `negative-bill-savings` is the same stale
stored-plan-vs-live divergence and should shrink after re-model — note it,
don't re-debug it. Confirm a sample's stored plan differs from `run_modelling_e2e`
(stored `air_source_heat_pump` for tens of £k where the live plan is a cheap
profitable `solar_pv`; stored works on a property the live model leaves at £0
because it already meets goal). The `negative-bill` certs that DO carry an
`air_source_heat_pump` are a genuine gas→electric fuel-switch (bills rise at
current price ratios) — expected by design.
- Adding a check is one decorated `(PropertyAudit) -> Optional[str]` function in
`scripts/audit/anomalies.py`; see its module docstring.