TSXV:CBR Cabral Gold Inc.

59 — medium, against August’s 70 and an earlier draft of this report’s 68. Neither of those was ever derived. This one is, below, and the derivation is what moved the number.

Two corrections, in opposite directions

The earlier draft replaced this pillar’s industry benchmark — which in August was AISC margin, an operating measure, scored 90 — with “P/NAV (junior gold producers)” scored 25. That is the valuation conclusion inserted into the Quality pillar. The same 1.27× multiple then drove Valuation down to 38 and dragged Quality below the 65 boundary, and because the decision matrix returns SELL for medium-quality-plus-expensive but HOLD for high-quality-plus-expensive, that double count was the entire difference between a SELL and a HOLD. It also cut the moat score from 58 to 22 with nothing in the text to support it.

The honest operating benchmark has improved: the prefeasibility study’s all-in sustaining cost is US$1,210 an ounce against spot of US$4,416 — a 72.6% margin, better than the ~70% August scored at 90. Quality is restored to 90 on that measure, and it is the strongest input in the build below.

The second correction runs the other way, and it is larger. Having removed the valuation contamination, this report set Quality at 68 — three points above the 65 boundary that decides the verdict — and published no derivation for it. Built explicitly from the framework’s own sub-signal list, it comes to 59. The table below is published precisely so the next reader can check it rather than take it on trust, and because a pillar that decides a signal at a three-point margin has no business being unauditable.

Start with what they did right, because it is genuinely impressive

Cabral received its operating licence on 13 August 2026, completed the leach plant, and poured first gold on 10 September — roughly 1,130 ounces, ahead of the schedule set out in its own 2025 prefeasibility study. It financed the build with a US$45m gold loan closed in November 2025 rather than by issuing equity at the bottom. Small mining companies miss these dates as a matter of routine. This one beat them.

That is real, it is why ROIC on projects scores 78 below, and it is not what most of this pillar measures.

The build, in full

Sub-signals selected from the framework’s four universals plus the Mining profile. The three sub-signals scored 50 are not-applicable-neutral: the company earned its first revenue this morning, so neither a revenue trajectory nor a realised margin exists to score.

The moat, scored

One caveat on the benchmark, stated because it cuts against the score. The US$1,210/oz all-in sustaining cost is the study’s figure struck at a US$2,500 gold assumption, and it is being divided into spot of US$4,416. Royalty and profit-share components scale with the metal, so the realised margin will be below 72.6%. It does not move the score: the band boundary is 40% of spot, which at US$4,416 means an AISC above US$2,650 — more than double the study’s figure. The precision is overstated; the band is not.

What else the pillar is measuring

• Asset concentration. One district, one country. Brazil is a reasonable mining jurisdiction and the Tapajós has produced gold for two hundred years, but a single-asset company carries the whole of its operational risk in one place.
• Mine life. Stage 1 is 6.2 years against a framework benchmark of eight for a producer — the single weakest sector-specific input above. That is a starter operation, not a franchise. The company’s future rests on Stage 2, which is discussed honestly at §4 and does not yet exist as a plan.
• Financial record. There isn’t one. Trailing earnings per share are −C$0.05 and the first revenue in the company’s history was earned this morning. Price to book of 27.5× is a number about accounting, not about worth. There is, however, a forward estimate — C$0.19, an 8.1× forward multiple — and §4 addresses why that is not used as the anchor and why it is nonetheless a real counter-argument.

Competitive Environment

Cabral does not compete for customers — gold sells itself. It competes for capital and for acquisition attention, against a Tapajós peer set that includes G Mining Ventures (Tocantinzinho, in production and far larger), Serabi Gold (producing, similar district), and a long tail of explorers. Against those, Cabral’s distinguishing feature is that it is now producing, which is a genuine step up in class.

Share trajectory: improving. Moving from developer to producer changes who can own the shares — a great many funds cannot hold a pre-revenue miner. That re-rating is real, it is happening, and §4’s argument is that the market has already paid for rather more of it than has been delivered.

46 — Full. And the first thing this section owes you is an explanation of why the last report had no number here at all.

The anchor the August report recorded but did not use

That report set Cabral’s valuation basis to “P/NAV (anchor N/A — pre-revenue)” and computed nothing — while recording the study in full elsewhere in the same file, and stating that the economic-study existence check had been run. An earlier draft of this report accused it of missing the study altogether. That accusation was false and is withdrawn. The real defect is narrower and still worth naming: the study was found and then not used as the anchor it plainly is. The updated prefeasibility study on the Stage 1 gold-in-oxide operation, dated July 2025, publishes exactly what an anchor needs:

• After-tax NPV US$73.9m and IRR 78% at US$2,500/oz gold
• After-tax NPV US$138m and IRR 139% at US$3,340/oz
• Initial capital US$37.7m; payback ten months at $2,500, seven at $3,340
• 113,000 ounces over 6.2 years

Two published points on the same curve give a slope: US$76.3m of net present value per US$1,000/oz of gold — US$7.63m per US$100/oz. At spot of US$4,416 that puts Stage 1 at US$257.81m, or C$355.20m. This is an extrapolation well beyond the study’s own published range and should be read as one — but the relationship is close to linear for an operation whose costs are largely fixed, and the direction is not in doubt.

What the rest of the district is worth

The global resource is 666,382 ounces indicated and 525,669 inferred, about 1.19 million in total — the company’s current statement, which combines a hard-rock estimate effective October 2022 with oxide estimates updated in October 2024 and July 2025. An earlier draft of this report used the superseded 2022 global figures of 604,000 and 534,500. Stage 1 mines 128,903 contained ounces of Probable Reserves at 0.65 g/t, from which it recovers 113,155. The remaining roughly 1.06 million ounces is primary hard rock at 1.14 g/t indicated — a higher grade than Stage 1’s 0.65 g/t reserve, not a lower one — that would need a mill, and there is no study, no capital estimate and no permit for it. So it is carried at an in-situ value per ounce, indicated at full weight and inferred at half, which is the disciplined treatment for ounces with no plan attached.

Adding C$26.1m of proceeds that would come in if the in-the-money options and warrants were exercised — they add shares, and they also add cash — and with net debt of approximately nil after the August placement, across 397.7m in-the-money shares:

NAV = C$1.17 per share. At C$1.48 that is 1.27× NAV.

Two corrections to this report’s own balance sheet, found in round six

The first was a sourcing error and it cut against the shares. Earlier rounds took the company’s July presentation figure of “C$35M cash at 29 June” over the C$22.4m a data provider reported, on the principle that primary beats provider. That was wrong, and instructively so. Cabral’s filed interim statements for the period ended 30 June 2026 — board-approved 26 August — report cash and cash equivalents of C$22,417,212. The provider figure was the filed figure. The deck’s C$35M reconciles to cash plus C$6.58m of gold deposits plus C$6.43m of non-current advances, and 29 June is not a period end. Worse, the C$67.4m of debt this report deducts comes from that same filed page: the liability was taken from the filing and the asset from a deck.

The second was larger and ran the other way: this report was deducting the mine twice. The Stage 1 value is the prefeasibility study’s net present value, and that study nets US$37.7m of initial capital as a future outflow. It is not future. The filed statements show property and equipment going from C$1.48m to C$57.83m, and the plant poured gold on 10 September. Deducting the gold loan that financed the build and using a net present value that has already subtracted the build charges it twice. Adding the spent capital back puts Stage 1 at US$257.81m, or C$355.20m.

The two corrections are one finding split by sign, and the second is four times the first: together they take net asset value to C$1.17 and the multiple to 1.27×. Fixing only the cash would have produced a SELL built on a double count.

The strongest argument against this report

On 24 August 2026 Cabral sold 34,582,754 units at C$1.30 to Alpayana S.A.C., a Peruvian miner, raising C$44,957,580 for a 9.99% stake and the right to nominate a director. Rounds one through three of this report did not mention it at all — a material corporate action three weeks before the stamp, found only by an independent auditor. It matters three ways, and the third is the uncomfortable one.

• The share count. Basic shares are 346,173,703, not the 301.7m from April, and fully diluted 415,024,289. This report values on 397.7m in-the-money shares, excluding the 17.3m warrants struck at C$1.70 while the stock trades at C$1.48 — crediting those would add proceeds at C$1.70 a share against a NAV near C$1.17, which flatters the answer on an exercise nobody would make. It would also produce a HOLD — so the verdict does not turn on the choice. It is not used because it credits an exercise nobody would make: a C$1.70 strike against a C$1.48 price.
• The balance sheet. Net debt of C$44.9m becomes approximately nil. §3 raises balance-sheet health from 40 to 62 accordingly. Note the cash figure predates the raise, and the ramp has been spending since, so this treatment is if anything generous.
• The thesis. This report marks Stage 2 at US$75 an ounce because it has no study, no capital estimate and no permit. A strategic industry buyer — one that operates mines for a living — paid C$1.30 a share, above the C$1.17 net asset value computed here, and earmarked the money for that very Phase 2. That is a considered party disagreeing with the central assumption of this analysis, in cash, three weeks ago, and it is the best reason to think this report marks the asset too conservatively.
• But the same transaction cuts the other way, and an earlier draft printed only the half that argued against this report. C$1.30 was a discount to the market, not a premium: the last close before the deal was C$1.37 on 21 August and the shares had touched C$1.40 the day before. So the strategic paid above what this report thinks the assets are worth, and below what the market was charging — which is rather closer to this report's position than to the market's. Both facts belong on the page.

The surface, because one cell is not an answer — and here the cell decides the signal

The in-situ mark is the input this whole report turns on, and it cannot be pinned tightly. So here is the grid rather than the convenient cell, at spot gold, fully diluted:

This table decides the signal, and an earlier draft of this report said it did not. That draft claimed the band moved across the surface while the signal stayed HOLD throughout — which was simply wrong, and wrong in the direction that flattered the shares. At medium business quality the decision matrix returns SELL against an Expensive valuation and HOLD against a Fair one, so the mark on unstudied rock is the decision-relevant input: a SELL would need the mark below about US$36 an ounce, which nobody argues for; from US$75 it is a HOLD, and only at a US$200 mark would the matrix return a BUY. Only the most generous cell would produce a BUY, and the difference between the published call and that cell is a single assumption — which is why the whole surface is printed rather than the one convenient cell.

Where this report puts itself, and the tension in doing so. The published call marks at US$75/oz, the disciplined treatment for ounces with no study, no capital estimate and no permit. But §4 goes on to establish that the market’s implied mark of US$142 sits inside the range advanced developers actually command. Those two statements pull against each other, and a reader is entitled to see that: mark the rock the way the market does — US$142/oz on the same full-ounce basis — and the multiple falls to about 1.0× (1.09× on the half-weighted basis the §4 surface uses). The report keeps the conservative mark because the disciplined treatment is the house rule for unstudied ounces — not because the alternative is unreasonable.

A basis error in this report’s own comparison, and it flattered the bear case

Earlier rounds compared the report’s US$75/oz mark with the market’s implied US$142/oz and concluded the market was paying more than twice what the asset is worth. Those two numbers are not on the same basis. The US$75 is applied to indicated ounces at full weight and inferred at half — a standard conservatism, and it stays — so it values 1.06 million ounces as if they were 800,314. The market’s figure is spread across the full count. Restated like for like, this report marks the remaining ounces at about US$56 an ounce, not US$75.

That matters, because §4 goes on to cite published 2026 work putting listed developers at an average of US$88 and advanced developers holding a study at US$100–300. US$56 is below the listed-developer average. To defend it this report would have to argue that Cabral’s 1.06 million remaining ounces — 1.14 g/t primary rock, in a district with a built plant, a permitted operation and a producing starter pit — are worth less per ounce than the average listed developer’s. It has no such argument, and it should not pretend otherwise. The mark is kept at US$75 on the surface basis because Stage 2 has no study, but it is now published as what it is: the most conservative defensible mark, not a neutral one, and the §4 surface shows what happens at every other.

The number that decided it

Run the calculation backwards. At C$1.48 fully diluted, after crediting Stage 1’s entire net present value at spot gold and adjusting for net debt and warrant proceeds, the market is paying about US$142 an ounce for the 1.06 million ounces that Stage 1 does not touch — ounces with no study, no capital estimate and no permit.

How rich is that? Less obviously rich than an earlier draft of this report claimed, and the correction matters. That draft asserted a “customary range of US$30–100” with no source and no gold-price era attached, which is exactly the unsourced-benchmark failure this report criticises elsewhere. Published 2026 work puts advanced developers holding a PEA or PFS at roughly US$100–300 an ounce, with a listed-developer average nearer US$88. On that basis US$142 sits inside the advanced-developer band, below its midpoint, rather than at 1.8× the top of a junior one.

So the honest statement is narrower than the draft’s: the market is marking unstudied rock below the midpoint of what advanced developers command, on a company whose Stage 2 has no study, no capital estimate and no permit. That is stretched. It is not absurd, and this report should not have implied it was.

The reverse-DCF implied-growth read, and why there isn’t one

The framework asks §4 to show the growth rate the current price implies, and a reverse discounted-cash-flow needs a forward earnings base. One exists, and this report chooses not to anchor on it — which is a different thing from it not existing, and an earlier draft got that wrong. The forward estimate is C$0.19 a share, an 8.1× forward multiple.

Why not anchor on it: it is a single forward year for a mine that poured its first ounce this morning, on a 6.2-year starter pit, from a thin estimate base — and a P/E on year one of a heap leach says almost nothing about a business whose value is the ore body. P/NAV is the right lens for a miner and this system applies it to miners as a rule. But 8.1× forward earnings is a real counter-argument to a report arguing the shares are dear, and it belongs in the open. The growth question is answered instead in the form the asset allows: the price implies an in-situ mark of US$142 an ounce on unstudied rock.

The four relative lenses

• Sector median (20%). Producers trade at 0.7–0.9× NAV on BMO’s January 2026 work; developers below. Cabral at 1.27× is not near that band. Peer band sourced; the Cabral figure computed here.
• Own history (15%). The shares are at C$1.48 against a 52-week close range of C$0.40–C$1.51 (the C$1.59 high is an intraday print made on 10 September, after this report’s stamp) — the 91st percentile, and 59% above the 200-day average of C$0.93.
• PEG / growth-adjusted (10%). Not computable. Trailing earnings are negative and the first revenue was earned this morning. There is no growth rate to divide by, and inventing one would be a fabrication.
• Analyst consensus (15%). A panel exists, it is tiny, and it disagrees with this report. An earlier draft said there was none; that was wrong — the framework’s own target tool returns two analysts, both STRONG BUY, mean target C$1.65 (range C$1.60–C$1.70). That is 11.5% above the stamped price and 35% above this report’s own base case of C$1.22. Two analysts is not a consensus and the lens is discounted heavily for it, but it is a genuine counter-datum on a report arguing the shares are dear, and burying it would have been the worst thing in this document.

What would make this wrong

Three things, all genuine, and none of them currently supported by a document:

1. Stage 2 could be worth far more than an in-situ mark, and an earlier draft of this report got the reason backwards. That draft said the mill feed was “0.87 g/t, a fifth of the oxide starter’s effective economics”. Wrong twice: 0.87 g/t is the global indicated grade, and the mill feed is the primary hard rock at 1.14 g/t indicated — against a Stage 1 reserve grade of 0.65 g/t. Stage 2 is roughly 1.75× the grade of the operation now running, not a fifth of it. What makes it uncertain is capital and permitting, not rock quality. Stage 1 turns 128,903 contained ounces into US$257.81m of present value; a mill would earn far less per ounce because a mill costs far more than a leach pad, but it would earn a great deal more than US$75. When a study exists, this section changes materially.
2. An updated global resource covering six deposits — up from three — is expected by year-end 2026. More ounces at the same in-situ mark is a higher NAV.
3. The gold price. Every figure here is struck at US$4,416. The NAV is highly geared to it: at US$5,000 the central-case NAV rises to about C$1.30.

61 — improving. The tape is unambiguously strong, and it is strong in a way that is itself part of the problem.

The tape

All five timeframes but one are in uptrends and the composite reads strongly bullish. Price is 7.2% above its 20-day average of C$1.38, 25.4% above its 50-day of C$1.18, and 59% above its 200-day of C$0.93. Daily RSI is 65.6, weekly 72.2 — and monthly RSI is 86.9, which is about as stretched as that indicator gets.

A stock 59% above its own 200-day average with a monthly RSI near 87 is not describing a name that is being discovered. It is describing one that has already been.

Risk-reward

Risk-reward 45. This is the one place the technical picture genuinely helps: the daily ATR is C$0.076, about 5.1% of price, and the nearest support at C$1.37 is only 1.5× ATR below — comfortably inside the framework’s 2.5 line. So a stop can do real work here, which is unusual for this report and is reflected at §12.

What it does not fix is the shape of the distribution below that support. The next meaningful level is C$1.25, and beneath it the chart is mostly air back to the C$0.99–C$1.02 shelf, which is where the shares traded in early August. Beta is 1.91.

Sentiment and catalysts

Sentiment rests on a panel of two, and it disagrees with this report. An earlier draft stated there was no analyst panel at all. There is: coverage of two, both Strong Buy, mean target C$1.65 — above the C$1.48 stamp, and well above this report’s C$1.15 base case. Two analysts is a thin panel and it is weighted as one, but it is a real reading and it points the other way. It is recorded here rather than buried because a report arguing the shares are dear owes the reader the fact that the only two people paid to cover them disagree.

What the news scan does show, in four articles, is the arc: the gold loan in October 2025, the “on budget, on schedule” update, and the first pour this morning. All of it good, all of it now behind the share price.

The catalyst calendar is genuinely light. The live process is the ramp to 3,000 tonnes a day and commercial production in Q4, which reports as it happens rather than on a date. The dated item worth watching is the updated six-deposit resource expected by year-end. One clear catalyst in a thirty-day window puts the clustering score at 62 — the framework’s focused band, normal position size. It modulates sizing, and below 30 would dock timing confidence; it is not the catalysts term in the timing formula, which is set separately at 55.

Upcoming events (next 30 days)

Recent surprises (last 7 days)

The company calendar is thin and everything on it has already happened. The one item that could move the valuation anchor is the six-deposit resource update due by year-end.

Four of five frames up, and the composite is as bullish as this report has recorded for the name. Read the RSI column before drawing comfort from it: 86.9 on the monthly frame is an extreme reading, and 72.2 on the weekly is not far behind. The monthly and weekly rows also rest on volume ratios of 0.28× and 0.48× on partial periods, so they carry less than they look. The useful number in this table is not a trend label — it is that support at C$1.37 sits only 1.5 ATR below the price, which is what makes a stop workable at §12.

Cabral Gold daily closes on the TSX Venture Exchange in Canadian dollars, 10 March to 9 September 2026, with the 50-day average. The series ends at the 9 September close of C$1.48 — the last completed session — and therefore does NOT include the 10 September first-pour announcement or the move that followed it. The near-vertical final stretch is the 42% re-rating this report is largely about.

Probability-weighted 12-month value: C$1.23 (0.20 × C$1.84 + 0.45 × C$1.22 + 0.35 × C$0.88), or −17% from C$1.48.

Sit with the shape rather than the number. The bull case is +24% and requires two independent things to go right; the bear is −41% and requires only one thing to go wrong. That asymmetry is not a view about Cabral — it is the arithmetic of paying 1.27× net asset value for a business with one mine, six years of starter reserves and a US$45m obligation repayable in gold.

A negative expected return sits oddly under a HOLD, so be precise about what it is and is not saying. It is not a SELL: that needs the in-situ mark below about US$36 an ounce, a level nobody argues for. It says the probability-weighted value is below the price on the report’s own deliberately conservative mark — and §4 concedes that mark is below the peer average it cites, so the expected return is the low end of a defensible range rather than a central estimate. It does not say the mine will fail, the resource is bad, or management has erred — §3 says the opposite on all three. It says the share price has moved further than the asset value, and that the gap is large enough to matter over twelve months.

Forecast: Zero of three groups are fully met — the Technical group fails only on its own overbought test, which is a fair reading of a monthly RSI at 86.9. The ladder is Wait. That is not a comment on the company, which §3 rates highly; it is a comment on paying 1.27× net asset value for it.

Forecast: By the next update on 24 September the ramp will be two weeks further on and gold will have been through the 16 September FOMC. My expectation is that Cabral is still a HOLD at that point — the valuation gap does not close in a fortnight, and nothing in the ramp is due to move the anchor before the year-end resource update. I would be wrong if the year-end resource update is pulled forward and lands materially above 1.19 Moz, which would raise the §4 anchor directly — or if gold clears US$5,000, which would raise the anchor by about C$0.13 a share in total — roughly C$0.02 per US$100 of gold, net of the metal-denominated loan.

Wait — zero of three entry groups met, and a HOLD on every horizon. There is no new position to size here.

For an existing holder, which is the case that matters, since Cabral has been a Donatien Pick since June: the framework says take the profit. The profit-target rule fired when the shares passed C$1.17, and they are now 27% above it. A holder from below C$1.00 has roughly doubled their money on a company that has executed everything it promised.

The argument for holding anyway is that the year-end six-deposit resource update could reset the §4 anchor upward, and that selling a newly-minted producer into a stagflationary gold market is the sort of decision that reads well for a quarter and badly for a cycle. That argument is legitimate. It is also a bet on a document that does not exist yet, at 1.27× the value of the assets that do.

If you hold, hold deliberately and small, with the C$1.37 level at §12 as a real line rather than a decoration — it is only 1.5 ATR below the price, which is the one genuinely favourable technical fact in this report.

And note what this report is not saying. It is not a call on the mine, the ore body, the jurisdiction or the management. It is a call on the price.

{
  "version": "v6",
  "brand": "",
  "ticker": "CBR.V",
  "company": "Cabral Gold Inc.",
  "currency": "CAD",
  "exchange": "TSXV",
  "exchange_ticker": "TSXV:CBR",
  "api_ticker": "CBR.V",
  "isin": "CA1271061022",
  "date": "2026-09-10",
  "analysis_status": "stopped",
  "macro_report_date": "2026-09-09",
  "primary_signal": "HOLD",
  "signal_short": "HOLD",
  "signal_medium": "HOLD",
  "signal_long": "HOLD",
  "base_signal_short": "HOLD",
  "base_signal_medium": "HOLD",
  "base_signal_long": "HOLD",
  "composite_short": 57,
  "composite_medium": 55,
  "composite_long": 55,
  "quality_score": 59,
  "valuation_score": 46,
  "timing_score": 61,
  "driver_score": 62,
  "driver_score_short": 60,
  "driver_score_medium": 62,
  "driver_score_long": 64,
  "driver_name": "The gold price",
  "driver_label": "Neutral",
  "economic_alignment_conviction": 76,
  "economic_alignment_stance": "Supportive",
  "economic_alignment_pressure": "tailwind",
  "economic_alignment_source": "MacroDriver-state-20260909 (Materials XLB Overweight on all three horizons)",
  "quality_confidence": 55,
  "valuation_confidence": 40,
  "timing_confidence": 50,
  "driver_confidence": 55,
  "economic_confidence": 65,
  "overall_confidence": 40,
  "overall_confidence_basis": "min(quality 55, valuation 40, timing 50) = 40 per SKILL L1329.",
  "price_at_rating": 1.48,
  "price_basis": "9 September 2026 TSXV close, the last COMPLETED session. It PREDATES the 10 September first-pour announcement; the shares traded near C$1.54 intraday afterwards on 3.6x normal volume.",
  "usd_cad_rate": 1.37775,
  "usd_cad_rate_asof": "2026-09-09 close",
  "lifecycle_stage": "materials gold just-commissioned single-asset producer",
  "lifecycle_note": "CHANGED from 'pre-production-developer'. First gold poured 10 Sep 2026. This is why Gates 1 and 5 are retired rather than carried: the going-concern and permitting arms fire on every PRE-production miner and would have been inherited unexamined.",
  "moat_score": 44,
  "moat_build": "pricing_power 20 (pure price-taker) + network_effects 50 (N/A-neutral; the SKILL names a gold miner as its own example) + switching_costs 25 (the SKILL names 'commodities' explicitly in the 0-49 band; a gold buyer has none) + cost_advantage 78 (72.6% AISC margin, but an oxide heap leach is replicable and depletes in 6.2 years) + intangible_assets 45 (operating licence + district land package; routine) = 218/5 = 43.6 -> 44. An earlier draft published moat 50 with NO components shown, and a draft before that cut it to 22 unevidenced.",
  "quality_build": "bench 90 x0.20 + moat 44 x0.20 + roic_on_projects 78 x0.10 + reserve_life_index 35 x0.10 + balance_sheet 62 x0.10 + cash_generation 50 x0.10 + revenue_trajectory 50 x0.10 + profitability_vs_peers 50 x0.10 = 59.3 -> 59. ROUND 7 CORRECTION, and it runs AGAINST this report's own conclusion, which is why it is a correction rather than steering: cash_generation was scored 25 - PENALISED for having no operating history - while revenue_trajectory and profitability_vs_peers were scored 50 as N/A-neutral for exactly the same fact. Three sub-signals, one underlying condition, treated two ways. All three are now N/A-neutral. Round 4 published 'NO defensible weighting reaches the 65 boundary'. THAT WAS FALSE and was withdrawn in round 5: from components this report itself publishes as defensible - moat at the previously-published 50, roic 85, reserve-life de-weighted - a build reaches the high 60s. THE HONEST STATEMENT: the defensible range spans roughly 57 to 68 and the 65 matrix boundary sits INSIDE it, so Quality is INDETERMINATE between medium and high on the published record. The published 59 is the straight reading of the framework's own sub-signal list; a reader who weights differently can reach high. That indeterminacy is the report's weakest leg and is stated as such rather than argued away. DISCLOSED ASYMMETRY: this report treats Cabral as a producer for gate purposes AND scores it against producer benchmarks (>8yr reserve life). A reader who rejects that symmetry scores Quality higher still.",
  "competitive_share_trajectory": "improving",
  "competitive_threat_level": "moderate",
  "industry_benchmark_name": "AISC margin (gold mining)",
  "industry_benchmark_value": 72.6,
  "industry_benchmark_score": 90,
  "eps_trailing": -0.05,
  "trailing_pe": null,
  "trailing_pe_note": "N/A - trailing EPS is negative (-C$0.05) and the company's first revenue was earned on 10 September 2026. An earnings multiple does not exist yet and none is published.",
  "price_to_book": 27.5,
  "shares_basic_m": 346.173703,
  "shares_fully_diluted_m": 415.024289,
  "shares_in_the_money_m": 397.732912,
  "exercise_proceeds_cad_m": 26.073,
  "share_count_note": "CORRECTED IN ROUND 4, and the error was self-inflicted. Basic 346,173,703 and fully diluted 415,024,289 per the company investor page, reconciling exactly (346,173,703 + 17,515,000 options + 48,898,993 warrants + 2,436,593 RSUs). An earlier draft used the 9 April structure of 301,723,341 / 364,072,658 and DISMISSED get_yahoo_quote's implied 346.2m as stale - the provider was correct to the share and the report reasoned away the right answer, because the megacap-sharecount heuristic says provider counts go stale. It cuts both ways: reconcile, never dismiss. VALUATION USES THE TREASURY METHOD - 397,732,912 in-the-money shares, excluding the 17,291,377 warrants struck at C$1.70 while the stock is C$1.48. Including them adds C$29.4m of proceeds at C$1.70 a share against a NAV near C$1.17, flattering NAV per share on an exercise nobody would make: it yields P/NAV 1.245x - still a HOLD, so the verdict consequence is nil. It is rejected because it credits an exercise nobody would make at a C$1.70 strike against a C$1.48 price, not because of where it lands.",
  "placement_2026_08": "MISSED BY ROUNDS 1-3, FOUND BY THE ROUND-3 AUDITOR. Cabral closed a C$44,957,580 non-brokered private placement on 24 August 2026 - 34,582,754 units at C$1.30 to Alpayana S.A.C., a Peruvian miner, for 9.99% undiluted (14.27% partially diluted), plus 17,291,377 warrants at C$1.70 expiring 24 Feb 2028. Alpayana may nominate one director while it holds 9.99%. Proceeds fund exploration and development including Phase 2. THREE CONSEQUENCES: (1) the share count above; (2) net debt of C$44.95m (total debt C$67.365m less cash of C$22.417m at the last reported balance sheet, 30 June 2026, which PREDATES the raise) becomes approximately NIL once the C$44.958m is added - construction and ramp spend since 30 June is NOT deducted, so this treatment FLATTERS the NAV and the true ratio is a little higher than published; (3) it is evidence AGAINST this report's own central assumption - a strategic industry buyer paid C$1.30, above the C$1.17 NAV computed here, with proceeds earmarked for the very Phase 2 this report marks at US$75/oz because it has no study, no capex and no permit. Published in section 4 as the counter-argument it is.",
  "valuation_model": {
    "basis": "P/NAV",
    "unit": "CAD_m",
    "share_count_field": "shares_in_the_money_m",
    "per_share_field": "nav_per_share",
    "terms": [
      {
        "id": "stage1_npv",
        "label": "Stage 1, marked to spot",
        "sign": "+",
        "kind": "asset_pv",
        "value": 355.2,
        "value_native": 257.81,
        "fx": 1.37775,
        "source": "src-pfs-2025",
        "nets": [
          "tax",
          "royalty",
          "sustaining_capex"
        ],
        "means": "the present value of the Stage 1 mine at spot gold, after tax, royalty and sustaining capital, WITH the US$37.7m of initial capital ADDED BACK because the study nets it as a future outflow and it is already spent - filed property and equipment went C$1.48m to C$57.83m"
      },
      {
        "id": "insitu",
        "label": "Remaining ounces in situ",
        "sign": "+",
        "kind": "asset_insitu",
        "value": 82.7,
        "source": "src-resource",
        "means": "the 1.06 Moz Stage 1 does not touch, marked at US$75 an ounce with inferred at half weight, which is the disciplined treatment for rock carrying no study, no capital estimate and no permit"
      },
      {
        "id": "net_cash",
        "label": "Net cash after the August placement",
        "sign": "+",
        "kind": "cash",
        "value": 0.0095,
        "source": "src-filed-interims",
        "means": "filed cash of C$22.417m plus the C$44.958m raised on 24 August, less the C$67.365m gold loan carried at book - a liability this term subtracts ONCE and no other term nets"
      },
      {
        "id": "proceeds",
        "label": "In-the-money exercise proceeds",
        "sign": "+",
        "kind": "dilution_proceeds",
        "value": 26.073,
        "source": "src-july-deck",
        "means": "cash arriving if the in-the-money options and warrants are exercised, credited alongside the shares they would create, with the C$1.70 warrants excluded from BOTH the count and the proceeds"
      }
    ],
    "no_double_count": "Checked pairs. initial_capex x stage1_npv/net_cash: the PFS NPV nets it as a future outflow and the gold loan that financed it is subtracted in net_cash, so it is ADDED BACK inside stage1_npv - counted once. Rounds 1-5 did not add it back and charged it twice, worth about C$0.13 a share. shares x proceeds: both use the in-the-money count of 397,732,912, so no proceeds are credited from instruments excluded from the denominator. tax/royalty/sustaining_capex: netted inside stage1_npv only, and no separate term carries them."
  },
  "nav_per_share": 1.17,
  "nav_basis": "COMPUTED BOTTOM-UP; corrected in round 6 against Cabral's FILED interim statements for the period ended 30 June 2026 (board-approved 26 Aug 2026). STAGE 1 = the July-2025 Ausenco PFS after-tax NPV marked to spot by interpolating between two of its five published sensitivity points (US$73.9m at US$2,500/oz, US$138m at US$3,340/oz; slope US$76.3m per US$1,000/oz) = US$220.11m at US$4,416, PLUS the US$37.7m of initial capital the study nets as a FUTURE outflow and which is now SPENT - filed property and equipment went C$1.48m to C$57.83m and the plant is pouring. Deducting the gold loan that financed the build AND using an NPV that already subtracted it charges the capex twice; rounds 1-5 did exactly that. Stage 1 = US$257.81m = C$355.20m at USD/CAD 1.37775. REMAINING OUNCES = the current resource (666,382 indicated + 525,669 inferred) less Stage 1's 128,903 contained Probable Reserve ounces, indicated at full weight and inferred at half, at US$75/oz = C$82.70m. NOTE THE BASIS: the half-weighting means the effective mark over the full 1,063,148 ounces is about US$56/oz, NOT US$75, and US$56 is below the US$88 listed-developer average this report cites. It is the most conservative defensible mark, not a neutral one. CASH AND DEBT, BOTH FROM THE SAME FILED PAGE. Cash and cash equivalents C$22,417,212. Gold loan payable C$18,577,432 current + C$48,787,841 non-current = C$67,365,273. Round 5 wrongly preferred the July deck's 'C$35M at June 29' - which reconciles to cash + C$6.58m gold deposits + C$6.43m non-current advances, and 29 June is not a period end - while taking the liability from the filing. Adding the C$44,957,580 placement of 24 Aug gives net debt of approximately NIL. Gold deposits and advances are NOT credited, a further conservatism. PROCEEDS = treasury method, in-the-money only: 17,515,000 options at C$0.46 + 31,607,616 warrants at C$0.57 = C$26.073m (July 2026 corporate presentation). The 17,291,377 warrants struck at C$1.70 are out of the money at C$1.48 and excluded from both count and proceeds. NAV = C$463.98m / 397,732,912 = C$1.1666; at C$1.48 that is 1.269x - the FULL band. Gate 3 does not fire (it needs 1.40). The SKILL's Materials/Miners guardrail line is P/NAV 1.5x, which this PASSES. TREATMENT MATRIX, each recomputed rather than carried forward: in-the-money only, as published, 1.269x (NAV C$1.1666); including the out-of-the-money warrants 1.245x (C$1.1888); basic shares only 1.170x (C$1.2650); strict treasury stock method 1.285x (C$1.1520). ALL FOUR return HOLD - the verdict no longer turns on the dilution treatment, which it did while the capex was double-counted. That is the difference between a conclusion that survives its own sensitivity and one that needed four conservatisms to break the same way.",
  "val_multiple_basis": "P/NAV (asset lens for a miner; earnings multiples do not exist here)",
  "sector_guardrail_note": "STATED RATHER THAN SKIPPED. The SKILL's sector guardrail line for Materials/Miners is P/NAV 1.5x. At 1.269x this name PASSES it, with room. lint_report's guardrail arm stays silent because it keys on a P/E or EV/EBITDA basis string and this name is valued on assets - that is the arm not applying, NOT the guardrail being unexamined. Do NOT silence the skip by declaring a P/E basis: forward_pe is 8.11 and against an EV/EBITDA line that would force a FALSE Expensive on a company whose first revenue was earned on 10 September.",
  "actual_multiple": 1.27,
  "warranted_multiple": 1.0,
  "warranted_ratio": 1.27,
  "val_band": "full",
  "warranted_note": "1.0x is GENEROUS by construction. BMO (Jan 2026) puts producers at 0.7-0.9x NAV and developers below that, so a single-asset business hours into its first pour would ordinarily warrant LESS than 1.0x. At 0.8x the ratio would be 1.59, which WOULD be Expensive and WOULD fire Gate 3 - stated in section 2 rather than glossed.",
  "fair_value_est": 1.17,
  "target_price": 1.22,
  "market_implied_insitu_usd_per_oz": 142.0,
  "market_implied_insitu_note": "Solved from the price: at C$1.48 fully diluted, after crediting Stage 1's entire NPV at spot and adjusting for net debt and proceeds, the market marks the remaining ~1.06 Moz at roughly US$142/oz. AN EARLIER DRAFT compared that to a 'customary US$30-100/oz' band that was UNSOURCED - the same failure this report criticises elsewhere. Published 2026 work puts advanced developers holding a PEA or PFS at about US$100-300/oz, with a listed-developer average near US$88. So US$142 sits INSIDE the advanced-developer band rather than at 1.8x the top of a junior one. The honest statement: the market marks unstudied rock below the midpoint of what advanced developers command. Stretched, not absurd.",
  "scenario_bull_target": 1.84,
  "scenario_base_target": 1.22,
  "scenario_bear_target": 0.88,
  "scenario_probabilities": {
    "bull": 20,
    "base": 45,
    "bear": 35
  },
  "probability_weighted_target": 1.23,
  "mtf_confluence": 74,
  "risk_reward_score": 45,
  "relative_strength_vs_spy": null,
  "relative_strength_vs_sector": null,
  "relative_strength_note": "Not computed as a benchmark spread this run. The informative comparison for a junior gold producer is against its own metal and its own history, and both are published: the shares are +42.3% since the 4 August report while gold rose 4.1% from the same base, they sit at the 91st percentile of a C$0.40-C$1.59 52-week range, and they are 59% above their own 200-day average. RS is not a term in the timing_score formula (SKILL L845).",
  "catalyst_clustering_score": 62,
  "catalyst_clustering_note": "One clear catalyst in a 30-day window - the updated six-deposit resource due by year-end. The ramp to commercial production is a PROCESS, not a date, and the 16 Sep FOMC is macro rather than company-specific. That is the framework's FOCUSED band (50-69): normal size, tighten around the event. It modulates sizing and, below 30, timing confidence; it is NOT the catalysts term in the timing formula, which is the FAVOURABILITY sub-signal and is set separately at 55.",
  "dynamic_macro_weight": 0.2,
  "timing_build": "mtf 74 x0.30 + risk_reward 45 x0.20 + macro 76 x0.20 + sentiment 45 x0.15 + catalysts 55 x0.15 = 61.4 -> 61. An earlier draft published this same string against a score of 58; the terms sum to 61.4, and the SCORE was corrected to match its own derivation rather than the reverse.",
  "sector_median_note": "BMO Capital Markets (January 2026): gold PRODUCERS trade at 0.7-0.9x NAV, royalty and streaming names at 1.5-2.0x. Junior single-asset producers sit at or below the producer band on execution risk. Cabral at 1.27x is well above it. Peer band SOURCED; the Cabral figure computed in section 4.",
  "own_history_percentile_52w": 91,
  "gold_price_usd": 4416.0,
  "gold_vs_50dma_pct": 3.8,
  "gold_vs_200dma_pct": -2.4,
  "driver_trend_note": "Gold is ABOVE its 50-day and BELOW its 200-day - a bounce inside a broken trend, not a confirmed tailwind. driver_score 62 therefore sits below the 65 at which a signal amplifies. That discipline costs nothing here, because amplification acts only on a BUY or a SELL and this report publishes neither.",
  "economic_study": "Updated Prefeasibility Study, July 2025, Stage 1 gold-in-oxide starter operation. After-tax NPV US$73.9m / IRR 78% at US$2,500/oz; US$138m / IRR 139% at US$3,340/oz. Initial capex US$37.7m, payback 10 months at base case. 113,000 oz over 6.2 years. CONFIRMED TO EXIST - the 4 August report classified this name 'pre-revenue' with 'anchor N/A' fourteen months after the study was published, which is the [[economic-study-existence-check]] failure that previously flipped CCI.CN.",
  "resource_indicated_koz": 666.382,
  "resource_inferred_koz": 525.669,
  "resource_note": "SOURCE: cabralgold.com/cuiu-cuiu-project, INDEPENDENTLY CONFIRMED against the primary page and corroborated verbatim in the 10 September 2026 first-pour release (an earlier draft cited this report's own auditor, which is circular provenance). Indicated 666,382 oz and Inferred 525,669 oz - the company's CURRENT statement, combining a hard-rock estimate effective October 2022 with oxide estimates updated October 2024 and July 2025. Primary/oxide split: indicated 450,200 oz at 1.14 g/t plus 216,182 oz at 0.50 g/t; inferred 455,100 at 1.04 g/t plus 70,569 at 0.34 g/t; 45.88 Mt for 1.19 Moz. AN EARLIER DRAFT used the superseded 2022 GLOBAL MRE of 604.0/534.5 koz.",
  "first_pour": "10 September 2026, ~1,130 oz, ahead of the 2025 PFS schedule. Operating licence 13 Aug 2026. Ramping toward 3,000 t/d; commercial production guided Q4 2026. ANNOUNCED THE MORNING AFTER THIS REPORT'S STAMPED PRICE.",
  "analyst_consensus_target": 1.65,
  "analyst_coverage_count": 2,
  "analyst_grades_consensus": "Strong Buy",
  "analyst_bullish_pct": 100.0,
  "analyst_grades_n": 2,
  "analyst_grades_detail": "A PANEL EXISTS and an earlier draft of this report said it did not. get_yahoo_analyst_targets(CBR.V) returns n=2, mean C$1.65, range C$1.60-C$1.70, distribution 2 strongBuy, recommendation 'buy'. The 4 August calibration had ALREADY recorded analyst_consensus_target 1.6 on coverage of 1 - so this run DELETED a datum rather than failed to find one. Two analysts is not a consensus and the 15% lens is discounted heavily, but the mean is 11.5% ABOVE the stamped price and 43% above this report's own base case. Published in section 4 as the counter-argument it is.",
  "earnings_quality_note": "No operating history to decompose - first revenue was earned 10 September 2026. A FORWARD estimate DOES exist, eps_forward C$0.19 for an 8.1x forward multiple, and section 4 states that declining to anchor on it is a CHOICE - one forward year on a 6.2-year starter pit says little about an ore body - rather than an absence. An earlier draft asserted no forward earnings base existed, which was wrong.",
  "clean_pe": null,
  "peg_computed": null,
  "implied_growth_pct": null,
  "eps_forward": 0.19,
  "forward_pe": 8.11,
  "gold_12m_return_pct": 21.2,
  "stage2_grade_note": "The mill feed is the PRIMARY hard rock: 450,200 oz indicated at 1.14 g/t and 455,100 oz inferred at 1.04 g/t, against Stage 1's Probable Reserve grade of 0.65 g/t - so Stage 2 is roughly 1.75x HIGHER grade, not lower. An earlier draft said the mill feed was '0.87 g/t, a fifth of the oxide starter's economics'; 0.87 is the superseded 2022 GLOBAL indicated grade and the comparison was inverted. What makes Stage 2 uncertain is capital and permitting, not rock quality.",
  "gold_loan_terms_note": "The US$45m gold loan repays 39 kg of gold per quarter from 31 March 2027 - a METAL-DENOMINATED liability carried at book while the entire asset side of this NAV is marked to spot gold of US$4,416. The NAV is therefore, if anything, FLATTERED at high gold prices. Disclosed because it cuts against this report's own conclusion.",
  "industry_benchmark_note": "AISC margin: the PFS all-in sustaining cost of US$1,210/oz against spot gold of US$4,416 = 72.6%. RESTORED. An earlier draft replaced this OPERATING benchmark with 'P/NAV (junior gold producers)' scored 25 - the VALUATION CONCLUSION inserted into the Quality pillar. The same 1.27x then drove Valuation to 38 AND dragged Quality below the 65 boundary, and since the matrix returns SELL for medium+expensive but HOLD for high+expensive, that double count was the entire difference between a SELL and a HOLD.",
  "gates_triggered": [],
  "gate3_withdrawn_note": "ROUND 5. Gate 3 fires at a warranted ratio of 1.40 and the corrected ratio is 1.27, so it DOES NOT FIRE. Earlier rounds published it as triggered on a NAV that omitted the 24 Aug placement cash and used superseded April strike prices. The valuation is Full rather than Expensive - still the dearest band this report has ever assigned Cabral, and still the reason there is no BUY.",
  "gates_caution": [
    "Gate 4 - Leverage (net debt near nil after the 24 Aug placement, but a US$45m gold-denominated obligation whose interest capitalises into principal until Dec 2026, on a mine that poured this morning)",
    "Gate 6 - Structural (single asset, single jurisdiction, 6.2-year starter life, Stage 2 unstudied)"
  ],
  "gates_retired_this_run": "WITHDRAWN IN ROUND 4. An earlier draft claimed Gate 1 (Financial Distress) and Gate 5 (Binary/Regulatory) were RETIRED from cautions the 4 August report carried. It carried neither: that calibration's gates_caution is ['Valuation Ceiling', 'Dilution / Leverage'] and gates_triggered is empty. The claim invented a prior state in order to take credit for improving on it. What is true without the invention: neither gate fires now - the company is funded, built and pouring, and the permitting binary resolved when the operating licence was granted on 13 Aug 2026.",
  "hard_gate_state": "caution",
  "do_not_buy_triggers": [],
  "dnb_note": "None fire, and trigger 2 deserves a word since the valuation IS the report. Trigger 2 fires on a valuation EXCESS against a resolvable anchor; that is exactly what Gate 3 has already adjudicated, and firing both would double-count one finding. The signal is HOLD on the matrix, which is already what a DNB trigger's cap would impose, so firing trigger 2 as well would add nothing. Trigger 4 (insider selling) cleared on a 4-article news scan read in full, with no disposal story.",
  "entry_groups_met": 0,
  "entry_conviction": "Wait",
  "short_entry_confirmed": false,
  "exit_groups_live": 1,
  "exit_action": "Trim",
  "stop_loss": 1.37,
  "stop_loss_note": "The C$1.37 support. Unusually for this framework it is a WORKABLE stop: 1.5x the daily ATR of C$0.0758 below the price, well inside the 2.5 comfort line. That is the one genuinely favourable technical fact in this report.",
  "next_update_date": "2026-09-24",
  "next_update_basis": "default +14d. The ramp to commercial production is a process rather than a date, and the resource update is guided only to year-end.",
  "prior_report_defects": "CORRECTED, and the correction is the important one. An earlier draft claimed the 4 August artifact had MISSED the July 2025 PFS and invoked the economic-study-existence-check precedent. FALSE, and withdrawn: the August calibration records the study in full - Ausenco 29 Jul 2025, NPV5 US$73.9M at US$2,500/oz, capex US$37.7M, AISC US$1,210/oz - and states the existence check was run. Its ACTUAL defect is narrower: it published val_band 'full' with warranted_multiple and warranted_ratio both null, which lint flags FW-BAND-NO-ANCHOR, because a band over an absent anchor vacuously satisfies the Gate-3 Expensive arm. This report COMPLETES the anchor; it does not discover the study. Claiming a delta without checking the committed prior artifact is what [[verify-deltas-against-committed-source]] forbids.",
  "threshold_fragility": "THE VERDICT IS HOLD. Rounds 1-4 published SELL; round 5 moved to HOLD for a reason that was itself wrong; round 6 corrected that reason and found a larger error running the other way; round 7 re-derived the scenarios and the Quality build. The call has been stable since round 5 and is robust to every input this report can test EXCEPT ONE, named at (2): the Quality build, whose defensible range straddles the boundary that decides the signal. The audit rounds since have moved prose and attribution, not the verdict. Round 9 narrowed this sentence to 'every NUMERIC input', which made it MORE precisely false - quality_score is a number, and item (2) says a defensible re-weighting of it returns BUY. That is the THIRD recurrence on this report of a self-certification its own next paragraph withdraws; it is recorded here rather than quietly corrected, because this is the field the next run diffs against. (1) THE VALUATION IS NOT NEAR A BOUNDARY ON THE WARRANTED MULTIPLE THIS REPORT USES - and section 2 now concedes the qualifier: at the 0.8x warranted multiple BMO's producer band would support, the ratio is 1.59, the band is Expensive and Gate 3 fires. P/NAV is 1.269x against the 1.40 Gate-3 line and the 1.5x Materials guardrail; both pass. All four dilution treatments return HOLD (1.170x-1.285x). A SELL needs the in-situ mark below about US$36/oz - a level nobody argues for. This was a knife-edge for four rounds only because the net asset value was deducting the mine twice. (2) THE QUALITY BUILD IS THE WEAKEST LEG AND IS INDETERMINATE. quality_score is 59; the defensible range spans roughly 57-68 and the 65 matrix boundary sits INSIDE it (see quality_build, where round 4's claim that no defensible weighting reaches 65 is WITHDRAWN as false). If Quality is high and Valuation Full the matrix returns BUY. This report does not go there because the straight reading of the framework's own sub-signal list gives 59 - but a reader who weights differently can reach high, and the indeterminacy cuts UPWARD, not downward. (3) EVERY UNRESOLVED MODELLING CHOICE RUNS ONE WAY, AND IT IS THE CONSERVATIVE ONE. The in-situ mark is US$56/oz effective against the US$88 listed-developer average this report itself cites; the PFS NPV is not rolled forward from its July-2025 base; gold deposits and non-current advances are not credited; gold deposits and non-current advances are not credited. Each of those is conservative. ONE ITEM RUNS THE OTHER WAY and an earlier round wrongly filed it with the rest: the gold loan is carried at its 30 June book value while every asset is at 9 September spot, which UNDERSTATES the liability by roughly C$0.017 a share and therefore FLATTERS the NAV. It is small, it is disclosed in gold_loan_terms_note, and it is the one input where the report's conservatism does not hold. and a reader should know the published figure is the low end of a defensible range rather than a central estimate. That is also why the probability-weighted return is negative under a HOLD. (4) GOLD. Assets are marked to US$4,416 spot; the US$45m loan is repayable in metal from March 2027 with 100% of interest capitalised to principal until December 2026. The scenarios now gear the loan with the metal, which damps the bull (net debt goes from nil to about C$9m at US$5,000 gold) as well as the bear.",
  "user_horizon": null,
  "user_allocation_pct": null,
  "portfolio_role": null,
  "post_audit_fixes_round1": "Round 1 from a fresh auditor: 4 BLOCKER, 10 MAJOR, 6 MINOR. The verdict FLIPS from SELL to HOLD on all three horizons, and the reason is an integrity failure of mine. B3, the one that matters: I replaced the Quality pillar's OPERATING industry benchmark - AISC margin, which the 4 August report scored 90 - with 'P/NAV (junior gold producers)' scored 25. That is the VALUATION CONCLUSION inserted into the Quality pillar. The same 1.45x multiple then drove Valuation to 38 AND dragged Quality below the 65 boundary, and because the Decision Matrix returns SELL for medium-quality-plus-expensive but HOLD for high-quality-plus-expensive, that double count was the ENTIRE difference between a published SELL and a published HOLD. I also cut moat_score 58 -> 22 with nothing in the text supporting it. Meanwhile the honest operating benchmark had IMPROVED: PFS AISC US,210/oz against spot US,416 is a 72.6% margin, better than the ~70% August scored 90. Quality restored to 68, moat to 50, and the 65 boundary is now named as fragility (1) - the input that actually decides this report, which an earlier draft said was the in-situ mark. B4, the second serious one: I claimed the 4 August report had MISSED the July-2025 PFS and invoked the economic-study-existence-check precedent and CCI.CN. FALSE. That calibration records the study in full - Ausenco 29 Jul 2025, NPV5 US.9M at US,500/oz, capex US.7M, AISC US,210/oz - and states the existence check was run. Its real defect was narrower: val_band 'full' published over a null warranted_multiple, which is FW-BAND-NO-ANCHOR. Withdrawn in all six locations. Claiming a delta without checking the committed prior artifact is exactly what the standing rule forbids, and I had stated the narrow defect CORRECTLY in prior_report_defects while asserting the wrong one in the prose. B1: I wrote that no analyst covers this name. get_yahoo_analyst_targets returns n=2, BOTH Strong Buy, mean C.65 - 11.5% ABOVE the stamped price and 43% above my own base case - and the 4 August calibration had already recorded 1.6 on coverage of 1, so this run DELETED a datum rather than failed to find one. On a report arguing the shares are dear, burying the only two analysts who cover it was the most damaging omission in the document. B2: the resource was the superseded 2022 global MRE; the company's current statement is 666,382 indicated and 525,669 inferred, and Stage 1 consumes CONTAINED reserve ounces (128,903) not recovered ones (113,155). MAJORS: the NPV slope was quoted per US/oz when it is per US,000/oz; the gearing was published as C/bin/zsh.55 a share per US of gold when it is C/bin/zsh.12, five times overstated, and it underwrote the 35% bear weight; the bear target's in-situ assumption was unstated while the bull's was stated - the one genuine thumb on the scale; the 'customary US-100/oz' in-situ band was UNSOURCED, the same failure this report criticises elsewhere, and published 2026 work puts advanced developers at US-300/oz with a listed average near US, which places the market's US INSIDE the band rather than at 1.8x its top; the Stage 2 grade claim was inverted - the mill feed is primary hard rock at 1.14 g/t against Stage 1's 0.65 g/t reserve grade, so it is 1.75x HIGHER grade, not a fifth; eps_forward C/bin/zsh.19 and an 8.1x forward multiple exist and I asserted no forward earnings base did; the Technical entry group rendered MET while the report said zero of three; timing_build summed to 61.4 against a published 58, and the SCORE was corrected to match its own derivation. MINORS: 'three weeks into its first pour' in four places when it poured that morning; gold's twelve-month return 24.4% -> 21.2%; the 52-week high was a post-stamp intraday print; and the gold loan repays 39 kg of gold quarterly from 31 March 2027 - a metal-denominated liability carried at book while the asset side is marked to spot, which FLATTERS the NAV and is now disclosed because it cuts against this report's own conclusion.",
  "post_audit_fixes_round2": "B1 THE RENDERED NARRATIVE LAYER WAS NEVER REGENERATED. Round 1 rewrote calibration_json but not sections/*, so 25 instances survived in the published prose: 5 SELL-verdict sentences, 5 'no analyst panel' claims, 5 superseded-resource figures, 3 'the August report said the study did not exist' claims, the unsourced US$30-100 in-situ norm, gold +24.4% (correct 21.2%), proceeds C$32m (correct C$29.5m) and 1.02 Moz (correct 1.06). All located by regex sweep of the BUILT artifact rather than from the finding list, and all fixed at source in the spec builder with assert-before-replace. B2 THE SECTION-4 SURFACE TABLE WAS WRONG IN 3 OF 5 ROWS AND ITS CENTRAL CLAIM WAS FALSE. The draft asserted 'no cell of this surface produces a BUY and none produces a SELL' and that the in-situ mark decided nothing. Verified against lint_report's own _base_signal: at the then-published quality 68 the US$100/150/200 rows returned BUY. The mark IS decision-relevant. Restated at the corrected quality: SELL below ~US$90/oz, HOLD above it - and section 4 now publishes the tension that the market's own implied US$176/oz sits INSIDE the range advanced developers command, so the conservative mark rather than the arithmetic is what makes this a SELL. B3 QUALITY 68 WAS UNAUDITABLE AND UNREACHABLE. The auditor declined to certify a pillar deciding the verdict at a 3-point margin with no build published, and section 3 rendered none of the three mandated tables. Built explicitly from the SKILL's four universal sub-signals plus the Mining profile - which contributed an input the report had missed entirely, the >8-year producer reserve-life benchmark against Cabral's 6.2-year starter pit - quality comes to 55. Moat rebuilt from its five components to 44 (the SKILL names 'commodities' explicitly in the 0-49 switching-cost band, which the draft had scored N/A-neutral at 50). A five-case weighting sensitivity is published in quality_build: the range is 53.5-63.7 and only a maximally generous set reaches 66.0, so 68 was not reachable from any published build. Section 3 now renders the sub-signal build table and the moat scorecard. VERDICT FLIP HOLD -> SELL on all three horizons; composites reflowed 57/55/58 -> 54/51/51. This returns the signal to where round 1's draft had it, for an entirely different and now-published reason: round 1 reached SELL by inserting the P/NAV valuation conclusion into the Quality pillar (a real double count, correctly caught by round 1's auditor); this reaches it from an operating-only weighted build containing no valuation input. M1 pillar chips in sections/* were author-typed and had drifted from the calibration (valuation 35 vs 38, timing 58 vs 61); both now match. M2 the AISC caveat is published: US$1,210/oz is the study's figure struck at a US$2,500 gold assumption and is being divided into US$4,416 spot, so the realised margin is below 72.6%. It does not move the band, whose boundary needs an AISC above US$2,650. M3 exit_action was Trim while the rendered ladder said Exit; the Profit-Target group had BOTH conditions met but state not_live. Corrected to live, exit_groups_live 2, exit_action Exit - which also reconciles the -20.3% expected return the draft published against a HOLD. M4 bear delta -53% -> -52% (0.71/1.48-1); -53% was the delta from a superseded C$1.51. M5 resource_note cited 'the independent auditor of this report' as a data source - circular provenance. Replaced with cabralgold.com/cuiu-cuiu-project, corroborated in the 10 Sep first-pour release. M6 Gate 3 was described as 'the reason this report says SELL'. A valuation-ceiling gate caps a BUY at HOLD and cannot manufacture a SELL; the matrix produces it. Reframed as corroborating rather than causing. M7 the section-14 claim that a missing analyst panel cost '25% of the pillar' - the consensus-target lens exists and is used (n=2, mean C$1.65, both Strong Buy, pointing AGAINST this report); only grades distribution and PEG are absent, 10%. MINOR 'six weeks ahead of schedule' appears in neither primary source (both say only 'ahead of schedule'); softened in all 7 places. MINOR post_audit_fixes_round1 was shell-corrupted ('C/bin/zsh.55', 'US,210/oz') by an unescaped dollar sign in a bash-quoted argument; this round's text is written from a Python argument list with no shell involved.",
  "post_audit_fixes_round3": "B1 A MATERIAL CORPORATE ACTION WAS MISSED ENTIRELY BY ROUNDS 1-2. Cabral closed a C$44,957,580 non-brokered private placement on 24 August 2026 - 34,582,754 units at C$1.30 to Alpayana S.A.C., a Peruvian miner, for 9.99% undiluted plus 17,291,377 warrants at C$1.70 and one board nomination right. Three weeks before the stamp, and not one earlier draft mentioned it. get_stock_news returned four articles over 160 days and not this one, so the absence of a tool hit again proved nothing. Verified against the company release and newsfilecorp; the capital structure was cross-checked against the company investor page, which reconciles exactly to 415,024,289. CONSEQUENCE 1 - THE SHARE COUNT. Basic 346,173,703 (not 301,723,341), fully diluted 415,024,289 (not 364,072,658). Worse than a stale number: section 15 had SURFACED the correct figure - 'market cap implies 346.2m shares' - and reasoned it away as provider staleness, applying the megacap-sharecount heuristic in the one direction where it was wrong. The provider was right to the share. Reconcile, never dismiss. CONSEQUENCE 2 - THE BALANCE SHEET. Net debt C$44.95m (total debt C$67.365m less C$22.417m cash at the last reported balance sheet, 30 June 2026, which PREDATES the raise) becomes approximately nil. Quality sub-signal balance_sheet 40 -> 62, so quality_score 55 -> 57 - still medium, verdict unchanged. Ramp spend since 30 June is not deducted, so the treatment flatters the NAV. CONSEQUENCE 3 - THE NAV, REBUILT ON THE TREASURY METHOD. In-the-money only: 397,732,912 shares and C$24.143m of proceeds, excluding the 17,291,377 warrants struck at C$1.70 while the stock is C$1.48. NAV C$1.02 -> C$1.03, P/NAV 1.45x -> 1.44x, still Expensive, SELL STANDS. The treatment is now the decisive fragility and is published as such: crediting the out-of-the-money warrants gives 1.3976x and a HOLD; basic shares alone give 1.3274x and a HOLD. Treasury method is used because the alternatives either over-credit an irrational exercise or ignore 51.6m deeply in-the-money instruments. CONSEQUENCE 4 - THE THESIS, and it cuts against this report. A strategic industry buyer paid C$1.30 a share, ABOVE the C$1.03 NAV computed here, with proceeds earmarked for the very Phase 2 this report marks at US$75/oz on the grounds that it has no study, no capex and no permit. Published in section 4 as the strongest argument against the report's own conclusion, not buried. B2/B3 ROUND 2'S B1 WAS NOT FULLY FIXED - six survivors reached the rendered artifact. The changes box still read 'fall from BUY to HOLD'; the section-14 summary still read 'Quality is 68, high' and 'HOLD on all three horizons'; the section-4 and section-7 leads read 35 and 58 under chips of 38 and 61, both of which round 2's M1 had certified as 'both now match'; a live exit rule still cited the withdrawn unsourced 'US$30-100 norm'; and base_signal_short/medium/long were still HOLD, implying a gate caused the SELL - the exact inference round 2's M6 withdrew. Round 3's auditor noted its own first sweep used grep -o with context anchors and reported ZERO hits on a defect that was present because the match began its line - a plausible mechanism for how round 2's sweep missed these. ALSO: a DUPLICATE quality_confidence key (60 and 55) meant the later silently won; resolved to 55. The resource sourcing caveat is retired - round 3 confirmed the figures against the primary page independently. NET EFFECT ON THE SIGNAL: none. SELL/SELL/SELL stands, composites 54/52/52.",
  "post_audit_fixes_round4": "THE DIAGNOSIS MATTERS MORE THAN ANY FINDING: round 4's auditor showed the FIX MECHANISM was failing, not the sweep regex. Three times a fix corrected the clause it was pointed at and left the adjacent clause in the same sentence or the same box - section 14 fixed '57, medium' then said 'high business quality' 200 words later; the changes box fixed 'BUY to SELL' and left 'high business quality is a HOLD row'; section 7 fixed its prose lead 58->61 and left the chip 12 characters away at 58. A string sweep structurally cannot catch that. SO THE CLASS IS NOW ELIMINATED BY CONSTRUCTION, not detected: build_stock_report.derive_section_scores() derives each section's chip AND the leading number in its own prose from the calibration pillar, because they are the same quantity and should never have had two copies. Proven both ways - it caught the live section-7 chip drift (58 vs timing_score 61) on the first run, and fires on injected prose-lead and chip drift. This is the same move as derive_composites. B1 SECTIONS 13, 14 AND THE CHANGES BOX TOLD THE READER IN PLAIN ENGLISH THAT THE FRAMEWORK RETURNS HOLD, under a SELL/SELL/SELL header. Section 14 said 'at every in-situ mark on the section-4 surface the answer is HOLD' - round 2's B2 defect resurrected verbatim, contradicting the surface table 300 lines above it. Rewritten to the truth: SELL below about US$84/oz in situ, HOLD above. B2 THE 9 APRIL SHARE STRUCTURE AND THE C$44.9m NET DEBT SURVIVED IN SECTIONS 2, 5, 11, 12 AND 15 after round 3 corrected only section 4 and the calibration. All updated to 346.2m basic / 415.0m fully diluted and to net debt near nil. B3 THE C$0.44 AND C$0.52 STRIKES WERE UNSOURCED and section 15 attributed them to a page that carries no exercise prices. Now sourced to the May 2026 corporate presentation, and - more usefully - BOUNDED: the April pool's total exercise value is fixed at C$29.456m, exercised strikes cannot be negative, so the 49.1m remaining instruments cannot yield more than that, a maximum blended C$0.5996 against the C$0.702 needed to reach the 1.40 boundary. At the absolute bound the ratio is 1.4170x. THE VERDICT IS ROBUST ACROSS THE ENTIRE FEASIBLE STRIKE RANGE. M1 THE PLACEMENT WAS DEPLOYED ONLY AS EVIDENCE AGAINST THIS REPORT, and the omitted half supports it. Alpayana paid C$1.30 against a C$1.37 close on 21 August, with the shares at C$1.40 the day before - a DISCOUNT to market, not a premium. So the strategic paid above what this report thinks the assets are worth and BELOW what the market was charging, which is nearer this report's position than the market's. Both halves now published. M2 THE GOLD LOAN CAPITALISES 10% INTEREST INTO PRINCIPAL UNTIL DECEMBER 2026 - undisclosed, and it cuts against the 'net debt nil' claim. Disclosed in section 2, section 11, the Gate 4 caution and nav_basis. M3 gates_retired_this_run WITHDRAWN. It claimed Gate 1 and Gate 5 were retired from cautions the 4 August report carried; that calibration's gates_caution is ['Valuation Ceiling', 'Dilution / Leverage'] and gates_triggered is empty. The claim invented a prior state in order to take credit for improving on it. M4 quality_build's SENSITIVITY DID NOT REPRODUCE - it carried pre-placement figures ('as authored 54.6') inside a string computing 56.8, and two cases could not be reconstructed from the published weights. Recomputed as six cases from the stated weights: range 54.8-63.1, and NO defensible weighting reaches the 65 boundary. Also disclosed the asymmetry the auditor identified: this report treats Cabral as a producer for gate purposes AND scores it against producer benchmarks, worth roughly half the gap to 65. M5 the PFS publishes FIVE sensitivity points, not two; the linear extrapolation between two of them is sound and the independent check puts its error at about C$0.003 a share. Stated. TREATMENT MATRIX now published in nav_basis, including the point that the published treatment is NOT the most adverse available - the strict treasury stock method gives 1.4626x against the published 1.4353x - which answers the charge that the method was chosen to reach SELL. NET EFFECT ON THE SIGNAL: none. SELL/SELL/SELL stands on Q57/V38/T61, composites 54/52/52.",
  "post_audit_fixes_round5": "THE VERDICT MOVED SELL -> HOLD ON ALL THREE HORIZONS. Round 5's auditor showed the SELL was reachable but NOT ROBUST, and it was right. The SELL required FOUR independent conservative choices to break the same way, and each one alone returns HOLD: the in-situ mark, the cash figure, the option/warrant strikes, and the dilution treatment. Stacking four conservatisms and publishing the result as a SELL is a thumb on the scale, not conservatism. B1 THE ROBUSTNESS CERTIFICATE WAS FALSE. Round 4 published 'NO defensible weighting reaches the 65 boundary'. Round 5 falsified it from components this report itself publishes as defensible - moat at the previously-published 50, roic 85, reserve-life and cash de-weighted - reaching 66.8, which is HIGH quality and therefore not a SELL. Round 2 had honestly published 'only a maximally generous set reaches 66.0'; round 4 narrowed that range in the direction protecting the verdict. WITHDRAWN. The honest statement is that the defensible range spans roughly 55-67 and the 65 boundary sits INSIDE it, so Quality is INDETERMINATE between medium and high - which is itself a reason not to publish a SELL. B2 SECTION 13 STILL SAID 'a HOLD on every horizon' UNDER A SELL HEADER - round 4's B1 recurring for the fourth consecutive round, in one of the three sections round 4 named. THE CLASS IS NOW ELIMINATED IN CODE: build_stock_report._verdict_contradictions() refuses to ship prose that asserts a verdict the calibration does not publish, exempting sentences that openly correct an earlier draft or state a counterfactual. It found this exact sentence on its first run. M0 THE STRIKES WERE STALE AND MIS-ATTRIBUTED. The company's July 2026 corporate presentation - linked from the very investor page section 15 cited - publishes options 18,160,000 at C$0.46 and warrants 31,817,949 at C$0.57, not April's C$0.44/C$0.52, and no 'May 2026 corporate presentation' exists on the site. Corrected: proceeds C$24.143m -> C$26.073m. M0b THE CASH FIGURE. The same presentation states C$35M at 29 June 2026; the data provider reports C$22.417m. Taking the PRIMARY figure - the project rule, and the one whose breach lost the share count in round 3 - and adding the 24 Aug placement gives NET CASH of about C$12.6m rather than net debt of nil. Both are published, because the choice straddles the band boundary: primary gives P/NAV 1.386 (Full), provider 1.429 (Expensive). RESULT: NAV C$1.03 -> C$1.07, P/NAV 1.44x -> 1.39x, val_band expensive -> FULL, valuation_score 38 -> 44. GATE 3 NO LONGER FIRES (it needs 1.40). Matrix Q57/V44/T61 returns HOLD. Composites 56/54/54. Exit action Exit -> Trim, since the thesis-invalidation trigger was keyed to the 1.40 breach. analysis_status stays 'stopped' - no BUY on any horizon, so the Donatien-Pick auto-Stop fires at HOLD exactly as at SELL. M6 A BASIS ERROR IN THIS REPORT'S OWN CENTRAL COMPARISON, and it flattered the bear case. The NAV marks at US$75/oz on 800,314 HALF-WEIGHTED ounces while the market-implied figure is spread over 1,063,148 FULL-count ounces - apples to oranges by 1.33x. Restated like for like the report marks at about US$56/oz, which is BELOW the US$88 listed-developer average it cites. The half-weighting stays (it is standard), but section 4 now publishes the effective mark and concedes it is the most conservative defensible mark rather than a neutral one. ALSO FIXED: section 5's stale C$539m market cap (April count x price); Gate 4's '62.9m' against its own 68.85m itemisation; bear -53% vs -52%; bull and forecast triggers set at 1.14 Moz below the report's own 1.19 Moz; 'upper half of US$100-300' for a figure below the 200 midpoint (5 places); section 12's gold gearing C$0.14 per US$100 against the report's own C$0.12; section 8's superseded 604,000 oz; 'on budget' asserted with no primary source saying it (5 places); and the section-14 post-news ratio, published as 1.51x on a superseded NAV, now 1.44x. WHAT ROUND 5 CONFIRMED AND MUST NOT BE CHURNED: the NAV build to the cent, all four dilution treatments, the strike-bound argument (valid, and the subset premise now evidenced), Gate 3's framing, derive_section_scores(), and the full live-verification set - gold loan, placement, first pour, resource, PFS, capital structure, the 21 Aug C$1.37 close, analyst panel, gold and its moving averages.",
  "post_audit_fixes_round6": "HOLD IS CONFIRMED, AND ROUND 6 IS THE FIRST ROUND IN WHICH IT IS ROBUST. Round 5 reached HOLD for a reason that was itself wrong; round 6 corrected that reason AND found a larger error running the other way. The two are one finding split by sign, and fixing only the first would have flipped the report back to SELL on a double count. B4 THE CASH CHOICE WAS WRONG, AND IT REPEATED ROUND 3'S LESSON IN REVERSE. Round 5 preferred the July deck's 'C$35M cash at 29 June' over the provider's C$22.417m on the principle that primary beats provider. Cabral's FILED interim statements for the period ended 30 June 2026, board-approved 26 August, report cash and cash equivalents of C$22,417,212 - the provider figure WAS the filed figure. The deck's C$35M reconciles to cash + C$6.58m gold deposits + C$6.43m non-current advances, and 29 June is not a period end. Worse: the C$67,365,273 of debt this report deducts (C$18,577,432 current + C$48,787,841 non-current) comes from THAT SAME FILED PAGE. The liability was taken from the filing and the asset from a deck. M10 THE NET ASSET VALUE WAS DEDUCTING THE MINE TWICE - the largest single error in six rounds, and every earlier round carried it. Stage 1 is valued at the PFS net present value, which nets US$37.7m of initial capital as a FUTURE outflow. It is spent: filed property and equipment went C$1.48m to C$57.83m and the plant poured gold on 10 September. Subtracting the gold loan that financed the build while using an NPV that had already subtracted it charges the capital cost twice. Worth about C$0.13 a share against the cash dispute's C$0.03. Stage 1 corrected to US$257.81m = C$355.20m. RESULT: NAV C$1.07 -> C$1.17, P/NAV 1.39x -> 1.269x, valuation_score 44 -> 46, still the FULL band. Gate 3 does not fire and the SKILL's 1.5x Materials guardrail PASSES - state it rather than skipping it. ALL FOUR dilution treatments now return HOLD (1.170x to 1.285x, each recomputed rather than carried forward), where at the double-counted NAV two of four returned SELL. A SELL would now need the in-situ mark below about US$36/oz. The verdict no longer turns on any single treatment choice. B1 section 4 claimed 'SELL only below about US$55/oz' against its own table; the real boundary was US$71 then and US$36 now. B2 the section-12 Rule Forecast still forecast 'still a SELL'. B3 section 2 read 'which is the Full band. Expensive ... is a SELL row' in adjacent sentences - the fourth consecutive round of that class. M7 MY OWN CLAIM THAT THE CLASS WAS 'ELIMINATED IN CODE' WAS OVERSTATED, and round 6 was right to flag it. The verdict guard had no pattern for 'is a SELL ROW' - its own docstring's motivating case - and it ran only on prose, never seeing the section-14 calibration embed. Both fixed: the pattern is added and the guard now scans calibration VALUES too, excluding the audit-trail fields whose job is to quote history. It found B2 and B3 immediately. M3 the strike-bound argument did not re-derive: C$29.46m was arithmetic on superseded April strikes (correct C$31.82m) and the 1.417x limit was computed on the double-counted NAV. Recomputed: the bound gives 1.253x against the published 1.269x, so the strike uncertainty no longer discriminates at all. ALSO: the four-treatment matrix published in round 5's nav_basis was wrong in three of four figures - caught by recomputing them here rather than carrying them forward. The gold loan's interest capitalisation to December 2026 is CONFIRMED verbatim in the loan release, so the Gate-4 caution stands.",
  "post_audit_fixes_round7": "HOLD CONFIRMED AND NOW ROBUST. Round 7 returned 4 BLOCKER / 14 MAJOR and explicitly endorsed the verdict: 'HOLD stands. Do not flip it.' It also judged the NAV construction SOUND on a primary read of the PFS - the capex add-back, the share count, all five surface cells, the US$36/oz SELL boundary and all four dilution treatments re-derive independently. B1 SECTION 2 RENDERED GATE 3 AS TRIGGERED AGAINST A CALIBRATION SAYING IT DOES NOT FIRE - two fields of the same spec file (sections.gates[2].state='triggered' vs gates_triggered=[], hard_gate_state='caution'). It carried four false claims with it: 'not what produces the SELL', 'against an Expensive valuation' (val_band is full), 'the decision matrix independently returns SELL', and a 'one gate triggered' roll-up. Nothing cross-checked the rendered gate cards against the gate arrays. Gate 3 is now rendered clear, and its text says plainly that rounds 1-5 published it as triggered on a NAV that omitted the placement cash, used superseded strikes and deducted the mine twice. B2 THE SCENARIOS WERE NEVER RE-DERIVED AFTER ROUND 6'S NAV CORRECTION. Round 6 recomputed the dilution matrix 'rather than carrying it forward' and did not do the same here. Section 4 claimed that at US$5,000 gold the central-case NAV rises to about C$1.15 - arithmetically impossible, since NAV is ALREADY C$1.1663 at US$4,416. Every scenario is now re-derived from the corrected NAV with its multiple assumption stated: bull NAV C$1.51 (gold US$5,000 AND in-situ US$150, the scenario's own two levers) x1.22 = C$1.84; base NAV C$1.1663 x1.05 = C$1.22; bear NAV C$0.91 (gold US$3,600, in-situ US$50) x0.96 = C$0.88. The metal-denominated loan is now GEARED WITH GOLD in every scenario, which damps the bull (net debt nil -> about C$9m at US$5,000) as well as the bear. Targets 2.05/1.15/0.71 -> 1.84/1.22/0.88; probability-weighted C$1.18 -> C$1.23; expected return -20.3% -> -17%. B3 the horizon table published 'the matrix returns SELL' under signal_medium=HOLD. B4 section 12 said 'at 45% more than the mine is worth' two sentences from 'C$1.48 against a net asset value of C$1.17' (27%). M14 THE VERDICT GUARD'S SIXTH RECURRENCE, AND THE SECOND DISTINCT HOLE. dnb_note carried a live 'The signal is SELL on the matrix' - the exact string round 6 claimed to have fixed. The guard scanned the field and exempted it because _COUNTERFACTUAL matched 'would' SIXTY CHARACTERS AFTER the assertion ('a DNB trigger would impose'). TWO FIXES: the counterfactual exemption is now scoped to the text BEFORE the claim - a hypothetical frames its consequent before it, and this also de-fangs the 'at US$' token that round 7 called a near-blanket exemption on a commodity report - and a 'returns X' arm was added, which round 7 identified as this report's DOMINANT construction with no arm at all. Adding that arm alone made the guard fire on five legitimate exposition sentences, so a CONDITION-MARKER exemption was added with it ('returns SELL FOR medium+expensive' is a mapping, not an assertion). The guard then found B3, M14 and the section-2 claim immediately. M1 sections 1 and 15 published different pillar confidences and overall_confidence_basis was wrong in four figures ('min(quality 60, valuation 45, timing 50) = 45' against the actual 55/40/50 = 40). M6 gold gearing was published as C$0.12 a share per US$400 in two places; re-derived NET OF THE METAL-DENOMINATED LOAN it is C$0.09. M11 the no-BUY justification applied the short technical-confirmation cap to Medium and Long; report-audit.md L32 says Medium/Long unaffected - the real reason is Quality 59 against the matrix's 65. M3/M4/M5/M8/M9 stale figures: NAV C$1.03 survived x3, section 15 still cited a 'May 2026 corporate presentation' round 5 proved does not exist, the PFS slope was stated 10x wrong, and two calibration notes carried superseded ratios. M7 A SELF-INFLICTED ONE WORTH RECORDING: round 5's global 604,000 -> 666,382 substitution corrupted a sentence that was CORRECTLY naming the superseded 2022 figure, making it read 'the superseded 2022 global figures of 666,382' - which is the CURRENT indicated figure. An over-broad regex damaged correct prose in two places. Restored. THE ONE SUBSTANTIVE ANALYTICAL CHANGE, and it runs AGAINST this report's own conclusion: cash_generation was scored 25 - PENALISED for having no operating history - while revenue_trajectory and profitability_vs_peers were scored 50 as N/A-neutral for exactly the same fact. Three sub-signals, one condition, two treatments. All three are now N/A-neutral: quality_score 57 -> 59, composites 56/54/54 -> 57/55/55. Still medium, so the verdict is unchanged, and the direction of the change is the argument that it is a correction rather than steering.",
  "post_audit_fixes_round8": "ROUND 8'S HEADLINE IS THE ONE THAT MATTERS: post_audit_fixes_round7 ASSERTED FIVE FIXES THE ARTIFACT FALSIFIED - the bull multiple, the base target, the section-15 confidences, the PFS slope and NAV C$1.03. report-audit.md L50 calls that worse than the original error, because the audit block is what a reader is asked to trust. The cause is that round 7's closing sweep was keyed on the strings it already knew about rather than on the class, so a second copy of each survived. Round 8 verified round 7's B1, B3 and B4 DID land; B2 did not. B6 IS THE MOST SUBSTANTIVE FINDING IN EIGHT ROUNDS AND IT IS A HONESTY DEFECT, NOT AN ARITHMETIC ONE. This report said twice that the BUY-to-HOLD downgrade 'is entirely about price'. Running the framework's own _base_signal on the report's own delta_vs_prior: Q70/V46/T61 returns BUY. At AUGUST's Quality, with TODAY's valuation and timing, the matrix still returns BUY. The downgrade is therefore carried ENTIRELY by this report's re-derivation of Business Quality from 70 to 59 - a pillar section 3 itself calls indeterminate across 57-69, straddling the boundary that decides the verdict. The report was attributing to the market a move the analyst made. Corrected in both places, and stated plainly: the analyst moved, more than the asset did. B2 REMAINDER: the drivers section still said gold at US$5,000 takes the central-case NAV 'to about C$1.15' when NAV is ALREADY C$1.1663 at US$4,416 - the same defect round 7 fixed in section 11 and missed here. Corrected to C$1.30. Section 11's bull multiple rendered 1.47x (computed on the superseded NAV) against the correct 1.22x; the base return rendered -22%, the return on the superseded C$1.15 target. target_price was never moved off 1.15 when scenario_base_target went to 1.22. B5 THE GATE 3 STRESS TEST WAS FALSE TWICE. 'At 0.8x the ratio would be 1.59 - still inside the Full band, and still not a trigger' is wrong on both counts: 1.59 is above the 1.40 line, so it is Expensive and it WOULD fire the gate. warranted_note gave a third figure (1.83) computed on the old NAV. Both corrected, and the point is now made properly - the gate is clear on the warranted multiple this report uses, not on every defensible one. B7 SECTION 4 SAID 'No cell produces a BUY at medium quality' TEN WORDS AFTER 'at US$200 the matrix returns a BUY'. The table was right and the sentence was wrong. Round 8 also identified the cause: round 7's new _CONDITION exemption matched ' at medium quality' in its 44-character lookahead and suppressed the guard. TWO CODE FIXES, because round 8 correctly invoked the standing rule that a fix is code AND agent. (1) _gate_state_consistency() now cross-checks the rendered gate CARDS against gates_triggered and hard_gate_state and fails the build on a mismatch - round 7 PRESCRIBED this assertion, round 8 found it had never been written, and B1 had been fixed by hand and could therefore simply recur. Proven to fire on an injected mismatch. (2) A NEGATION exemption was added to the verdict guard: 'No cell produces a BUY' is the opposite of asserting BUY, and the guard had no way to tell. A NOTE ON MY OWN REPLACEMENT PROSE: the B6 correction initially tripped the verdict guard three times, because it reasons counterfactually ('at August's Quality the matrix returns BUY') without a marker the guard recognises. Rather than loosen the guard, the prose was reworded to say what it actually means - 'if August's Quality were held, the matrix WOULD return BUY'. More accurate English, and it trips the existing exemption. Loosening a guard to accommodate one's own prose is how a guard dies. ALSO: section 15 published Valuation 45% and Quality 60% against the calibration's 40 and 55; the section-15 PFS note stated the slope as US$76.3m per US$100/oz, ten times wrong; placement_2026_08 still carried NAV C$1.03; the analyst panel was compared to a superseded C$1.15 base case (35% above C$1.22, not 43%); and threshold_fragility filed the gold-loan-at-book item under 'every unresolved choice is conservative' when that item FLATTERS the NAV by about C$0.017 a share - the one input where the report's conservatism does not hold, now stated as such. VERDICT UNCHANGED: HOLD/HOLD/HOLD. Round 8 endorsed it explicitly ('No flip - HOLD stands') and confirmed the scenario multiples (1.22/1.05/0.96) and the 20/45/35 probability set re-derive.",
  "post_audit_fixes_round9": "ROUND 9 CONFIRMED ALL SEVEN OF ROUND 8'S BLOCKERS GENUINELY FIXED IN THE RENDERED HTML, and proved both code fixes fire rather than accepting the claim. It returned 3 BLOCKER / 9 MAJOR - down from 7/14 - and every one of the three was INTRODUCED OR SHARPENED BY ROUND 8'S OWN FIXES, which is report-audit.md L56 exactly. B1 THE FIX I ASKED THE AUDITOR TO CHECK HARDEST DID INTRODUCE A FRESH DEFECT. The new 'not about price' paragraph sits in the changes box; paragraph 5 of the SAME box, 250 words later, still opened 'Now the part that changes the signal' and ran straight into the valuation anchor. Both cannot be true and they share a screen. The opener now reads 'Now the anchor this report completes - which is not what moved the signal, but is what the signal is now measured against'. B2 THE NEW PARAGRAPH CITED SECTION 3 FOR A CLAIM SECTION 3 DOES NOT MAKE, and promoted an undemonstrated number above the fold. 'A pillar section 3 itself calls indeterminate across 57-69' - section 3 contains neither the word nor the range; that language lives in quality_build and threshold_fragility, which render in section 14. And NO PUBLISHED BUILD REACHES 69: the highest ever shown is 66.8, landing near 68 with cash_generation at 50. The auditor explicitly RAISED its own round-8 grade on this from MINOR to BLOCKER because round 9 had promoted it into the lead paragraph as the justification for the headline framing. Pointer corrected to section 14; range corrected to 57-68 everywhere. B3 SECTION 14'S POST-NEWS RATIO CONTRADICTED THE PARAGRAPH ROUND 8 REWROTE. It read 'at C$1.54 the ratio is 1.44 - still inside the Full band'. 1.54/1.1666 = 1.32, and 1.44 is NOT inside Full ([1.20,1.40)). So the report taught the band rule correctly in the newly-rewritten section 2 and broke it in section 14, on a number 9% too high. THE STRUCTURAL FINDING, AND IT IS THE RIGHT ONE: EIGHT OF ROUND 8'S FOURTEEN MAJORS WERE NEITHER FIXED NOR DISPOSITIONED. report-audit.md L47 requires every BLOCKER *and MAJOR* to be cleared before re-audit; I fixed the blockers and six majors and left eight standing, unmentioned. Two of them were made WORSE by fixes landing around them. All eight are now closed: the 1.40x/1.245x split in share_count_note (stale round-4 residue, and not innocuous - 1.40 is the exact Gate-3 line); the dilution rejection rationale in BOTH copies, which read 'we rejected a treatment because it gave the answer we publish' and is now stated properly (rejected because it credits an exercise nobody would make; the verdict consequence is nil); threshold_fragility's 'robust to every input' and 'NO LONGER NEAR A BOUNDARY', both falsified by the section 2 paragraph round 8 had just rewritten; THREE different answers to 'what decides this report', now separated into the two different questions they answer (what moved the SIGNAL = the Quality re-derivation; what the LEVEL is sensitive to = the in-situ mark; the dilution treatment decides neither); four SELL-frame residues that carry no verdict lead-in and so are structurally invisible to the guard; and section 3's 'the two universals scored 50' where the table beneath it shows three. THE CODE FINDING IS THE MOST IMPORTANT AND IT WAS MINE. The _NEGATED exemption added in round 8 REPRODUCED THE EXACT SCOPE BUG THAT ROUND 7'S FIX EXISTED TO CLOSE - it searched a 30-CHARACTER WINDOW before the claim rather than the words that govern it. The auditor demonstrated it against the live calibration rather than arguing it: 'No part of this is close: the signal is SELL on the matrix' and 'There is not much doubt left; the verdict is SELL' were both silently exempted. It also found that _CONDITION was NEVER NARROWED as round 8 asked - a third exemption had been added beside it instead, leaving four exemption patterns against one claim pattern. BOTH ARE NOW FIXED STRUCTURALLY RATHER THAN PATCHED. A shared _clause_before() helper scopes BOTH positional exemptions to the GOVERNING CLAUSE - text back to the last colon, semicolon, dash, or complementiser ('that'/'which') - so the bug cannot be fixed in one exemption and left in the other, which is exactly what happened last round. _CONDITION's bare 'at' token, which exempted round 8's blocker B7 ('produces a BUY AT medium quality'), now only qualifies when it introduces a level (at US$, at C$, at a digit). Proven 9/9 against a suite containing round 7's M14, all four of round 9's M8 reproductions, and four false-positive guards. VERDICT UNCHANGED: HOLD/HOLD/HOLD, endorsed again. The auditor agreed the 345 kg gold-loan balance should be LEFT ALONE - disclosed, immaterial, and not published in the report, so no reader is misled.",
  "post_audit_fixes_round10": "ROUND 10 RETURNED 0 BLOCKERS - the first round in ten with no defect in the report's numbers, framework application or conclusions. All three blockers and all nine majors from round 9 verified fixed in the rendered HTML, and for the first time the replacement prose introduced NO substantive defect. 2 MAJOR, both fixed here. MAJOR-1 THE FALSE-ROBUSTNESS CERTIFICATE, THIRD RECURRENCE ON THIS REPORT. Round 9 narrowed threshold_fragility from 'robust to every input' to 'robust to every NUMERIC input', which made it MORE PRECISELY FALSE: quality_score is a number, and item (2) of the same field says its 57-68 range straddles the deciding boundary and that a defensible re-weighting returns BUY (_base_signal(65,46,61) = BUY, confirmed). The opening sentence certified what the next two paragraphs withdrew. Prior recurrences: round 4's 'NO defensible weighting reaches 65' (withdrawn as false in round 5) and round 9's M3. It now reads 'robust to every input EXCEPT ONE, named at (2)' and records the recurrence, because this is the field the next run diffs against. MAJOR-2 SIX DEMONSTRATED HOLES IN THE CLAUSE-SCOPING - a framework finding, not an artifact one: the guard passed cleanly on the shipped prose and none of the six shapes appears in the document. _CLAUSE_BREAK carried no comma and no coordinator, so a negation or modal in a comma-joined clause still exempted a claim it did not govern ('Although nothing has changed, the matrix returns SELL'). And _COUNTERFACTUAL still carried 'at US$' and 'treatment', which are TOPIC WORDS on a commodity report rather than hypothetical markers - 'At US$4,416 spot, the matrix returns SELL' names TODAY's spot, the opposite of a counterfactual. THE AUDITOR'S DIAGNOSIS IS THE PART WORTH KEEPING: this is the THIRD ITERATION OF ONE BUG. Round 7 scoped sentence -> preceding text, round 8 used a 30-character window, round 9 used clause boundaries with an incomplete set. Each fix changed the UNIT and each kept the same structural assumption - that a governing word can be found by SEGMENTATION rather than by GOVERNMENT. The auditor validated its patch at 17/17 before recommending it; I re-ran the full suite independently and got 17/17 (six holes now caught, five historical reproductions incl. round 7's M14 still caught, six legitimate constructions still exempt). AND THE PATCH CAUGHT A LIVE ONE THE AUDITOR'S SUITE DID NOT. With commas as clause breaks the shipped prose fired once: section 4 read 'at US$200 the matrix returns a BUY' - a HYPOTHETICAL mark stated as fact. The guard was right and the prose was loose. Reworded to 'only at a US$200 mark WOULD the matrix return a BUY', and the same shape in section 11 with it. That is the second time the answer has been to fix the prose rather than loosen the guard. MINORS CLOSED: cash_generation's justification - the auditor accepted the mark of 50 but not the reason, and was right that negative free cash flow IS measurable here unlike revenue trajectory and realised margins. It now says so: what is not measurable is OPERATING cash generation, one day into production, and scoring a development-stage build-out as a cash-generation failure would penalise the company for having built the asset section 4 values. That matters because the 25->50 move is what took Quality to 59, and Quality is what the verdict turns on. Also fixed: a participle left dangling off 'section 14' by round 9's B2 fix, and section 4's 'about 1.0x' now names its basis (US$142/oz full-ounce; 1.09x on the half-weighted basis the surface uses). VERDICT UNCHANGED: HOLD/HOLD/HOLD. The auditor states it expects to return PASS next round.",
  "post_audit_fixes_round11": "PASS. Round 11 returned an explicit, unconditional PASS - 0 blockers, 0 majors, 10 minors logged and none blocking - after ELEVEN rounds. Both of round 10's majors verified fixed. A CORRECTION I OWE THE TRAIL, and it is the same class this report has been fighting, in miniature. post_audit_fixes_round10 said the US$200 rewording was applied 'and the same shape in section 11 with it'. The auditor diffed section 11 across both rounds, found it BYTE-IDENTICAL, and asked whether a substantive change had silently failed to land. It had not: BOTH edits were in the same SECTION 4 sentence ('a SELL would need the mark below about US$36 ... only at a US$200 mark would the matrix return a BUY'), and the note simply mislabelled the location. Nothing was missed and section 11 needed no change - but an audit-trail field named a section it did not touch, which is exactly the shape of round 8's B1 without the live defect behind it. Recorded rather than quietly dropped. THE AUDITOR CORRECTED ITSELF FIRST, AND MORE SEVERELY THAN I HAD. Its round-10 harness sampled isinstance(v, str) over spec['sections'] - a 16-member dict of which only four are strings - so it read 4 of 16 sections and NEVER SAW SECTION 4. Its 'no regression, shipped prose clean' claim was not evidence of anything. That is report-audit.md L65 (a cleared check is not evidence a check RAN), and it means the patch it recommended was better than its own test could show: the live section-4 defect the comma-patch caught was one its harness had no way of finding. ITS ANSWER TO MY QUESTION, AND I THINK IT IS RIGHT: the comma set IS still segmentation. It bought six true positives and costs at least five false ones, and the false ones are the most natural English forms of what the exemptions exist to permit ('If gold fell to US$3,000, the matrix returns SELL' now fires; 'No cell, on any reading of the surface, produces a BUY' now fires). The error did not go away - it moved from false-negative to false-positive. The reason is structural: every version looks BACKWARDS from the claim for a nearby marker, but a conditional is antecedent-INITIAL - 'if' at the start governs the whole sentence, comma or not - and looking backwards cannot see that. ITS RECOMMENDATION, WHICH I AM NOT IMPLEMENTING IN THIS ROUND AND HAVE CARRIED TO THE IMPROVEMENT PLAN INSTEAD: stop tuning the regex. Four exemption patterns now guard one claim pattern across six rounds of oscillation. INVERT THE BURDEN - keep the guard narrow and loud, and let the author mark deliberate exemptions in the artifact, so an exemption becomes a RECORDED DECISION rather than a regex inference. That is the same move as derive-don't-retype, applied to judgement instead of arithmetic. NINE MINORS LOGGED FOR THE NEXT CBR.V RUN, not fixed here: the audit_route_note duplication (now reproducing once per round), nav_basis's 'interpolating', the oxide/hard-rock characterisation, the relative-strength subtitle, a stale 'Quality 60', the 45/45/40 horizon confidences, and the undisclosed NPV base date. The durable fix for the confidences is extending derive_section_scores() to them - the same move that permanently closed pillar-chip drift. FINAL STATE: HOLD / HOLD / HOLD on Quality 59 / Valuation 46 / Timing 61; NAV C$1.17, P/NAV 1.269x, Full band; Gate 3 clear; analysis_status 'stopped' (Donatien-Pick auto-Stop, no BUY on any horizon); probability-weighted C$1.23, -17%.",
  "audit_route_note": "Round 1: a FRESH general-purpose agent (never a fork) \u2014 a FRESH general-purpose agent, never a fork. Verdict FAIL (4 blockers, 10 majors, 6 minors). Round 2: a FRESH general-purpose agent (never a fork) \u2014 a FRESH general-purpose agent, never a fork. Verdict FAIL (3 blockers, 9 majors, 4 minors). Round 3: a FRESH general-purpose agent (never a fork) \u2014 a FRESH general-purpose agent, never a fork. Verdict FAIL (3 blockers, 6 majors, 5 minors). Round 4: a FRESH general-purpose agent (never a fork) \u2014 a FRESH general-purpose agent, never a fork. Verdict FAIL (5 blockers, 18 majors, 9 minors). Round 5: a FRESH general-purpose agent (never a fork) \u2014 a FRESH general-purpose agent, never a fork. Verdict FAIL (2 blockers, 13 majors, 9 minors). Round 6: a FRESH general-purpose agent (never a fork) \u2014 a FRESH general-purpose agent, never a fork. Verdict FAIL (5 blockers, 10 majors, 6 minors). Round 7: a FRESH general-purpose agent (never a fork) \u2014 a FRESH general-purpose agent, never a fork. Verdict FAIL (4 blockers, 14 majors, 11 minors). Round 8: a FRESH general-purpose agent (never a fork) \u2014 a FRESH general-purpose agent, never a fork. Verdict FAIL (7 blockers, 14 majors, 10 minors). Round 9: SendMessage to the SAME auditor \u2014 the same auditor continued via SendMessage - patched, not rebuilt. Verdict FAIL (3 blockers, 9 majors, 8 minors). Round 10: SendMessage to the SAME auditor \u2014 the same auditor continued via SendMessage - patched, not rebuilt. Verdict FAIL (0 blockers, 2 majors, 10 minors). Round 11: SendMessage to the SAME auditor \u2014 the same auditor continued via SendMessage - patched, not rebuilt. Verdict PASS (0 blockers, 0 majors, 10 minors).",
  "delta_vs_prior": "vs calibration-CBR.V-20260804-1329.json. signal_medium BUY -> HOLD. signal_long BUY -> HOLD. quality 70 -> 59. valuation 45 -> 46. timing 56 -> 61. economic alignment 64 -> 76. price 1.04 -> 1.48. warranted_ratio set to 1.27 (absent in the prior report). entry_conviction Half-Size -> Wait. entry_groups_met 1 -> 0. exit_action Hold -> Trim. analysis_status donatien-pick -> stopped. next_update 2026-08-18 -> 2026-09-24. base target 1.4 -> 1.22. bull target 2 -> 1.84. bear target 0.8 -> 0.88. relative strength vs sector REMOVED (was strong outperform). moat 58 -> 44. competitive share trajectory gaining -> improving. overall confidence 45 -> 40. exit_groups_live 0 -> 1. probability-weighted target set to 1.23 (absent in the prior report). scenario weights set to {'bull': 20, 'base': 45, 'bear': 35} (absent in the prior report). gates_caution 2 -> 2 (added Gate 4 - Leverage (net debt near nil after the 24 Aug placement, but a US$45m gold-denominated obligation whose interest capitalises into principal until Dec 2026, on a mine that poured this morning); Gate 6 - Structural (single asset, single jurisdiction, 6.2-year starter life, Stage 2 unstudied); removed Valuation Ceiling; Dilution / Leverage). Carried without change: signal_short, driver, val_band, short_entry_confirmed, hard_gate_state, industry benchmark score.",
  "lint": {
    "exit": 0,
    "assertions": 67,
    "errors": 0,
    "warnings": 6,
    "could_not_run": [
      "CCY-PE-SKIPPED",
      "FW-GUARDRAIL-SKIP"
    ],
    "at": "2026-09-11T05:31",
    "report": "CBR.V_Signal_v6_20260910_1600.html"
  }
}

In one paragraph. Cabral Gold poured first gold at Cuiú Cuiú on 10 September 2026 — roughly 1,130 ounces, ahead of the schedule in its own feasibility study, funded by a US$45m gold loan rather than by dilution. That is a small company executing well and this report credits it: Business Quality is 59, medium, on an all-in sustaining cost of US$1,210 an ounce against spot of US$4,416 — a 72.6% margin. The signal nonetheless comes down to HOLD on all three horizons from HOLD / BUY / BUY — and the honest attribution is the Quality re-derivation, not the price: if August’s Quality of 70 were held, the matrix would return BUY on today’s valuation and timing. Marking the company’s own July-2025 prefeasibility study to spot gold, adding the rest of the district at US$75 an ounce in situ, crediting C$26.1m of in-the-money exercise proceeds against net debt of approximately nil across 397.7m in-the-money shares gives NAV of C$1.17 a share. At C$1.48 that is 1.27× net asset value — the Full band, which against medium business quality is a HOLD row in the decision matrix. Gate 3 does not fire: the valuation ceiling needs 1.40× and the corrected ratio is 1.27×. Earlier rounds published it as triggered on a net asset value that omitted the August placement cash and used superseded April strike prices. Three corrections to earlier drafts of this report belong here rather than in a footnote. First, an earlier draft scored Quality 60 and published SELL; that 60 was produced by putting the valuation conclusion inside the Quality pillar — replacing an operating benchmark with P/NAV — so the same multiple was counted twice, and that double count was the whole difference between SELL and HOLD. Second, an earlier draft said the 4 August report had missed the feasibility study; it had not, and the claim is withdrawn — that report recorded the study in full and its real defect was narrower, a valuation band asserted over a null anchor. Third, an earlier draft said no analyst covered this name; two do, both Strong Buy, mean target C$1.65 — 11.5% above the stamped price and squarely against this report’s conclusion, which is why it is now in §4 rather than absent. Two different questions get two different answers, and earlier rounds ran them together. What moved the signal since August is the Quality re-derivation, 70 to 59 — at August’s Quality the matrix would still return BUY on today’s numbers. What the level is most sensitive to is the in-situ mark, which is what the §4 surface maps. The dilution treatment decides neither: all four treatments return HOLD. The §4 surface returns SELL only below about US$36 an ounce in situ — a level nobody argues for — HOLD from US$75, and a BUY at US$200. The SELL this report carried for four rounds was never robust — it needed four independent conservative choices to break the same way, and each one alone returns HOLD: the in-situ mark, the cash figure, the strike prices and the dilution treatment. All four are now named in the fragility list. Because no horizon carries a BUY, the house rule takes Cabral off the active Donatien Pick list: it moves to Stopped, stays visible, and re-activates on any future BUY. Finally, the stamped price of C$1.48 is the 9 September close and predates this morning’s announcement; the shares traded near C$1.54 afterwards, at which the ratio is 1.32 — still inside the Full band.