NYSE:NOW ServiceNow, Inc.

Do-Not-Buy triggers — all five clear

1. Leverage into rising rates: clear. Net debt/EBITDA ~1.1× against a 4× trigger, with 20× interest cover.

2. Valuation extreme: clear on both arms. 26.19× forward is 1.12× warranted — nowhere near the 2.0× deep-expensive arm — and below the 33× sector guardrail, so the expensive-plus-catalyst arm cannot fire either. Worth stating plainly because the trailing GAAP multiple of 81.9× looks extreme and is not the right lens (§4).

3. Persistent negative earnings revisions: clear, with an honest qualification about the evidence. The consensus non-GAAP EPS ladder runs up — FY26 $4.07 → FY27 $5.02 → FY28 $6.09 → FY29 $7.10 — but that is a level ladder, i.e. expected growth. The trigger asks whether estimates are being cut over time, which needs a revisions series this run did not pull. The conclusion is very likely right; the evidence is indirect and is labelled so. Grades consensus is Buy (60 buy / 8 hold / 1 sell, n=69, 87% bullish).

4. Insider selling spike: clear. A 21-day news scan surfaced no insider-disposal story. It is not an exhaustive list of NOW coverage in the window — analyst actions alone include BofA raising its target (21 Aug), Cramer advising profit-taking (27 Aug) and BTIG at $170 (7 Sep). Absence in a news scan is weaker than a filings search, and is stated as such rather than dressed up.

5. Structural business-model threat: raised as a Gate 6 caution above, deliberately not as a Do-Not-Buy. The agentic-AI transition is real and underway — AI ACV past $1bn, about half of net new business already non-seat-based — but the company is so far monetising it rather than being displaced by it. That is a caution on a 3–5 year view, not an impairment.

Read the income statement on two bases, because they disagree. On the operating measures ServiceNow is excellent: revenue +24% year on year, gross margin 74.8%, free cash flow of $5.15bn at a 35% margin. On GAAP net income it is not: operating income has fallen four quarters running — $572m, $443m, $503m, $162m — while net income held up on non-operating items. In Q2 2026, net income of $298m exceeded operating income of $162m. Both facts are true and the report uses each where it belongs.

Industry benchmark — Rule of 40: 59. Revenue growth of 24% plus a free-cash-flow margin of 35%. Anything above 40 passes and 60 is exceptional, so this sits just under exceptional and comfortably in the top quartile of platform software. It is the single strongest argument for owning the business, and it is why Quality stays at 74 despite the GAAP margin deterioration. Benchmark score 82.

Competitive Environment. The threat to ServiceNow is not a rival taking its accounts. It is whether the unit it sells — a seat on a workflow platform — survives agentic AI in its current form.

The seat question is not hypothetical any more, and this is the strongest fact in the report. With Q2 2026 ServiceNow disclosed that AI ACV has crossed $1bn and that roughly half of net new business is already non-seat-based. The transition the bear case worries about and the bull case depends on is underway, not pending. It cuts both ways and that is precisely the difficulty: it proves the company can monetise AI outside the seat model, and it proves the seat model is being displaced. What it removes is the option of treating this as a distant risk.

Which earnings number, and why it matters more here than almost anywhere. ServiceNow's trailing GAAP P/E is 81.9×. Its forward non-GAAP P/E is 26.2×. Neither is a trick; they measure different things. Non-GAAP excludes stock-based compensation, which for this company is large and real. GAAP trailing income is inflated by non-operating items to the tune of 39.9% (§2, Gate 4). We anchor on the forward non-GAAP multiple — the basis the 7 August report used and the one consensus is quoted on — and we say so wherever the number appears. The trailing 81.9× is shown, not used.

What the basis change cost this name: almost nothing. The downgrade is the market and the rate. It is worth walking this carefully, because the first draft of this report got it wrong in an instructive way.

The 7 August report published actual 23.5× against warranted 27.7× — a ratio of 0.85, attractive. That 27.7× was the trailing-form warranted multiple. Restated forward at August's own discount rate (10-year 4.63%) it is 27.7 ÷ 1.15 = 24.09×, so August's consistent-basis ratio is 23.5 ÷ 24.09 = 0.976.

That still sits in the 0.80–1.00 band — Attractive, score range 65–77 — and August published a valuation score of exactly 65. Band unchanged, score range unchanged, and High/Attractive/Improving is still a BUY row. The basis restatement changes nothing about August's call.

What moved the ratio to today's 1.115 is the 11.7% price rise plus 15bp on the 10-year, which lifts r from 9.13% to 9.28% and trims the warranted multiple from 24.09× to 23.49×. This downgrade is the market and the discount rate, not a correction of last month's arithmetic.

The error worth naming: the first draft compared August's actual against September's warranted — mixing the time axis while accusing the prior report of mixing the trailing/forward axis. Exactly the fault it was prosecuting.

Embedded optionality. Genuine, and deliberately not scored as a reason the stock is cheap: the AI agent products are early in monetisation and consensus models very little of it, and the platform's move toward consumption pricing could decouple revenue from seat count — which is the same trend Gate 6 flags as a threat. That symmetry is the honest way to hold it: the identical development is the bull case and the bear case, and which one it becomes is not knowable from here. Worth a few points of tilt, already inside the 56.

Where price actually sits

At $131.11 the stock is above its daily 50-day ($118.36) and 200-day ($118.28) averages — which is what makes the daily read strong — but below its monthly SMA20 ($151.58) and SMA50 ($139.52), which is what keeps the monthly in a downtrend. Both are true. It has recovered 61% from the $81.24 low and remains 32.7% below the $194.73 high.

Position risk

Daily ATR is $6.73, about 5.1% of price — high, and it means ordinary noise is large. The nearest meaningful daily support is $106.05, which is 3.7 ATR below spot. That is beyond the framework's 2.5-ATR comfort line: there is no tight stop available, so any position taken here is sized against a wide one.

Momentum is rolling over at the short end: the daily MACD histogram is −0.81 and the hourly RSI is 34.65. Volume at 0.62× the 20-day average gives no confirmation in either direction.

Relative strength

Over twelve months ServiceNow has badly lagged — and by considerably more than this report said a draft ago. On price, NOW is −29.0% against SPY at +16.9% and XLK at +39.1%: 45.9 points behind the market and 68.1 behind its own sector. The figures previously shown here, −35 and −28, were inferred from the drawdown rather than computed, and the sector gap in particular was understated by forty points.

But twelve months is not the window the framework scores, and on the window it does score the picture inverts. Over one month NOW is +2.9% against SPY −1.4% and XLK +0.8%; over three months +23.6% against +5.1% and +6.4%. That is outperformance of both benchmarks on both timeframes — the top band of the relative-strength table, not the bottom. It does not lift the Timing score of 45, and that is by design rather than by oversight: relative strength is not a term in the timing formula, which is built from the multi-timeframe trend, risk-reward, macro, sentiment and catalysts. It is carried as an additional short-horizon input — context for the trade, not a component of the score.

Both readings are true, and they are one fact seen from two distances: a name still 29% below where it stood a year ago and 23.6% above where it stood three months ago. The three-month window opens at $106.06 on 10 June, but that is an arithmetic base, not a floor: price fell on to a closing low of $89.52 on 25 June and closed below $106 on 25 sessions before reclaiming the level in late July. The rally from that June low is +46.5%, and it is why the risk-reward above reads unfavourable — it has carried price 3.7 ATR clear of the nearest support at $106.05. (§12 is right that a tested bounce off $106 has not happened; price went straight through it.) Momentum has turned on the one- and three-month view — which is not the daily picture above, where MACD is negative and the hourly RSI is oversold; both are true at their own horizons. It has not repaired the twelve-month damage, and the macro still rates the sector Underperform. All figures are price returns: NOW pays no dividend while SPY and XLK do, so on a total-return basis the twelve-month gaps are about a point wider again.

Sentiment and catalysts

Forty-six analysts cover the price target, mean $142.28; the framework’s grades tool reads Buy — 60 buy / 8 hold / 1 sell on 69 analysts, 87% bullish. Set the target against a range of $72 to $248 and the "consensus" is thinner than the label suggests. The company calendar is quiet — Q3 results are late October — but the calendar that matters here is not ServiceNow’s. The FOMC meets on 16 September and the Bank of Japan on the 18th: two high-impact macro events two days apart, and §8 rates both directly relevant to a duration-sensitive name. That is the framework’s own clustering case, so the clustering score is marked down to 45 — the framework’s “noisy” band for two or three events inside fourteen days, where it had been sitting at 70, which is the band for a calendar with no major catalyst at all. To be clear about what that does and does not do: the clustering score modulates position sizing and, below 30, timing confidence. It is not the catalyst term in the timing formula, so the Timing score of 45 is unchanged by this correction. Since no entry path is open in any case, the sizing guidance is moot here — but it would not be if one were. It is still not a reason to pre-position — but the reason is that two clustered macro events sit inside the window, not that nothing is happening.

Upcoming events — next 14 days (this report's own risk window)

Date	Event	Impact	Forecast	Previous	Relevant?	Why
2026-09-11	US CPI (Aug)	High	+0.4% MoM, 3.4% YoY	core 3.34%	⚠ Indirect	Technology is a LOW macro-sensitivity sector, so the 3-day WAIT-override does not apply. It matters here only through the discount rate
2026-09-16	FOMC Decision	High	48–65% implied HIKE	3.50–3.75%	✅ YES	The single most relevant event. A hike raises the 4.78% risk-free rate that sets the warranted multiple — and long-duration software is where that bites hardest. The macro report names exactly this for NOW
2026-09-16	US Retail Sales (Aug)	Medium	+0.3% MoM	—	⚠ Indirect	A demand read; enterprise software lags consumer by quarters
2026-09-18	BoJ Rate Decision	High	25bp hike to 1.25% (97%)	1.00%	✅ YES	The carry unwind the macro report names as hitting long-growth names. A disorderly yen move is a direct risk-off channel for this cohort

Recent surprises (last 7 days)

Date	Event	Actual	Forecast	Surprise	Impact
2026-09-04	US Non-Farm Payrolls (Aug)	162k	53k (Dow Jones)	+109k, large upside	Hardens the hike case, which raises the discount rate feeding the warranted multiple. The macro report attaches a SINGLE-RELEASE CAUTION: one BLS release plus its own revisions is not independent corroboration, so it downgraded the driver rather than promoting it
2026-09-04	US Unemployment (Aug)	4.1%	4.2%	−0.1pp, below consensus	Unchanged from July. A stronger labour market, which points the same way

Two events matter to this name, and both cut the same way. The FOMC on 16 September carries a 48–65% implied probability of a hike, and the BoJ on 18 September is near-certain to hike into an already-fragile carry trade. The macro report names both mechanisms for ServiceNow specifically: duration sensitivity and the carry unwind.

The WAIT-FOR-EVENT override was evaluated and does not apply. It is scoped to High macro-sensitivity sectors; Technology is Low (weight 0.10). Recorded because the rule was tested, not skipped.

Both events land BEFORE this report's 24 September next-update date — the FOMC on the 16th and the BoJ on the 18th. If it hikes, the 4.78% risk-free input to §4 rises, the warranted multiple falls, and a ratio already at 1.12× moves toward Full. That is the single most likely path to a further downgrade.

Timeframe	Trend	Direction	RSI	MACD	Key S/R	Breakout	Vol
Monthly	Downtrend ↓	Bearish	47.8	−10.36, hist −3.79	S: $105.45   R: $163.06	Resistance breakout	0.30×
Weekly	Uptrend ↑	Bullish	55.8	+3.71, hist +5.52	S: $98.00   R: $139.20	Resistance breakout	0.22×
Daily	Strong uptrend ↑	Bullish	51.5	+5.94, hist −0.81	S: $106.05   R: $149.60 (the $129.75 level is already cleared)	Resistance breakout	0.62×
Hourly	Downtrend ↓	Bearish	34.7	−1.42, hist +0.15	S: $130.85   R: $138.98	Support breakdown	0.01×
15-minute	Strong downtrend ↓	Bearish	52.6	−0.03, hist +0.11	S: $130.85   R: $131.64	Resistance breakout	0.20×
Confluence: Bearish · MTF Score 45

The stack does not agree, and the disagreement is the information. The monthly frame is in a downtrend because price sits well below its own 20- and 50-month averages after a 32.7% drawdown; the daily frame is in a strong uptrend because price has recovered 61% off the low and sits above both daily averages. Those are not contradictory readings of the same thing, they are two different time horizons giving two different answers. What tips the balance is that the two frames closest to now — hourly and 15-minute — are both bearish, the daily histogram has turned negative, and the hourly RSI is 34.7. A recovery within a broken longer-term trend, currently losing short-term momentum. The level usually quoted here is a weekly close above $139.20. It has already happened twice and failed — weekly closes of $144.71 on 28 August and $141.26 on 4 September, and an August monthly close of $147.99 above the $139.52 monthly SMA50 — and the monthly frame is still a downtrend. A failed breakout is bearish confirmation, not a pending bullish trigger. What would actually change the read is a weekly close above $139.20 that HOLDS for two weeks, which neither attempt did.

Six months of daily closes with the 50-day average. Price is above both daily averages and far below the monthly ones — the recovery is real, and so is the drawdown it is recovering from.

Probability-weighted 12-month value: $140.00 (0.25 × $175 + 0.50 × $145 + 0.25 × $95), or +6.8% from $131.11. Worth sitting with that number: a 6.8% expected return, on a name with a 5.1%-of-price daily ATR and no stop closer than 3.7 ATR, into a macro the report itself rates Underperform. The distribution is mildly positive and the path is wide. That combination is what HOLD means.

Forecast:

Rule Forecast — when each condition is likely to resolve

ENTRY — Fundamental: price below ~$117.60 fair value.
→ FORECAST: needs a 10.3% fall, or growth in the forward number.
→ BASIS: two routes. Price falls to $117.60 — roughly 2 ATR, well within a normal quarter for a name this volatile — or forward EPS rises enough to lift fair value to today's price, which needs about $5.58 against the current $5.006. At ~20% earnings growth that is roughly two to three quarters away.
→ CONFIDENCE: Moderate. One of the two happens on most reasonable paths.

ENTRY — Technical: volume-confirmed strength with a positive histogram.
→ FORECAST: not close. Volume at 0.62× would have to more than double.
→ BASIS: the histogram is negative and the hourly frame is in a downtrend, so both missing legs are moving the wrong way. The pullback branch — a tested bounce off $106 with a higher low — is the more likely of the two, and it needs a further 19% fall.
→ CONFIDENCE: Low on the breakout branch; Moderate on the pullback branch.

ENTRY — Catalyst: Q3 results, late October.
→ FORECAST: catalyst-dependent, and the single most likely route to a signal change in either direction.
→ BASIS: the AI monetisation question gets its next real datapoint there. A net-expansion figure above 120% would move the bull case; below 115% would trigger the thesis-invalidation floor in §5.
→ CONFIDENCE: High that the date holds; no view on direction.

EXIT — Stop-Loss: two consecutive closes below $106.05.
→ FORECAST: possible on a hawkish FOMC; 3.7 ATR away.
→ RISK TRIGGER: 16 September. A hike is the clearest path there.
→ CONFIDENCE: Moderate.

EXIT — Profit-Target: $140 median target with RSI > 70.
→ FORECAST: the price leg is only 6.8% away, but RSI at 51.5 is far from 70.
→ BASIS: both legs are required, so this is unlikely to fire without a sharp move.
→ CONFIDENCE: Low.

Position sizing not computed — no allocation or portfolio role was specified, and the framework does not assume one. Volatility context only.

The §12 Conviction Ladder reads Wait: 0 of 3 entry paths are open. That is the sizing answer — there is no size, because there is no entry.

Daily ATR is $6.73, about 5.1% of price — high, and roughly double what a typical large-cap carries. Beta is 0.97, which understates the real experience: the drawdown from $194.73 to $81.24 and the recovery to $131.11 is a far wider ride than a beta near 1 suggests. Size for the ATR, not the beta.

{
  "ticker": "NOW",
  "company": "ServiceNow, Inc.",
  "currency": "USD",
  "date": "2026-09-10",
  "version": "v6",
  "brand": "",
  "exchange": "NYSE",
  "exchange_ticker": "NYSE:NOW",
  "isin": "US81762P1021",
  "api_ticker": "NOW",
  "finder_ticker": "NOW",
  "finder_exchange": "NYSE",
  "analysis_status": "stopped",
  "price_at_rating": 131.11,
  "signal_short": "HOLD",
  "signal_medium": "HOLD",
  "signal_long": "HOLD",
  "base_signal_short": "HOLD",
  "base_signal_medium": "HOLD",
  "base_signal_long": "HOLD",
  "primary_signal": "HOLD",
  "short_hold_reason": "full_hold",
  "short_entry_confirmed": false,
  "quality_score": 74,
  "valuation_score": 56,
  "timing_score": 45,
  "driver_score": 52,
  "driver_score_short": 48,
  "driver_score_medium": 52,
  "driver_score_long": 58,
  "driver_label": "Neutral",
  "driver_name": "Enterprise IT spending cycle & AI budget reallocation",
  "amplification_note": "Driver 52 is inside the 36-64 no-amplification band AND economic pressure is a Headwind, so neither arm of Stage 2 is available on any horizon. A HOLD never amplifies in any case.",
  "composite_short": 54,
  "composite_medium": 59,
  "composite_long": 64,
  "lifecycle_stage": "growth",
  "overall_confidence": 50,
  "economic_alignment_stance": "Contrarian",
  "economic_alignment_pressure": "Headwind",
  "economic_alignment_conviction": 40,
  "economic_alignment_source": "watchlist-signal",
  "macro_report_date": "2026-09-09",
  "eps_trailing": 1.6,
  "eps_forward": 5.00551,
  "trailing_pe": 81.94,
  "actual_multiple": 26.19,
  "warranted_multiple": 23.49,
  "warranted_ratio": 1.115,
  "val_multiple_basis": "clean forward non-GAAP P/E (SBC-excluded, per the company's own definition)",
  "val_band": "fair",
  "discount_rate_r": 0.0928,
  "risk_free_10y": 0.0478,
  "g_near": 0.15,
  "g_term": 0.03,
  "warranted_forward_uncapped": 23.49,
  "warranted_trailing_form": 27.01,
  "fcf_yield": 3.7,
  "fcf_yield_on_ev": 3.7,
  "fcf_yield_on_mktcap": 3.8,
  "ev_ebitda": 39.9,
  "price_to_book": 10.83,
  "nonop_pct_of_net_income": 39.9,
  "clean_pe": 113.7,
  "clean_peg": 1.12,
  "implied_growth_rate": 18.5,
  "earnings_quality_note": "GAAP net income is INFLATED by non-operating items - the usual direction, and the opposite of WCP.TO. TTM non-operating income is $666m of $1,670m net (39.9%); in Q2 2026 net income of $298m EXCEEDED operating income of $162m, and operating income has fallen 572 -> 443 -> 503 -> 162 over four quarters. Every valuation figure in section 4 is therefore struck on non-GAAP earnings or free cash flow. The trailing GAAP P/E of 81.94x is shown but NOT used - and note it is itself FLATTERED: on an operating-basis clean EPS of about $1.15 the clean trailing multiple is ~113.7x, WORSE than reported. Carried forward from the 2026-08-07 run, which recorded 60%.",
  "competitive_share_trajectory": "stable",
  "competitive_threat_level": "elevated",
  "moat_score": 66,
  "industry_benchmark_name": "Rule of 40 (platform software)",
  "industry_benchmark_value": 59,
  "industry_benchmark_score": 82,
  "analyst_consensus_target": 142.28,
  "analyst_target_high": 248.0,
  "analyst_target_low": 72.0,
  "analyst_target_upside_pct": 8.5,
  "analyst_coverage_count": 69,
  "analyst_grades_consensus": "Buy",
  "analyst_bullish_pct": 87.0,
  "analyst_grades_n": 69,
  "analyst_grades_detail": "0 strongBuy / 60 buy / 8 hold / 1 sell, from get_grades_consensus. NOT Yahoo's 'strong_buy' recommendation label on 46 analysts - that is a different field and an earlier draft published it by mistake, along with a bullish_pct of 78 sourced from neither.",
  "analyst_dispersion_note": "High $248 against low $72 is a 3.4x spread - past the framework's 2x trigger. Valuation confidence docked to 70%. Note the two analyst counts differ by source (grades tool 69, price-target feed 46); that is a source difference, not coverage loss.",
  "implied_growth_pct": 18.5,
  "peg_computed": 1.13,
  "own_history_percentile_52w": 44,
  "sector_median_note": "Large-cap application software ~26-30x forward; ESTIMATED from the peer set, not sourced from a data feed this run.",
  "threshold_fragility": "Two live fragilities. (1) The MTF confluence verdict rests on the hourly and 15-minute frames, whose volume ratios are 0.01x and 0.20x; a re-pull hours later flipped the 15-minute frame and the whole confluence from bearish to bullish. (2) timing_score 45 is TEN POINTS below the 55 at which the band turns Neutral -> Improving, and High/Fair/Improving is a BUY row on all three horizons. This HOLD is closer to flipping than a flat verdict suggests - and the newly computed relative strength points the same way, since a name outperforming both benchmarks over one and three months is not what a Neutral timing band usually describes.",
  "relative_strength_vs_spy": 18.5,
  "relative_strength_vs_sector": 17.3,
  "relative_strength_note": "COMPUTED this run from raw Yahoo closes, not inferred. The fields carry the 3-MONTH spread, the framework's scored window (SKILL L809 scores 1-month and 3-month; L1343 carries RS as an additional short-horizon input, and it is NOT a term in the timing_score formula at L845, so this correction changes no score). 1-month: NOW +2.9 vs SPY -1.4 vs XLK +0.8. 3-month: NOW +23.6 vs SPY +5.1 vs XLK +6.4, i.e. +18.5 and +17.3 (differenced unrounded: 23.62 - 5.10 and 23.62 - 6.36). 12-month, for context only: NOW -29.0 vs SPY +16.9 vs XLK +39.1, i.e. -45.9 and -68.1. PRICE returns, not total returns - NOW pays no dividend, SPY and XLK do, so the 12-month gaps are about a point wider on a total-return basis. The prior -35.0/-28.0 were inferred from the drawdown and were never computed; the sign flip is a BASIS CORRECTION, not a momentum reversal.",
  "catalyst_clustering_score": 45,
  "dynamic_macro_weight": 0.1,
  "mtf_confluence": 45,
  "risk_reward_score": 35,
  "fair_value_est": 117.6,
  "stop_loss": 106.05,
  "target_price": 145.0,
  "scenario_base_target": 145.0,
  "scenario_bull_target": 175.0,
  "scenario_bear_target": 95.0,
  "scenario_probabilities": {
    "bull": 25,
    "base": 50,
    "bear": 25
  },
  "probability_weighted_target": 140.0,
  "entry_groups_met": 0,
  "entry_conviction": "Wait",
  "exit_groups_live": 0,
  "exit_action": "Hold",
  "gates_triggered": [],
  "gates_caution": [
    "Gate 4 - Dilution/Accounting (high SBC; GAAP net income non-operating-inflated at 39.9%)",
    "Gate 6 - Structural business-model threat (agentic AI vs the seat-based model)"
  ],
  "do_not_buy_triggers": [],
  "hard_gate_state": "caution",
  "next_update_date": "2026-09-24",
  "next_update_basis": "default +14d (Q3 earnings late October; Technology is LOW macro-sensitivity so the 3-day CPI/FOMC carve-out does not apply)",
  "user_horizon": null,
  "user_allocation_pct": null,
  "portfolio_role": null,
  "post_audit_fixes_round1": "Two blockers: the inverted basis claim corrected in five locations, and the Donatien-Pick auto-Stop executed. M9 was WITHDRAWN by the auditor after challenge, and its replacement is more useful: Yahoo eps_trailing oscillated 1.56<->1.60 and trailing_pe 84.04<->81.94 at an IDENTICAL price within one session. The published figures are right; the FIELD IS UNSTABLE.",
  "post_audit_fixes_round2": "Residual BLOCKER: the refuted claim survived in a SIXTH location, the section-4 pillar headline, four paragraphs above the body refuting it. M2-1: Gate 1 CLEARED - the arm is 'below 0.8 AND DECLINING' and at 0.7006 against August's 0.70 it is not declining; an earlier draft reversed a CORRECT August adjudication by quoting half an arm. gates_caution 3 -> 2. M2-2 false exhaustiveness dropped. M2-3 all three scenario cards on FY28 with the 2.4x consensus dispersion disclosed. M2-4 stale 24.1x -> 23.8x. M2-5 sector median now reads as the LOW END of the band. M2-6 clean_peg and implied_growth_rate added so section 14 stops contradicting section 4. M2-7 coverage 46 -> 69. MINORs 1, 2, 4 fixed.",
  "audit_route_note": "Round 1: a FRESH general-purpose agent (never a fork) \u2014 first round. Verdict FAIL (2 blockers, 13 majors, 9 minors). Round 2: SendMessage to the SAME auditor \u2014 artifact patched, not rebuilt; the auditor held every derivation. Verdict FAIL (1 blocker, 8 majors, 4 minors). Round 3: SendMessage to the SAME auditor \u2014 patched, not rebuilt; both defects were in section 14. Verdict FAIL (1 blocker, 1 major, 4 minors). Round 4: SendMessage to the SAME auditor \u2014 patched, not rebuilt; the auditor read section 14 as a unit rather than checking the round-3 fixes. Verdict FAIL (0 blockers, 1 major, 5 minors). Round 5: SendMessage to the SAME auditor \u2014 patched, not rebuilt; the auditor verified the audit-trail set field-by-field and then swept the sections attention had left. Verdict FAIL (0 blockers, 2 majors, 4 minors). Round 6: SendMessage to the SAME auditor \u2014 patched, not rebuilt; the auditor checked a downstream claim instead of trusting it, and left the report to do it. Verdict FAIL (1 blocker, 1 major, 1 minor). Round 7: SendMessage to the SAME auditor \u2014 patched, not rebuilt; the auditor reproduced all nine relative-strength returns to the decimal and found the defect in the sentence that reconciles them. Verdict FAIL (0 blockers, 1 major, 3 minors). Round 8: SendMessage to the SAME auditor \u2014 patched, not rebuilt; a full end-to-end re-read of section 7 plus a 17-assertion integrity pass and the three-axis sweep. Verdict PASS (0 blockers, 0 majors, 2 minors).",
  "delta_vs_prior": "vs calibration-NOW-20260807-1030.json. signal_short BUY -> HOLD. signal_medium BUY -> HOLD. signal_long BUY -> HOLD. quality 78 -> 74. valuation 65 -> 56. timing 58 -> 45. driver 62 -> 52. economic alignment 52 -> 40. price 117.35 -> 131.11. warranted_ratio 0.85 -> 1.115. val_band attractive -> fair. entry_conviction Full-Size -> Wait. entry_groups_met 2 -> 0. short_entry_confirmed true -> false. analysis_status donatien-pick -> stopped. next_update 2026-08-21 -> 2026-09-24. base target 135 -> 145. bull target 175 -> 175. bear target 82 -> 95. relative strength vs SPY set to 18.5 (absent in the prior report). relative strength vs sector set to 17.3 (absent in the prior report). gates_caution 2 -> 2 (added Gate 4 - Dilution/Accounting (high SBC; GAAP net income non-operating-inflated at 39.9%); Gate 6 - Structural business-model threat (agentic AI vs the seat-based model); removed Accounting/Dilution (high SBC; GAAP net income non-operating-inflated \u2014 scored on non-GAAP/FCF); Structural Threat (agentic-AI contest \u2014 currently a NOW tailwind, monitored not fired)). Carried without change: exit_action, hard_gate_state. NOTE: The relative-strength sign change from -35.0/-28.0 to +18.5/+17.3 is a BASIS CORRECTION, not a momentum reversal. The prior pair was inferred from the drawdown and never computed; the new pair is the framework's own 3-month scored window (SKILL L809). On the same 12-month basis as the prior pair the figures are WORSE, not better: -45.9 vs SPY and -68.1 vs the sector. Do not diff this as performance.",
  "post_audit_fixes_round3": "Both defects were in section 14 and both were the same failure mode - prose written in an earlier round that per-finding fixes had no reason to revisit. BLOCKER: the refuted claim survived in a SEVENTH location, section 14's opening clause ('the reason splits about evenly between the market and our own arithmetic'), contradicted TWO SENTENCES LATER in its own paragraph. The corrected sentence was inserted and the clause it was meant to replace was left standing. M3-1: the section-14 EMBEDDED delta_vs_prior was stale ('gates_caution 2 -> 3 ... Gate 1 newly raised') while the on-disk field was correct - because the embed renders from the SPEC and close_audit_round.py writes to DISK. One field, two authors. Fixed structurally: the spec's delta_vs_prior is now synced FROM disk before rendering, so close_audit_round.py is its sole author. close_audit_round.py now also warns when a rendered report embeds an audit-trail field that no longer matches the calibration. MINOR 1 the balance-sheet sub-signal no longer refers to a liquidity caution that no longer exists. MINOR 2 the not-declining leg now states that August's 0.70 is a rounded prose figure, so flatness is established to two decimals not four. MINOR 3 the bull card now gives 28.7x on the same FY28 basis as the other two.",
  "post_audit_fixes_round4": "M4-1: post_audit_fixes_round3 made a FALSE claim about its own fix, and contradicted itself inside one sentence - 'the spec no longer carries delta_vs_prior at all' AND 'the spec is synced FROM disk before rendering' cannot both be true. The second half was the true one. Corrected, and then made true structurally rather than by prose: build_stock_report.py now takes the calibration as an optional third argument and syncs delta_vs_prior, audit_route_note and every post_audit_fixes_round* FROM disk before rendering, so disk wins by construction instead of by my remembering to copy it across. close_audit_round.py's divergence warning widened to threshold_fragility and the post_audit_fixes_round* fields, since restate_round_claims rewrites round counts on disk while the spec still authors the sentence - the same two-author shape one field later. MINOR 1 section 14 now states the one ACTION this report triggers: no BUY on any horizon takes the name off the active Donatien Pick list as Stopped, visible, re-activating on a later BUY. It appeared only in the header badge and delta_vs_prior. MINOR 2 the round-3 fix had turned a contradiction into a repetition - the same point made twice in near-identical words four sentences apart; one dropped. MINOR 3 threshold_fragility no longer opens with round-artefact framing ('both undisclosed in the first draft') - the field states the fragility, and the next run diffs it. MINOR 4 the round-1 audit_route_note entry duplicated itself because --route and --route-reason said the same thing.",
  "post_audit_fixes_round5": "Both majors were the STATES axis: a correction landed where the finding NAMED it and nowhere else. M5-1: section 7 published 'a Strong Buy consensus at 1.45' - the exact Yahoo label that section 4, section 15 and analyst_grades_detail all explicitly disown. M2-5 corrected section 4 and the calibration; nothing re-read section 7. No quantifier moved, so no count sweep could see it. Section 7 now gives the framework's own grades read (Buy, 60/8/1 on 69, 87 percent bullish) and keeps 46 analysts only where it is correct, on the target range. M5-2: section 15's get_yahoo_quote line said the live endpoint 'does not return' 1.56/84.04, contradicting post_audit_fixes_round1 two sections away, which correctly records the field as UNSTABLE - it returns 1.56/84.04 AND 1.60/81.94 at an identical 131.11 within one session. Section 15 now records the instability. M5-3 (tooling): the divergence guard was BLIND to a dropped field - the 'k in html' precondition skipped any watched field the embed had lost entirely, which is precisely the failure mode round 4 created by moving these fields out of the spec. Precondition removed; the check pulled out of main() into embed_divergence() with a --self-test that proves all four cases, because three bugs in three rounds all shipped as clean passes. build_stock_report.py now AUTO-RESOLVES the calibration beside the output file instead of taking it as an optional argument, so the trail cannot vanish by omission, and it syncs threshold_fragility too - by construction rather than by warning, which was round 4's own lesson. MINOR 2: section 7 now carries the relative-strength figures WITH the caveat that they are derived from the drawdown, not computed, and section 15 has a source line saying so - the last open piece of round-1 M4. MINOR 3: the section-11 bull card's Wells Fargo 175 / BTIG 170 sourcing now has a section-15 entry.",
  "post_audit_fixes_round6": "BLOCKER (outside the report, live on donatien.ca): the auto-Stop was executed in the calibration and the report header, delta_vs_prior asserted the propagation as done, and NONE of it had happened. The watchlist row still read Donatien Pick, still linked to the 7 August report, and the Portfolio grid tile was still live advertising a Full-Size conviction at 117 dollars with a 175 bull for a name now rated HOLD/HOLD/HOLD at 131.11 and badged Stopped. Both repos were pushed. Five audit rounds never looked outside the report because nothing asked them to. Fixed by running build_watchlist.py --write (row now status-stopped, HOLD x3, Wait, linking to the 10 September report) and build_portfolio_grid.py --write (tiles 3 -> 2; the US Tech cell is now empty, which is honest). Then fixed the CLASS, not just the instance: preflight_report.py now reads the PUBLISHED page - WL-BADGE-STALE, WL-LINK-STALE, WL-TILE-STALE, WL-STALE-ROW, WL-MISSING-ROW - and all five are proven to fire, three of them against the exact watchlist that was live at the moment of the blocker, recovered from git. Two bugs were found in the new check while proving it: a fixed byte-window spanned neighbouring rows so a sibling's badge could satisfy the check, and the first 'reports/NOW/' match in the document is the GRID TILE rather than the table row, so it read the wrong element entirely. The row is now identified by what only a row has, a status cell. M6-1: section 7 called the catalyst calendar 'calm' with 'no near-term event to size around' while section 8 calls the 16 September FOMC 'the single most likely path to a further downgrade', section 11's bear trigger is dated to it and section 12's stop-loss names it as THE risk trigger. It also inverted the framework's own clustering rule (SKILL L911/L958), whose worked example is a Fed meeting: FOMC on the 16th and BoJ on the 18th are two high-impact events two days apart. catalyst_clustering_score 70 -> 50, which changes no signal since clustering only bites below 30. The CONCLUSION survives - do not pre-position - but the reason was backwards. M6-2: 'thinner than either label suggests' had two antecedents until round 5 removed one.",
  "post_audit_fixes_round7": "M7-1: I turned the 3-month ARITHMETIC BASE into a support event. Section 7 said the name was 'three months into a hard bounce off 106.06' - but 106.06 is simply the close on 10 June, the denominator of the 3-month return. Price then fell to a closing low of 89.52 on 25 June and closed BELOW 106 on 25 separate sessions before reclaiming the level in late July. Worse, the invented mechanism falsified a status cell two sections away: section 12 marks 'a tested bounce off 106 support with a higher low' as NOT MET, and section 12 is right. The real rally is +46.5% from 89.52, and 89.39 was already sitting in the MTF daily support list, so the true level was in the tool output the whole time. The reconciliation now states the arithmetic base as an arithmetic base and agrees with section 12 explicitly. MINOR 1: 'momentum' was used in opposite directions two paragraphs apart in a section whose whole architecture is that horizons disagree - now qualified to the one- and three-month view, against the daily MACD and hourly RSI above it. MINOR 2: section 7 put 'the top band of the relative-strength table' inside a pillar scored 45 and left the reader without the reconciliation the calibration note gives the machine; it now says relative strength is not a term in the timing formula and is carried as context for the trade rather than a component of the score. MINOR 3: relative_strength_vs_sector was differenced from ROUNDED inputs (23.6 - 6.4 = 17.2) in the round that was about computing properly; unrounded 23.62 - 6.36 = 17.26, so 17.3. The auditor's open question sent me back to the density table and found catalyst_clustering_score 50 was in the WRONG BAND - 50-69 is '1 clear catalyst within 30 days', the calm-ish reading the previous round was correcting, while two events inside fourteen days is 20-49 'noisy'. Now 45, above the 30 at which confidence is docked. And the ambiguity itself is now closed in the SKILL: the clustering score is NOT the catalysts term in the timing_score formula - the formula's term scores whether the calendar is favourable, clustering scores how crowded it is, and they move independently. Changing clustering does not move timing_score. The report now says so.",
  "post_audit_fixes_round8": "No fixes required. Round 8 returned PASS at 0 blockers and 0 majors. Two minors logged and accepted rather than resolved: the report HEADER badge still renders status-stopped as badge-purple (the watchlist renders correctly; the header needs a builder and stylesheet change affecting every report), and the section-10 lint assertion (plotted array length, first, last, min, max against the price bars) is a next-name ratchet - the auditor verified THIS artifact by hand, 127 closes with 0 mismatches, first 116.61 on 2026-03-10, last 131.11 on 2026-09-09, min 83.00, max 147.99, which closes the last surface neither of us had ever examined. Carried and non-blocking: the 3.8bn net debt figure is lease-inclusive and caveated, no gate or Do-Not-Buy trigger turns on it, and it should be resolved from the balance sheet at the next refresh rather than carried further; the late-October earnings date cannot reach a 14-day gate from seven weeks out; the judgement-scored sub-scores and confidence percentages are coherent with no rule constraining them; and FY28-as-a-12-month-target is sound because by September 2027 the market's forward year IS FY28."
}

In one paragraph. ServiceNow is a HOLD on all three horizons, down from BUY on all three, and the reason is the market and the discount rate, not our own arithmetic. The business is in good shape: revenue +24%, a Rule of 40 of 59, $5.15bn of free cash flow at a 35% margin, and switching costs deep enough that no competitor has moved a large estate. But the price has risen 11.7% since August and the 10-year has added 15bp, and those two together take the warranted ratio from 0.976 to 1.115 — out of Attractive and into Fair. Restating the anchor onto a consistent forward basis moves August's recorded 0.85 to 0.976 but leaves its band, score range and BUY signal untouched. Against that, the macro has turned: the report now rates ServiceNow Underperform on all three horizons, duration-sensitive into a possible hike and into the yen carry unwind, which makes STRONG BUY unreachable and going long explicitly Contrarian. No entry path is open — fair value of about $117.60 sits below the market price, so even the cheapest route in is shut. Gate 4 and Gate 6, both carried forward from August, remain live and unresolved: high stock-based compensation with GAAP income inflated by non-operating items, and the structural question of whether agentic AI compresses the seat-based model this company bills against. That last one is the same development as the bull case — and with AI ACV past $1bn it is already happening, which is why it sits as a caution rather than a distant risk. One thing follows automatically from three HOLDs: with no BUY on any horizon, ServiceNow comes off the active Donatien Pick list as Stopped — still visible on the watchlist, still losing its Portfolio-grid tile, and re-activating the moment a later report produces a BUY anywhere.