• Height-limit input — drives apartment floors upper bound and a hard check that configured floors × ~3m do not exceed the scheme height.
• Setback inputs (front, side, rear) — feed directly into the net buildable footprint for townhouse and apartment scenarios; currently baked into site_cover as an approximation.
• Infrastructure charges — split into trunk (adopted charges from the council's Charges Resolution, capped by Schedule 16) and non-trunk (service extensions/augmentation from the services availability check). Trunk charges are deterministic per council; non-trunk costs are auto-populated when services don't reach the frontage.
• Environmental offset cost — optional, tickbox-gated. Required under the Environmental Offsets Act 2014 only when ecological overlays trigger a prescribed environmental matter loss. Tickbox in the "Optional cost adjustments" group (A10), off by default, auto-ticked only when the overlay set (T1) includes a matter that triggers an offset obligation. When ticked, dependent inputs (Offset area (m²), Offset rate ($/m² or $/ha)) appear; offset_area × offset_rate is added as a project-cost line item. T2/T3 confidence badge and a "Why this is on" hover when auto-ticked.
• Community title / body corporate setup — optional, tickbox-gated. Required for community-scheme products (townhouse/apartment/mixed-use); not for Torrens detached/duplex. Tickbox in A10, off by default, auto-ticked only when the product type is a community scheme. When ticked, dependent inputs (CMS prep, body corp setup, per-lot registration) appear and are added to the cost stack.
• GST treatment toggle — margin scheme vs. standard, with the downstream revenue figure adjusted.
• [NEW] Car-parking ratio and GFA inputs — minimum parking per dwelling/bedroom by zone (and visitor ratio) plus a Gross Floor Area override.
• [NEW] Services and trunk infrastructure availability check — water, sewer, power, NBN capacity at the site. Sewer augmentation or transformer relocation can add $100–400k and months of lead time.
• [NEW] Contamination / environmental screening tier — Phase 1 / Phase 2 / remediation estimate. Candidate for the optional/tickbox pattern in A10, auto-ticked only when site history or overlay set indicates contamination risk (T1).
• [NEW] "Optional cost adjustments" group — reusable tickbox/opt-in pattern. A collapsible group hosting any rarely-applicable cost input. Contract: (a) when unticked, zero contribution and excluded from the <<<FEASIBILITY_UPDATE>>> payload; (b) when manually ticked, dependent inputs appear; (c) when auto-ticked (from planning-scheme, product-type, or overlay evidence), values are prefilled, a T1–T4 badge is shown on the tickbox, and a "Why this is on" hover explains the trigger in one sentence; (d) unticking an auto-ticked box shows a soft "AI had recommended this based on X; are you sure?" tooltip. Hosts A4, A5, A9 and any future rarely-used cost.
• [NEW] AI-authored "triggered inputs" summary line in the Executive Summary — one line listing any opt-in inputs the AI auto-ticked for this site, or "No optional cost adjustments triggered." if none.

Every prefilled value in A1–A11 must carry a T1–T4 tier and a retrieval date. The "Why this value?" hover (F2) displays the source record ID.

• Pre-sale threshold input (e.g. 60% pre-sales required by senior lender) and a calculated field for pre-sale revenue needed before drawdown. Defaults come from the user's own lender term sheet (user-supplied T1) or are left blank — no AI default.
• Interest capitalisation toggle — currently interest is calculated on a straight average balance; toggle for S-curve drawdown (more realistic for construction).
• Land holding cost input — rates, maintenance, insurance during the acquisition/DA period. Uses the council's published rates and a named insurance quote (T1); no AI estimates.
• Contingency input (typically 5–10% of construction) — currently folded into $/m² rather than surfaced.
• [NEW] Cash-flow waterfall timeline with S-curve drawdown — a month-by-month chart of acquisition → DA → construction → sales receipts, with peak-debt and funding-gap lines. Shared backbone for B7 and B8.
• [NEW] Live interest-rate and inflation feed — construction-loan rate pulled from a named, dated public source (RBA cash-rate series, published big-4 reference rates, ABS CPI). T1 only; no AI-estimated rates.

B7 — [NEW] JV / syndication waterfall — separate feasibility mode (refined in attempt 3)

A JV mode toggle that opens a parallel feasibility view next to the default owner-developer view. Inputs are largely shared — site, scheme, yield, cost, revenue — but the output set and flow of cash change. All hurdle/promote defaults are blank until the user enters the term sheet; no "typical" AI defaults inside the calculation.

• Parties — two-party (developer / capital partner) or three-party (developer / land-vendor vendor-finance / capital partner), user-configurable.
• Capital contributions — each party's equity plus any vendor-finance or land-in-kind contribution, with derived ownership percentages and peak exposure per party.
• Waterfall tiers — return of capital → preferred return → catch-up → promote split above the hurdle.
• Per-party outputs — IRR, MOIC, cash-on-cash return, distribution timing, per party. A deal that is 22% project-wide but 11% to the capital partner is flagged red.
• Drawdown rules — capital-partner equity drawn first (toggle) or pari-passu with the developer, feeding the same S-curve (B5).
• Sensitivity — tornado chart (E2) extends to preferred return, hurdle, and promote split.
• Hand-off — G6 term-sheet preview gains a JV variant with the capital partner's return profile stated in plain English.

JV mode wraps the base feasibility — it does not rewrite it.

B8 — [NEW] Build-to-Rent (BTR) / hybrid-tenure — first-class standalone investor category (refined in attempts 3 and 4)

A BTR / hybrid-tenure mode that opens a parallel feasibility view. First-class, standalone for the investor who acquires the land outright (market purchase, option, put-and-call) and develops to hold — no JV agreement with the land owner required. The JV-embedded variant remains available but is optional and not the default framing.

• Acquisition assumption, explicit. A toggle at the top picks the acquisition structure: (i) Outright purchase — default; investor buys the land at an agreed price; purchase price flows into the base cost stack. (ii) Option / put-and-call — option period, option fee, and strike price; option fee carried as sunk cost if the deal does not proceed. (iii) JV with land owner — only when the user explicitly selects it (this is the B7 path).
• Tenure mix toggle — 100% BTR, 100% BTS, or a hybrid split (e.g. 60% BTS / 40% BTR).
• Rental yield inputs — gross rental yield by bedroom count, vacancy rate, gross-to-net opex ratio. Rental-yield defaults are T1 from CoreLogic / SQM / REA median rents for the suburb and bedroom count, dated. Stabilised NOI computed per dwelling and aggregated.
• Cap rate input and terminal sale year — net stabilised value = NOI / cap rate. Cap rate is user-supplied (from a named agent or valuer); no AI cap-rate default feeds IRR.
• Hold-period cash flow — the S-curve timeline (B5) extends from practical completion to the terminal sale, with rental income, opex, and hold-period debt service; ungeared and geared IRR both reported.
• Sell-all / rent-hold / hybrid comparison — three-column summary on the Executive Summary showing margin-on-cost, day-one peak debt, 5-year IRR, 10-year IRR for each tenure option.
• "Investor desire" presentation layer. Executive Summary in BTR mode leads with stabilised yield on cost, day-one LVR, stabilised LVR, ICR at stabilisation, ungeared and geared IRR over 5 and 10 years, terminal-sale sensitivity. Developer-style margin-on-cost / gross profit numbers remain available but secondary.
• Tax and state-concession flags — MIT tax, land-tax surcharges, stamp duty, foreign-investor surcharges. Surfaced as flags, not calculation inputs; the user is prompted to have a tax adviser confirm.
• Optional JV embedding — if the user is also in JV mode (B7), BTR appears as a development-option toggle inside the waterfall. Each party's IRR is then computed against the BTR hold-period cash flow.

• Impact Assessable — "Recent DA Precedent" sweep for maximum realistic yield. Load-bearing. Returns actual DA numbers, decision dates, approved yields, decision paths, performance-outcome arguments, and conditions from the council's register (T1). A precedent cannot be inferred — if the register does not return it, it is not cited. Any uplift without a supporting precedent defaults back to the scheme-permissible maximum or is flagged as a novel argument with elevated risk. Output: a "Precedent Summary" block, new precedent_confidence and precedent_refs fields in <<<FEASIBILITY_UPDATE>>>, cached precedent findings in the knowledge base.
• "Assessment category" display field in the Executive Summary — makes it clear whether the numbers are Code or Impact based.
• Pre-lodgement meeting fee input ($800–$2500 typical), T1 from the current Fees & Charges schedule.
• Appeal contingency input — only shown when Impact Assessable is active; defaulted from the AI's assessment.
• Referral-agency cost inputs — SARA, DES, QFES, DTMR — surfaced automatically when the AI detects an overlay that triggers a referral. T1 from the published schedule.
• [NEW] Submission count from the council's DA register (T1 only). Downgraded in attempt 5 from a predictive "submission-risk score" to a simple count of prior submissions on comparable DAs in the catchment. No demographic inference, no AI-estimated probability.
• [NEW] Rezoning / spot-rezoning / planning-scheme-amendment scenario — descriptive by default. No probability of rezoning success is presented unless drawn from the state's published scheme-amendment register (T1) for comparable amendments in the same LGA. Otherwise it is a descriptive "what would need to change" view only, with no hit-rate.
• [NEW] Affordable-housing / social-housing uplift modeller — only where the scheme contains an explicit density-bonus provision (T1); otherwise not offered.

• "Refresh suburb comparables" button — re-queries the property-search function. CoreLogic / RP Data / Pricefinder pull with a dated snapshot stored with the feasibility.
• Sale-price range field (low / base / high) — statistical bands from the actual comparable set, not AI-authored ranges.
• Construction-cost source tag — mandatory, not optional. Allowed sources: QNCC, Rawlinsons, a named QS quote. An AI estimate is T4 and must be labelled as such in every export.
• [NEW] Absorption-rate / days-on-market overlay — T1 only, from CoreLogic / REA listings data for the subject suburb and product type.
• [NEW] Competing-supply pipeline — T1 only, within a configurable radius: count DAs lodged / approved / under construction for comparable product, surfaced as a "supply risk" signal.
• [NEW] Independent data-source tag on every prefilled value — beyond T1–T4 tiers, show the source (ArcGIS Brisbane PlanRegister / QLD State Planning Policy / CoreLogic comparables / DA-precedent sweep / AI assumption). Mandatory.

• Side-by-side Code vs. Impact comparison — two-column comparison of yield, profit, ROI, and timeframe with additional yield gain / extra risk quantified. Numbers on the Impact side come only from the Realistic band (see I2) unless the user explicitly switches to Stretch. Cross-links to the Precedent Summary from C1.
• Tornado sensitivity chart — input ranges taken from sourced data in D, not hand-picked.
• Save-scenario / compare-scenarios — save inputs as a named scenario; compare multiple on one screen.
• Monte-Carlo risk view — optional advanced view distributing sale price, construction cost, and interest rate; returns a break-even probability. Input distributions are sourced (CoreLogic sale dispersion, RBA/big-4 rate history, QNCC cost bands). No hand-picked distributions.
• [NEW] Staged-development scenario — split the project into two or more stages with separate DA cost, construction cost, and sales windows; consolidate the cash-flow waterfall across stages.
• [NEW] Exit-strategy comparison — compare "sell individual lots/units", "bulk sell to a builder", "retain and rent (BTR)", "land-bank and re-analyse in 24 months" on a single chart.

• Confidence badges on each prefilled input (T1 Verified / T2 Discovered / T3 Learned / T4 Assumption). Mandatory on every prefilled input.
• "Why this value?" hover on each AI-prefilled input — shows the one-sentence reason from <<<FEASIBILITY_UPDATE>>>, the scheme provision cited, and the source record ID. Mandatory.
• Diff view when prefill is applied — show old → new with per-field revert.
• "Reset to AI recommendation" button per input.
• Inline chart showing dwelling yield broken down into gross land → environmental deductions → road/open-space deductions → net developable → final yield.
• Group collapsing state persistence across sessions.
• Currency formatting in number inputs — display $2,300,000 while storing 2300000.
• Validation warnings — soft warnings when inputs breach scheme limits; yellow banners without blocking calculation.
• [NEW] Narrative link-outs on every number — each prefilled value links to the specific paragraph in the AI's prose analysis that justifies it.
• [NEW] Unit-economics card — one compact card per dwelling/unit showing revenue, direct costs, share of fixed costs, and per-unit profit margin.

• "Export feasibility as PDF" — a branded PDF of the Executive Summary plus the AI's prose analysis. Every numeric value is printed with its tier tag and source record ID. A "Cost adjustments applied" line lists any optional-group tickboxes that were on.
• "Send to my quantity surveyor" — email the construction-cost assumptions as a single verification request.
• Versioning — timestamped snapshot of inputs + outputs on every Code/Impact run.
• Side-by-side comparison with the saved CRM property version — what has changed since last save, including any change to optional-group tickboxes between versions.
• [NEW] Pre-lodgement brief export — Word/PDF skeleton pre-lodgement brief combining the site information, scheme provisions cited, Precedent Summary from C1, and performance-outcome arguments. Cannot include any T4 value without an explicit user override captured in the audit log (I5 enforces).
• [NEW] Financier one-page term-sheet preview — editable one-pager with deal summary, peak debt, LVR/LCR, pre-sale requirement, key risks, exit. JV and BTR variants available. Same T4 constraint as G5.

• "Book pre-lodgement meeting" button that opens the relevant council's pre-lodgement booking URL with the site address pre-populated.
• Referral-agency contact card that appears automatically when a triggering overlay is detected.
• Infrastructure-charges resolution link — deep-link to the council's current Charges Resolution PDF for the zone.
• [NEW] Integrated title / survey / plan-of-subdivision requests — one-click requests out to a nominated surveyor and conveyancer for title search, identification survey, and plan of subdivision, pre-populated from lastSearchData.
• [NEW] Live DA-portal deep-link and tracking — once a DA is lodged, the feasibility page shows the DA's status (lodged / information request / decision) pulled from the council's DA-tracking portal.

All H-links are to live council systems; if a link breaks or a register changes URL, the system degrades to "source unavailable" rather than guessing.

The mechanism by which the reality rule is enforced over time. Several items connect to C1, C6, C7, E3, F8, G3, G5 — Section I is the glue that turns those into a reliability discipline.

1. Council-approval track record — T3 hit-rates only. Every Impact-Assessable recommendation is persisted (G3) with argument, uplift, precedents cited, and band. When a DA is later lodged on the same or a comparable site, the outcome — approved, approved with conditions, refused, appealed, withdrawn — is pulled from the council's DA-tracking portal (H5) and written back against the original recommendation. The number shown next to every uplift is a real historical rate (e.g. "this argument has carried in 11 of 13 comparable DAs in Moreton Bay since 2022 (85%)"), never an AI-estimated probability. If sample size is below threshold (n ≥ 10), no rate is shown and the Realistic band is not offered for that council/zone/argument type.
2. Three-band recommendation — Conservative / Realistic / Stretch, all three shown every time.
   • Conservative — scheme-permissible maximum only, no performance-outcome uplift. Approval probability approaches 100%.
   • Realistic — performance-outcome uplift backed by two or more T1 precedents in the same scheme since the last relevant scheme amendment AND a T3 hit-rate above threshold. If either fails, only Conservative is produced.
   • Stretch — further uplift justified by a thinner evidentiary base. Requires at least one T1 precedent OR a P&E Court decision (T1); without one, Stretch is not offered. Flagged with lower confidence and an explicit "this is a novel argument; expect IR or appeal" note.
3. Precedent-weight scoring — deterministic published weights. No model-authored weights. Weights combine: Recency (current scheme cycle > older); Scheme-amendment proximity (post-amendment > pre-amendment); Distance from subject site (same catchment > same ward > same LGA); Similarity of argument (same performance outcome, similar overlay set, similar uplift %); Decision path (delegated < full committee < P&E Court). The Realistic band requires a minimum weighted score. If insufficient, the system returns only the Conservative band — it does not invent a Realistic band.
4. (Cut — council-officer style matching; see "What has been removed" above.)
5. Automated council-submission sanity check (pre-flight). Before G5 or G6 exports, a pre-flight runs: every numeric input satisfies either the code-assessable limit or sits at a level the Realistic band's precedents support; every performance-outcome argument cites a current (not superseded) scheme provision; referral agencies are all costed and timelined; yield is achievable on net developable after environmental deductions and parking; the combination of uplifts exists in at least one approved precedent. Red outputs block export until resolved or explicitly overridden with a reason captured in the audit log. Blocks export of any T4 value without override.
6. Court and appeal decision ingestion — T1 only. Queensland P&E Court (and equivalents in other states) decisions are ingested and tagged by argument type for win/loss. An independent evidentiary source distinct from I1 and the authoritative source for novel arguments.
7. Explicit "what would make this refusal-proof?" output. For any recommendation below a threshold hit-rate (e.g. Realistic 60–70%), output the minimum changes that push it over — e.g. "drop the rear unit from 3 to 2 storeys to match precedents DA-2023-2451 and DA-2024-1108". Each suggested change must be tied to a specific T1 precedent or court decision. No generic "reduce density" suggestions.
8. Versioned, immutable assumption ledger. Every recommendation is tied to a ledger of the inputs and evidence it used at the moment it was made (planning scheme version, precedent set as at date, court decisions as at date, track-record hit-rate as at date, rate feeds as at date). A re-run next month can be diffed against the original and the user can see exactly which piece of evidence moved.
9. Creativity budget — system-controlled, not user-controlled. (Replaced in attempt 6.) The precedent-weight threshold that I3 uses moves automatically as a function of T1 precedent count and T3 hit-rate. The UI reports the current budget ("Realistic band available: 3 T1 precedents, hit-rate 62%") rather than exposing a dial. The only user-facing control is "show me Conservative / Realistic / Stretch", and Stretch remains gated by evidence. Aggressive settings never lower tier requirements — fewer T1 precedents may be admissible, but T4 inferences are never admitted into the Realistic or Stretch bands.
10. Approval-confidence number on the Executive Summary — visible every run. A single number, computed only from T1/T2/T3 inputs. T4 values are excluded from the calculation and the UI states so. Over time its calibration against realised outcomes becomes the north-star metric for the system's reliability.

Together these convert the framing "the system should be reliable and creative" into an operational rule: creativity is allowed only in the direction the evidence supports, and the evidence must be retrievable from a named source at a named date.

From the user (J1–J3)

1. Spreadsheet headers and section delineation. The four result tables (DA summary, Costs, Finance, Returns) currently read as one clustered block. Each needs a clear caption, a visible column header row (e.g. Item / Value / Units / Source), visual banding between sections, a sub-total row at the bottom of each group, and whitespace between groups. Input side gets the same treatment — one caption per group, column headers for Input / Value / Units / Source tier, and dividers between Land & Overlays, Purchase & Fees, Finance, and Profit & Equity.
2. Creativity budget is system-controlled, not user-controlled. A user-facing slider would be pulled toward Aggressive and undermine the realism rule. Replaced by a system-set creativity budget derived from evidence (see I9 above). No slider, no override dial. The only user-facing control remains "show me Conservative / Realistic / Stretch", with Stretch gated by evidence.
3. Sections collapsed and hidden when not triggered. Every group — inputs and results — defaults to collapsed. A group is entirely hidden (not just collapsed) if it is not triggered for the specific feasibility (no environmental-offset row if no overlay hit, no contamination row if no EMR/CLR hit, no JV block if no JV flagged, no BTR block if exit is sale). A single master tick-box ("Show all sections, including ones not triggered for this site") reveals hidden groups for power users.

Further agent suggestions (J4–J20)

4. One answer row per table, visually promoted — Residual Land Value, Total Cost, Peak Debt, Project IRR / Margin on Cost — rendered larger, bolded, with a coloured left border.
5. Units in the header, not on every row — move "$", "%", "m²", "months" into the column header or a dedicated units column so values stay aligned.
6. Consistent number formatting across all tables — thousands separators on money, two decimals on percentages, no trailing zeros on areas, negatives in parentheses and red. One formatter, applied everywhere.
7. Source tier badge as a column, not a hover. F1/F2 put T1–T4 badges on hover; promote them to a first-class Source column on every spreadsheet row so the user sees at a glance how much is T4 assumption vs. T1 fact.
8. Triggered-by column on group headers — when a group is revealed because something triggered it, the header says why (e.g. "Environmental offset (triggered by: Koala Habitat overlay, T1)"). When hidden groups are revealed via J3, their headers say "Not triggered for this site".
9. Sticky headers on long tables — as the Costs table grows, the column header row sticks to the top of the viewport.
10. Section completion indicator — each collapsed input group header shows complete / partial / empty at a glance.
11. Changed-from-default highlighting — inputs the user has manually overridden get a subtle left-border highlight and a "reset to source value" affordance.
12. Code vs. Impact diff inside the spreadsheets. E1 already proposes a side-by-side comparison; inside the tables, show both values in the same row with a Δ column and a visual cue on rows that actually differ.
13. Print/export layout distinct from screen layout. On-screen is dense; the exported PDF switches to one-group-per-page with the headline row repeated at top, Source column retained, hidden (not-triggered) groups omitted entirely.
14. Keyboard navigation across cells — arrow keys between input cells; Tab to next group; Enter to collapse/expand.
15. Warning and error rows use row-level treatment, not modal dialogs — amber/red row with an inline message; no popups. I5 pre-flight aggregates at export time.
16. Formula transparency on click — clicking any derived result cell opens a small drawer showing the inputs and the formula (e.g. Residual Land Value = GDV - Total Cost - Target Profit), with each term linkable back to its input cell.
17. Narrow-screen fallback — on tablet/phone, the four-column spreadsheet collapses to a stacked card layout (one card per row, label-above-value) rather than forcing horizontal scroll.
18. Empty-state copy when a group is hidden — if the J3 master toggle is off, nothing shows. If on but the group is not triggered, render an empty state ("No triggers apply to this site — showing for reference. Most inputs are optional.") rather than a silent empty table.
19. Scenario save/restore visible at the spreadsheet level — a small "Scenario: Base / Realistic / Stretch / Custom" pill above the first table; clicking switches the whole sheet and animates the changed rows.
20. Last-updated and source-as-at stamps per table — footer line such as "Comparables as at 2026-04-18 (CoreLogic), rates as at 2026-04-15 (RBA), precedents as at 2026-04-12 (council register)." Ties the spreadsheet to the assumption ledger (I8).

Not a line item on the list, but the framework the user asked for on calculation confidence. Layered independent checks so a bug has to slip past all of them:

1. A frozen "reference property" with known-good outputs. 51 Greenslill Road, Albany Creek is the first fixture. Yield, GFA, construction cost, revenue, gross profit, margin, IRR, peak debt are recorded as expected-output fixtures. Every code change runs the fixtures; any drift beyond tolerance (e.g. 0.5%) fails the build.
2. Inline "show-your-working" panel on the Executive Summary — every headline number gets a one-click expansion showing inputs → intermediate values → formula → result.
3. Dual-path recomputation — for the highest-stakes numbers (gross profit, margin on cost, IRR, peak debt), compute two ways and show a red banner on disagreement.
4. Unit tests for every formula — site coverage, net developable area, yield, GFA, construction $, interest on average balance, S-curve interest, IRR, NPV, margin on cost, margin on revenue.
5. Third-party benchmark comparison — periodic comparison against Estate Master DF / Feastudy / a trusted QS spreadsheet.
6. Soft validation warnings in the UI (already F8).
7. Source-tag every prefilled value (already D6 and the T1–T4 tier rule).
8. Versioned snapshots of inputs + outputs on every run (already G3, and the I8 assumption ledger).

The agent needs from the user: a short list of reference sites with expected outputs; tolerance bands per output; and a source-of-truth benchmark tool if one exists, for the dual-path comparison in check 5.

Strict reading — can the software itself become RPP / RPIA / CPP?

No. Professional registration is granted to natural persons who meet education, experience, CPD, and professional-conduct requirements and who are personally accountable (including for PI insurance). A software system cannot hold those credentials.

Loose reading — can the output be treated as an RPP-signed deliverable?

Yes, by pairing the system with a registered town planner who reviews and signs. The "AI-assisted, planner-signed" model — the same shape as QS reports, survey plans, and signed financial statements. The system's deliverables (G5, C1 Precedent Summary, G1 PDF) are designed to be reviewable and signable by an RPP because they carry the evidence, the scheme provisions cited, and the I8 assumption ledger. The RPP's signature is the legal deliverable; the system is the drafting tool.

Three commercial routes (each worth weighing)

1. In-house RPP panel — one or more RPPs (employed or contracted) who review every Impact-Assessable output going out under the business's name, sign it, and carry PI. Fastest to market; recommended for the first 12–18 months regardless of the other routes.
2. PIA endorsement / Planners Declared AI Tools programme (if and when one exists). The reliability stack in Section I (I1, I3, I6, I8) is the evidentiary base a programme would look for. Publishing a transparency report (overall approval rate vs. recommended rate, per LGA) builds credibility a programme would later ratify.
3. Sign-off service as a platform feature — on-demand RPP review and signature inside the product, SLA'd, returning a signed, deliverable artefact. Monetises the pairing in (1).

What the system needs to build to be credibly RPP-signable (beyond Section I)

• A review mode in which an RPP sees the full evidence chain next to each recommendation, can edit, accept, and sign.
• Immutable post-sign archival — once signed, output is frozen and archived with the planner's credentials, the I8 ledger, and a tamper-detection hash.
• PI insurance alignment — the business carries a policy covering the combined AI-drafted + RPP-signed deliverable; panel RPPs carry their own cover. Commercial arrangement, not a software task, but needs the audit trail.
• Scope discipline in every signed deliverable — states clearly what it is (planning advice, feasibility analysis, pre-lodgement brief) and what it is not (legal advice, tax advice, QS report, valuation). The sign-off scope matches.

Short answer: the software cannot itself be an accredited town planner. It can be — and should be designed to be — the drafting tool an accredited town planner signs. Section I is the credibility base; the RPP-panel + signed-output arrangement is the commercial model; PIA-style recognition, if it exists later, is the ratification. None of this requires waiting — the in-house RPP panel approach works from day one.

Highest ROI / lowest effort

Mandatory T1–T4 tier tags on every prefilled value (F1), "why this value?" hovers with source record ID (F2), height-limit input (A1), assessment-category tag (C2), construction-cost source tag enforced (D3), narrative link-outs (F9), triggered-inputs summary line (A11), three-band output (I2), system-controlled creativity budget reported to the user (I9 / J2), approval-confidence number gated on T1/T2/T3 inputs only (I10), spreadsheet headers and section delineation (J1), collapse-and-hide-untriggered pattern (J3), one answer row per table promoted (J4), units in header (J5), source tier as a column not a hover (J7), consistent number formatting (J6), last-updated stamps per table (J20).

Medium effort, high value

Optional cost adjustments group (A10) hosting A4/A5, side-by-side Code vs. Impact drawing only from sourced bands (E1), infrastructure-charges input from the Charges Resolution (A3), pre-sale threshold from user-supplied term sheet (B1), S-curve cash-flow waterfall (B5), save-scenario (E3), BTR standalone with outright-purchase default and sourced rental-yield defaults (B8), absorption + competing-supply overlays from T1 data (D4 + D5), precedent-weight scoring (I3), pre-flight submission sanity check (I5), "what would make this refusal-proof?" output tied to T1 precedents (I7), versioned assumption ledger (I8), triggered-by column on group headers (J8), sticky headers (J9), section completion indicator (J10), changed-from-default highlighting (J11), Code vs. Impact diff inside tables (J12), formula transparency on click (J16), warning/error rows as row-level treatment (J15), scenario save/restore at spreadsheet level (J19).

Higher effort, deferred

Recent DA Precedent sweep from the council register (C1), pre-lodgement brief export (G5), JV waterfall with user-supplied terms (B7), reframed rezoning scenario (C7), Monte-Carlo with sourced distributions (E4), DA-portal tracking (H5), council-approval track record reaching sample threshold (I1), court decision ingestion (I6), print/export layout distinct from screen (J13), narrow-screen fallback (J17), keyboard navigation (J14), and RPP review/sign-off mode (Part 3).